Control of Power Electronic Converters and Systems, Volume 3, explores emerging topics in the control of power electroni
282 82 32MB
English Pages 718 [720] Year 2021
Table of contents :
Front Cover
Control of Power Electronic Converters and Systems
Control of Power Electronic Converters and Systems
Copyright
Contents
Contributors
Preface
1  Advanced control of power electronic systems—an overview of methods
1.1 Introduction
1.2 Linear controller implementation and analysis
1.2.1 Implement PI controller
1.2.2 Implement PR controller
1.2.3 Implement repetitive controller
1.2.4 Pole placement control
1.2.5 Linear quadratic regulator
1.3 Nonlinear controller implementation and analysis
1.3.1 Hysteresis control
1.3.2 Lookup table–based direct power control
1.3.3 Model predictive control
1.3.4 Backstepping control
1.3.5 Sliding mode control
1.3.6 Feedback linearization
1.3.7 Passivity control
1.3.8 Adaptive control
1.3.9 H∞ control
1.3.10 Artificial intelligence
1.4 Summary
References
2  Robust design and passivity control methods
2.1 Introduction
2.2 State space model of power electronic systems
2.2.1 Introduction
2.2.2 Gridconnected voltage source inverter
2.2.3 An example of a VSC (STATCOM)
2.3 Robust controller design
2.3.1 Voltagemodulated direct power control
2.3.2 Tracking controller
2.3.3 Robust control design
2.3.4 Simulation results
2.4 Robust smallsignal stabilization control design
2.4.1 Introduction
2.4.2 Lur’e type model
2.4.2.1 Example 1: DC/DC buck converter model
2.4.2.2 Example 2: DC microgrid model
2.4.3 Robust smallsignal stability analysis problem
2.4.4 Robust stability analysis method
2.4.5 Robust stabilization control design
2.4.5.1 Example 3: robust smallsignal stability of a DC microgrid
2.5 Passivitybased control design
2.5.1 Introduction
2.5.2 Portcontrolled Hamiltonian system
2.5.3 Passivitybased control design in a microgrid
2.5.4 Passivity control design in STATCOM
2.6 Conclusions and future prospective
Acknowledgment
References
3  Sliding mode controllers in power electronic systems
3.1 Introduction
3.2 System dynamics of a voltage source inverter
3.3 Introduction to sliding mode control
3.3.1 Reachability condition in the sliding mode control
3.3.2 Control input calculation
3.4 Classical sliding mode control
3.5 Boundarylayer sliding mode control
3.5.1 Adaptive boundarylayer sliding mode control
3.6 Adaptive sliding mode control
3.7 Integral sliding mode control
3.8 Terminal sliding mode control
3.9 Secondorder sliding mode control
3.9.1 Twisting SMC
3.9.2 Supertwisting
3.9.3 Adaptive supertwisting
3.10 Comparison of different SMCs
3.11 Experimental results
3.11.1 Steadystate performance
3.11.2 Dynamic performance
3.12 Conclusions
References
4  Model predictive control of power converters, motor drives, and microgrids
4.1 Introduction on MPC
4.2 MPC for permanentmagnet synchronous motor drives
4.2.1 Model of PMSM
4.2.2 FCSMPC for PMSM drives
4.2.3 Performance evaluations of MPC with simulations
4.2.4 Summary
4.3 MPC for microgrids
4.3.1 Dynamics and predictive model of VSCbased MG
4.3.2 MPC for robust and fast operation of an islanded AC MG
4.3.3 Dynamic stabilization of a DC MG using MPC
4.3.4 Performance evaluation with experimental results
4.3.5 Summary
4.4 MPC for renewable energy applications (PMSG wind turbine)
4.4.1 Modeling of PMSG wind turbine system with backtoback power converter
4.4.2 Direct model predictive current control
4.4.3 FCSMPC for threelevel PMSG systems
4.5 Conclusions and future trends
References
5  Adaptive control in power electronic systems
5.1 Introduction
5.2 System dynamics on a UPS system
5.3 System identification
5.3.1 Parameters identification
5.3.1.1 Recursive leastsquare estimation
5.3.2 State identification
5.3.2.1 Observability condition
5.3.2.2 Luenberger observer
5.3.2.3 Kalman filter
5.3.2.4 Disturbance estimation based on an adaptive observer
5.4 Indirect adaptive predictive control
5.4.1 Experimental results
5.5 Model reference direct adaptive control of UPS
5.5.1 The outer voltage control loop
5.5.2 The inner current control loop
5.5.3 Experimental results
5.6 Conclusion
References
6  Machine learning technique for lowfrequency modulation techniques in power converters
6.1 Introduction
6.2 Cascaded Hbridge active power filter configuration
6.3 ANN for the asymmetric selective harmonic current mitigationPWM
6.4 The proposed technique simulation and experimental results
6.4.1 Simulation results
6.4.2 Experimental results
6.5 Conclusion
References
7  Overview of stability analysis methods in power electronics
7.1 Introduction
7.2 Smallsignal stability analysis methods
7.2.1 Modeling of power converter
7.2.1.1 State space averaging method
7.2.1.2 Generalized state space averaging method
7.2.1.3 Harmonic state space method
7.2.1.4 Black box method
7.2.2 Eigenvalue method
7.2.2.1 Component modeling and system integration
7.2.2.2 Eigenvalue stability analysis
7.2.2.2.1 Eigenvalue analysis
7.2.2.2.2 Participation factor analysis
7.2.2.2.3 Sensitivity analysis
7.2.2.3 Verification
7.2.3 Impedancebased method
7.2.3.1 Impedance modeling
7.2.3.1.1 Impedance modeling
7.2.3.1.2 Network partitioning
7.2.3.2 Stability analysis
7.2.3.3 Verification of analysis
7.2.4 Comparison of methods
7.3 Largesignal stability analysis methods
7.3.1 Timedomain simulations
7.3.2 Lyapunovbased analytical methods
7.3.2.1 Takagi–Sugeno multimodel method
7.3.2.2 Brayton–Moser’s mixed potential
7.3.2.3 Optimal Lyapunov function generation
7.4 Case studies with practical examples
7.4.1 Smallsignal stability analysis
7.4.1.1 Eigenvalue method
7.4.1.1.1 System modeling
7.4.1.2 Impedancebased method
7.4.1.2.1 Impedance modeling
7.4.1.3 Verification
7.4.2 Largesignal stability analysis on a power electronic system
7.5 Summary
References
8  Cyber security in power electronic systems
8.1 Introduction
8.2 Cyber physical architecture of power electronic converters
8.2.1 Physical stage
8.2.2 Cyber stage
8.3 Vulnerability analysis of cyber attacks on control of VSCs
8.3.1 Cyber security
8.3.2 Vulnerability assessment
8.4 Cyber attack detection and mitigation mechanisms in power electronic systems
8.4.1 Detection
8.4.2 Mitigation
8.5 Test cases
8.5.1 Test case I
8.5.2 Test case II
8.6 Conclusions and future challenges
References
9  Advanced modeling and control of voltage source converters with LCL filters
9.1 Introduction
9.2 Modeling of the VSCs with LCL filters
9.2.1 Modeling methods in balanced threephase systems
9.2.2 Modeling methods in unbalanced threephase systems
9.2.2.1 Generalized averaged model
9.2.2.2 Harmonic state space model
9.2.3 Simulation examples
9.3 Alternative current control of the VSCs with LCL filters
9.3.1 Control in synchronous reference (dq) frame
9.3.2 Resonance damping technique
9.3.2.1 Passive damping technique
9.3.2.2 Active damping technique using filter capacitor current feedback
9.3.2.3 Active damping technique under grid current feedback
9.3.2.4 Active damping technique under converter current feedback
9.3.3 Control under unbalanced grid voltages
9.4 Impedancebased stability analysis under weak grid conditions
9.4.1 System control of the LCLfiltered VSCs in αβ frame and dq frame
9.4.2 Impedancebased stability analysis
References
10  Phaselocked loops and their design
10.1 Introduction
10.2 PLL’s control and design
10.3 Threephase PLLs
10.3.1 Conventional synchronous reference frame PLL
10.3.2 Moving average filter–based PLLs
10.3.3 Notch filter–based PLLs
10.3.4 Sinusoidal signal integrator–based PLLs
10.3.5 Secondorder generalized integrator–based PLLs
10.3.6 Complex coefficient filter–based PLLs
10.3.7 Delayed signal cancellation–based PLLs
10.3.8 Multiple SRF filter–based PLLs
10.3.9 Other threephase PLLs
10.3.10 Performance comparison and recommendation
10.4 Singlephase PLLs
10.4.1 Standard PPLLs
10.4.2 Low pass filter–based PPLLs
10.4.3 Moving average filter–based PPLLs
10.4.4 Notch filter–based PPLLs
10.4.5 Doublefrequency and amplitude compensation–based PPLLs
10.4.6 Modified mixer PDBased PPLLs
10.4.7 Transfer delayed–based PLLs
10.4.8 Inverse park transformation–based PLLs
10.4.9 Generalized integrator–based PLLs
10.4.10 Synthesis circuit PLLs
10.4.11 Performance comparison and recommendation
10.5 Summary
References
11  Stability and robustness improvement of power converters
11.1 Introduction
11.2 Stability and robustness improvement of current control
11.2.1 Smallsignal modeling
11.2.1.1 Linearization of the converter power stage
11.2.1.2 Smallsignal model of VSC with converterside current control
11.2.1.3 Smallsignal model of VSC with gridside current control
11.2.2 Passivitybased stability analysis
11.2.2.1 Passivitybased stability analysis for converterside current control
11.2.2.2 Passivitybased stability analysis for converterside current control
11.2.3 Robustness enhancement
11.2.3.1 Time delay reduction
11.2.3.2 Design of passive filters
11.2.3.3 Active damping
11.3 Stability and robustness improvement of outerloop control
11.3.1 Smallsignal modeling
11.3.1.1 Linearization of the PLL
11.3.1.2 Smallsignal model of VSC with PLL
11.3.1.3 Linearization of the DClink voltage control
11.3.1.4 Smallsignal model of VSC with DClink voltage control loop
11.3.2 MIMObased stability analysis
11.3.2.1 Dynamic impact of PLL
11.3.2.2 Dynamic impact of DVC
11.3.3 Robustness enhancement
11.3.3.1 PLL design
11.3.3.2 DVC design
References
12  High switching frequency threephase currentsource converters and their control
12.1 Challenges of high switching frequency CSCs control
12.1.1 Multiple timescale dynamics of high switching frequency CSCs
12.1.2 Control methods of high switching frequency CSCs
12.2 Stability analysis of the singleloop DClink current control
12.2.1 Smallsignal modeling
12.2.2 Stable region of singleloop DClink current control
12.2.3 Experimental validation
12.3 Active damping methods for high switching frequency CSCs
12.3.1 Virtual impedance analysis
12.3.1.1 Capacitorvoltage feedback (Fig. 12.12)
12.3.1.2 Capacitorcurrent feedback (Fig. 12.13)
12.3.1.3 Inductorcurrent feedback (Fig. 12.14)
12.3.1.4 Time delay effect on virtual impedance
12.3.2 Experimental validation of active damping
12.3.2.1 Capacitorvoltage feedback (see Fig. 12.12)
12.3.2.2 Capacitorcurrent feedback (see Fig. 12.13)
12.3.2.3 Inductorcurrent feedback (see Fig. 12.14)
12.4 Summary
References
13  Highpower current source converters
13.1 Introduction
13.2 Current source converters and applications
13.2.1 Thyristorbased technology
13.2.2 PWM CSCbased mediumvoltage drives
13.2.3 More potential application
13.3 Parallel CSC system and modulation strategies
13.3.1 Parallel CSC topology
13.3.2 CSC modulation strategies
13.3.2.1 Space vector modulation
13.3.2.2 Selective harmonic elimination
13.3.2.3 Direct dutyratio pulse width modulation
13.3.2.4 Comparison of different CSC modulations
13.4 Parallel CSC and circuit analysis
13.4.1 CM loop circuit of parallel CSC
13.4.2 DClink circuit of parallel CSC
13.5 DC current balance and CMV reduction methods
13.5.1 SVMbased methods
13.5.1.1 Interleaved SVM
13.5.1.2 Multilevel SVM
13.5.2 Carriershifted SPWMbased methods
13.5.3 Case study results
13.6 Conclusions
References
14  Parallel operation of power converters and their filters
14.1 Introduction
14.2 Circulating current modeling
14.2.1 Parallel converters with a common DC bus
14.2.1.1 Impact of the modulator mismatch
14.2.1.2 Impact of the impedance mismatch
14.2.2 Parallel converters with separate DC bus
14.3 Circulating current control
14.3.1 Current sharing schemes
14.3.2 Droop control scheme
14.3.3 Zero vector dwell time control
14.4 Harmonic performance improvement through interleaved operation
14.4.1 Modulation of parallel interleaved converters
14.4.2 Symmetrically interleaved converters
14.4.3 Harmonic performance evaluation
14.4.3.1 Two symmetrically interleaved VSCs
14.4.3.2 Three symmetrically interleaved VSCs
14.4.4 Nearest three vector modulation
14.5 Circulating current suppression in parallel interleaved converters
14.5.1 Galvanic isolation
14.5.2 Coupled inductor
14.5.2.1 Equivalent electric circuit
14.5.2.2 Impact of the modulation scheme
14.5.3 Commonmode inductor
14.5.4 Integrated inductor
14.6 Summary
References
15  Advanced power control of photovoltaic systems
15.1 Introduction
15.2 Overview of PV inverter control
15.2.1 Control structure
15.2.2 MPPT algorithm
15.2.2.1 Perturb and observe MPPT
15.2.2.2 Fractional open circuit voltage MPPT
15.3 Requirement of advanced control functionality
15.3.1 Grid code
15.3.1.1 Requirements under normal grid conditions
15.3.1.2 Requirements under abnormal grid conditions
15.3.2 Active power control requirement
15.4 Constant power generation control strategy
15.4.1 Direct power control (PCPG)
15.4.2 Currentlimiting control (ICPG)
15.4.3 Perturb and observe–based control (P&OCPG)
15.5 Benchmarking of constant power generation control strategy
15.5.1 Dynamic responses
15.5.2 Steadystate responses
15.5.3 Tracking error
15.5.4 Stability
15.5.5 Complexity
15.6 Summary
References
16  Low voltage ridethrough operation of singlephase PV systems
16.1 Introduction
16.2 Low voltage ridethrough operations
16.2.1 LVRT control using the singlephase PQ theory
16.2.2 LVRT control based on the powervoltage curve
16.3 Reactive power injection strategies under LVRT
16.3.1 Constant average active power control strategy (Const.P)
16.3.2 Constant active current control strategy (Const.Id)
16.3.3 Constant peak current control (Const.Igmax)
16.4 Summary
References
17  Gridfollowing and gridforming PV and wind turbines
17.1 Introduction
17.2 PV and wind turbine systems
17.2.1 PV structures
17.2.2 Grid converter for wind turbine systems
17.3 Gridfollowing power converters
17.3.1 Definition
17.3.2 Synchronization strategies
17.3.2.1 Synchronous reference frame phaselocked loop
17.3.2.2 Stationary reference frame frequencylocked loop
17.3.3 Current controllers
17.3.3.1 PI controller on the SRF
17.3.3.2 Resonant controller in a stationary reference frame
17.4 Gridforming power converters
17.4.1 Definition
17.4.2 Control schemes for gridforming power converter
17.4.3 Droop control in gridforming power converters
17.4.3.1 Grid impedance influence on droop control
17.4.3.1.1 Inductive grid
17.4.3.1.2 Resistive grid
17.4.3.1.3 General case
17.4.3.2 Virtual impedance control
17.4.4 The synchronous power controller
17.5 Conclusions
References
18  Virtual inertia operation of renewables
18.1 Introduction
18.2 Evolution of green energy transition
18.2.1 Lowinertia grid challenges
18.2.2 Control of power converter–interfaced renewables
18.3 Virtual inertiabased control
18.3.1 Virtual synchronous machine
18.3.2 Concepts and fundamentals
18.4 VSM implementation
18.4.1 NSG penetration level
18.4.2 VSM performance
18.4.3 Fault right through capability
18.5 Summary and future trend
References
19  Virtual inertia emulating in power electronic–based power systems
19.1 Introduction
19.2 Inertia concept
19.3 Inertia challenges of power electronic–based power systems
19.4 Adaptive inertia for gridconnected VSGs
19.4.1 VSG principle
19.4.2 Adaptive inertia
19.5 Simulation and experimental results
19.5.1 Simulation results
19.5.2 Experimental results
19.6 Summary
References
Further reading
20  Abnormal operation of wind turbine systems
20.1 Introduction
20.1.1 Classification of grid faults
20.1.2 Grid code requirements on lowvoltage ridethrough
20.1.3 Fundamental wind turbine configurations
20.2 Control of type III wind turbine during grid faults
20.2.1 Existing challenges during grid faults
20.2.1.1 Internal challenge from DFIG configuration
20.2.1.2 External challenge from grid codes
20.2.2 Control strategies during symmetrical grid faults
20.2.3 Control strategies during asymmetrical grid faults
20.3 Control of type IV wind turbine during grid faults
20.3.1 Modeling and control of gridside converter
20.3.1.1 Converter current control
20.3.1.2 Phaselocked loop for grid synchronization
20.3.1.3 DClink voltage controller
20.3.2 Symmetrical grid fault control
20.3.2.1 Protection of DClink capacitor through chopper control
20.3.2.2 Verification of symmetrical fault control
20.3.2.3 Zerovoltage ridethrough capability
20.3.3 Asymmetrical grid fault control
20.3.3.1 Grid synchronization during asymmetrical faults
20.3.3.2 Currentreference generation methods
20.3.3.3 Calculation of phase current magnitude
20.3.3.4 Selection of k1,k2,P∗,Q∗
20.3.3.5 Verification of asymmetrical fault control
20.4 Summary
References
21  Wind farm control and optimization
21.1 Introduction
21.2 Wind farm active dispatch
21.2.1 Wake effect
21.2.2 Single MPPT and global MPPT
21.2.3 Noise impact reduction
21.3 Wind farm reactive dispatch
21.3.1 Regular method
21.3.2 Loss minimization
21.3.3 Levelized production cost minimization
21.4 Wind farm layout optimization
21.4.1 Topography and wind direction impact for layout
21.4.2 Layout optimization for minimum levelized production cost
21.5 Conclusion
References
22  Power converters and control of LEDs
22.1 Introduction
22.2 Characteristics of LEDs and drivers
22.2.1 Physical principle of LEDs
22.2.2 Optoelectrical properties of LEDs
22.2.2.1 Current–voltage characteristic
22.2.2.2 Luminous flux emission
22.2.3 Characteristics of drivers for LEDs
22.3 Color control with a power converter
22.4 Efficiency and lifetime improvement
22.5 Current sharing schemes
22.5.1 Passive current sharing schemes
22.5.2 Active current sharing schemes
22.6 Reliability assessment
References
Index
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
Z
Back Cover
Control of Power Electronic Converters and Systems Volume 3
This page intentionally left blank
Control of Power Electronic Converters and Systems Volume 3
Edited by Frede Blaabjerg
Academic Press is an imprint of Elsevier 125 London Wall, London EC2Y 5AS, United Kingdom 525 B Street, Suite 1650, San Diego, CA 92101, United States 50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom Copyright Ó 2021 Elsevier Ltd. All rights reserved No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions. This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).
Notices Knowledge and best practice in this ﬁeld are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary. Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility. To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein. Library of Congress CataloginginPublication Data A catalog record for this book is available from the Library of Congress British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library ISBN: 9780128194324 For information on all Academic Press publications visit our website at https://www.elsevier.com/booksandjournals
Publisher: Brian Romer Acquisitions Editor: Lisa Reading Editorial Project Manager: Joshua Mearns Production Project Manager: Sojan P. Pazhayattil Cover Designer: Greg Harris Typeset by TNQ Technologies
Contents Contributors Preface
1.
xv xvii
Advanced control of power electronic systemsdan overview of methods Chao Wu and Frede Blaabjerg 1.1 1.2
Introduction Linear controller implementation and analysis 1.2.1 Implement PI controller 1.2.2 Implement PR controller 1.2.3 Implement repetitive controller 1.2.4 Pole placement control 1.2.5 Linear quadratic regulator 1.3 Nonlinear controller implementation and analysis 1.3.1 Hysteresis control 1.3.2 Lookup tableebased direct power control 1.3.3 Model predictive control 1.3.4 Backstepping control 1.3.5 Sliding mode control 1.3.6 Feedback linearization 1.3.7 Passivity control 1.3.8 Adaptive control 1.3.9 HN control 1.3.10 Artificial intelligence 1.4 Summary References
2.
1 5 5 6 7 9 9 10 11 11 14 15 19 23 23 24 25 26 28 28
Robust design and passivity control methods Yonghao Gui, Jianzhe Liu, Jan Dimon Bendtsen and Jakob Stoustrup 2.1 2.2
Introduction State space model of power electronic systems 2.2.1 Introduction 2.2.2 Gridconnected voltage source inverter 2.2.3 An example of a VSC (STATCOM)
35 36 37 38 40 v
vi Contents 2.3
Robust controller design 2.3.1 Voltagemodulated direct power control 2.3.2 Tracking controller 2.3.3 Robust control design 2.3.4 Simulation results 2.4 Robust smallsignal stabilization control design 2.4.1 Introduction 2.4.2 Lur’e type model 2.4.3 Robust smallsignal stability analysis problem 2.4.4 Robust stability analysis method 2.4.5 Robust stabilization control design 2.5 Passivitybased control design 2.5.1 Introduction 2.5.2 Portcontrolled Hamiltonian system 2.5.3 Passivitybased control design in a microgrid 2.5.4 Passivity control design in STATCOM 2.6 Conclusions and future prospective Acknowledgment References
3.
42 43 44 46 47 49 49 51 57 58 58 61 62 62 63 67 71 73 73
Sliding mode controllers in power electronic systems Hosein GholamiKhesht, Pooya Davari and Frede Blaabjerg 3.1 3.2 3.3
Introduction System dynamics of a voltage source inverter Introduction to sliding mode control 3.3.1 Reachability condition in the sliding mode control 3.3.2 Control input calculation 3.4 Classical sliding mode control 3.5 Boundarylayer sliding mode control 3.5.1 Adaptive boundarylayer sliding mode control 3.6 Adaptive sliding mode control 3.7 Integral sliding mode control 3.8 Terminal sliding mode control 3.9 Secondorder sliding mode control 3.9.1 Twisting SMC 3.9.2 Supertwisting 3.9.3 Adaptive supertwisting 3.10 Comparison of different SMCs 3.11 Experimental results 3.11.1 Steadystate performance 3.11.2 Dynamic performance 3.12 Conclusions References
77 78 79 81 82 82 85 85 85 87 87 89 89 90 91 91 93 94 96 96 98
Contents vii
4.
Model predictive control of power converters, motor drives, and microgrids Zhenbin Zhang, Tomislav Dragicevic, Yu Li, Yongdu Wang, Changming Zheng, Minrui Leng and Jose Rodriguez 4.1 4.2
Introduction on MPC MPC for permanentmagnet synchronous motor drives 4.2.1 Model of PMSM 4.2.2 FCSMPC for PMSM drives 4.2.3 Performance evaluations of MPC with simulations 4.2.4 Summary 4.3 MPC for microgrids 4.3.1 Dynamics and predictive model of VSCbased MG 4.3.2 MPC for robust and fast operation of an islanded AC MG 4.3.3 Dynamic stabilization of a DC MG using MPC 4.3.4 Performance evaluation with experimental results 4.3.5 Summary 4.4 MPC for renewable energy applications (PMSG wind turbine) 4.4.1 Modeling of PMSG wind turbine system with backtoback power converter 4.4.2 Direct model predictive current control 4.4.3 FCSMPC for threelevel PMSG systems 4.5 Conclusions and future trends References
5.
101 102 106 108 109 109 110 111 112 113 113 115 115 116 118 119 121 122
Adaptive control in power electronic systems Hosein GholamiKhesht, Pooya Davari and Frede Blaabjerg 5.1 5.2 5.3
Introduction System dynamics on a UPS system System identification 5.3.1 Parameters identification 5.3.2 State identification 5.4 Indirect adaptive predictive control 5.4.1 Experimental results 5.5 Model reference direct adaptive control of UPS 5.5.1 The outer voltage control loop 5.5.2 The inner current control loop 5.5.3 Experimental results 5.6 Conclusion References
6.
Machine learning technique for lowfrequency modulation techniques in power converters Amirhossein Moeini, Morteza Dabbaghjamanesh, Tomislav Dragicevic, Jonathan W. Kimball and Jie Zhang
125 127 128 129 130 135 136 136 137 142 142 145 145
viii Contents 6.1 6.2 6.3
Introduction Cascaded Hbridge active power filter configuration ANN for the asymmetric selective harmonic current mitigationPWM 6.4 The proposed technique simulation and experimental results 6.4.1 Simulation results 6.4.2 Experimental results 6.5 Conclusion References
7.
149 151 152 157 157 162 164 165
Overview of stability analysis methods in power electronics Qianwen Xu 7.1 7.2
Introduction Smallsignal stability analysis methods 7.2.1 Modeling of power converter 7.2.2 Eigenvalue method 7.2.3 Impedancebased method 7.2.4 Comparison of methods 7.3 Largesignal stability analysis methods 7.3.1 Timedomain simulations 7.3.2 Lyapunovbased analytical methods 7.4 Case studies with practical examples 7.4.1 Smallsignal stability analysis 7.4.2 Largesignal stability analysis on a power electronic system 7.5 Summary References
8.
169 171 172 174 177 179 180 180 180 183 183 192 195 195
Cyber security in power electronic systems Subham Sahoo 8.1 8.2
8.3
8.4
8.5
Introduction Cyber physical architecture of power electronic converters 8.2.1 Physical stage 8.2.2 Cyber stage Vulnerability analysis of cyber attacks on control of VSCs 8.3.1 Cyber security 8.3.2 Vulnerability assessment Cyber attack detection and mitigation mechanisms in power electronic systems 8.4.1 Detection 8.4.2 Mitigation Test cases 8.5.1 Test case I 8.5.2 Test case II
199 200 201 202 203 203 204 205 205 210 212 212 213
Contents
8.6 Conclusions and future challenges References
9.
ix 217 217
Advanced modeling and control of voltage source converters with LCL filters Bochen Liu, Dao Zhou and Frede Blaabjerg 9.1 9.2
Introduction Modeling of the VSCs with LCL filters 9.2.1 Modeling methods in balanced threephase systems 9.2.2 Modeling methods in unbalanced threephase systems 9.2.3 Simulation examples 9.3 Alternative current control of the VSCs with LCL filters 9.3.1 Control in synchronous reference (dq) frame 9.3.2 Resonance damping technique 9.3.3 Control under unbalanced grid voltages 9.4 Impedancebased stability analysis under weak grid conditions 9.4.1 System control of the LCLfiltered VSCs in ab frame and dq frame 9.4.2 Impedancebased stability analysis References
10.
221 222 222 226 236 238 239 242 251 256 258 262 266
Phaselocked loops and their design Wenzhao Liu and Frede Blaabjerg 10.1 10.2 10.3
10.4
Introduction PLL’s control and design Threephase PLLs 10.3.1 Conventional synchronous reference frame PLL 10.3.2 Moving average filterebased PLLs 10.3.3 Notch filterebased PLLs 10.3.4 Sinusoidal signal integratorebased PLLs 10.3.5 Secondorder generalized integratorebased PLLs 10.3.6 Complex coefficient filterebased PLLs 10.3.7 Delayed signal cancellationebased PLLs 10.3.8 Multiple SRF filterebased PLLs 10.3.9 Other threephase PLLs 10.3.10 Performance comparison and recommendation Singlephase PLLs 10.4.1 Standard PPLLs 10.4.2 Low pass filterebased PPLLs 10.4.3 Moving average filterebased PPLLs 10.4.4 Notch filterebased PPLLs 10.4.5 Doublefrequency and amplitude compensationebased PPLLs 10.4.6 Modified mixer PDBased PPLLs 10.4.7 Transfer delayedebased PLLs
269 270 271 271 277 278 278 279 280 282 283 284 285 287 287 288 288 289 290 291 291
x Contents 10.4.8 Inverse park transformationebased PLLs 10.4.9 Generalized integratorebased PLLs 10.4.10 Synthesis circuit PLLs 10.4.11 Performance comparison and recommendation 10.5 Summary References
11.
292 293 295 296 298 298
Stability and robustness improvement of power converters Xiongfei Wang, Yicheng Liao and Zichao Zhou 11.1 11.2
Introduction Stability and robustness improvement of current control 11.2.1 Smallsignal modeling 11.2.2 Passivitybased stability analysis 11.2.3 Robustness enhancement 11.3 Stability and robustness improvement of outerloop control 11.3.1 Smallsignal modeling 11.3.2 MIMObased stability analysis 11.3.3 Robustness enhancement References
12.
303 304 305 309 313 319 319 326 332 336
High switching frequency threephase currentsource converters and their control Dapeng Lu and Frede Blaabjerg 12.1
Challenges of high switching frequency CSCs control 12.1.1 Multiple timescale dynamics of high switching frequency CSCs 12.1.2 Control methods of high switching frequency CSCs 12.2 Stability analysis of the singleloop DClink current control 12.2.1 Smallsignal modeling 12.2.2 Stable region of singleloop DClink current control 12.2.3 Experimental validation 12.3 Active damping methods for high switching frequency CSCs 12.3.1 Virtual impedance analysis 12.3.2 Experimental validation of active damping 12.4 Summary References
13.
339 339 340 342 342 343 346 348 350 357 364 365
Highpower current source converters Li Ding and Yunwei Li 13.1 13.2
Introduction Current source converters and applications
367 368
Contents
13.2.1 Thyristorbased technology 13.2.2 PWM CSCbased mediumvoltage drives 13.2.3 More potential application 13.3 Parallel CSC system and modulation strategies 13.3.1 Parallel CSC topology 13.3.2 CSC modulation strategies 13.4 Parallel CSC and circuit analysis 13.4.1 CM loop circuit of parallel CSC 13.4.2 DClink circuit of parallel CSC 13.5 DC current balance and CMV reduction methods 13.5.1 SVMbased methods 13.5.2 Carriershifted SPWMbased methods 13.5.3 Case study results 13.6 Conclusions References
14.
xi 368 370 371 373 374 375 381 381 385 387 387 394 397 400 401
Parallel operation of power converters and their filters Ghanshyamsinh Gohil 14.1 14.2
Introduction Circulating current modeling 14.2.1 Parallel converters with a common DC bus 14.2.2 Parallel converters with separate DC bus 14.3 Circulating current control 14.3.1 Current sharing schemes 14.3.2 Droop control scheme 14.3.3 Zero vector dwell time control 14.4 Harmonic performance improvement through interleaved operation 14.4.1 Modulation of parallel interleaved converters 14.4.2 Symmetrically interleaved converters 14.4.3 Harmonic performance evaluation 14.4.4 Nearest three vector modulation 14.5 Circulating current suppression in parallel interleaved converters 14.5.1 Galvanic isolation 14.5.2 Coupled inductor 14.5.3 Commonmode inductor 14.5.4 Integrated inductor 14.6 Summary References
15.
403 405 405 409 410 411 412 415 419 419 419 425 428 430 432 433 437 440 442 444
Advanced power control of photovoltaic systems Ariya Sangwongwanich, Jinkui He and Yiwei Pan 15.1 15.2
Introduction Overview of PV inverter control
447 448
xii Contents 15.2.1 Control structure 15.2.2 MPPT algorithm 15.3 Requirement of advanced control functionality 15.3.1 Grid code 15.3.2 Active power control requirement 15.4 Constant power generation control strategy 15.4.1 Direct power control (PCPG) 15.4.2 Currentlimiting control (ICPG) 15.4.3 Perturb and observeebased control (P&OCPG) 15.5 Benchmarking of constant power generation control strategy 15.5.1 Dynamic responses 15.5.2 Steadystate responses 15.5.3 Tracking error 15.5.4 Stability 15.5.5 Complexity 15.6 Summary References
16.
448 450 453 453 454 455 457 458 460 463 463 466 466 467 467 468 468
Low voltage ridethrough operation of singlephase PV systems Zhongting Tang and Yongheng Yang 16.1 16.2
Introduction Low voltage ridethrough operations 16.2.1 LVRT control using the singlephase PQ theory 16.2.2 LVRT control based on the powervoltage curve 16.3 Reactive power injection strategies under LVRT 16.3.1 Constant average active power control strategy (Const.P) 16.3.2 Constant active current control strategy (Const.Id) 16.3.3 Constant peak current control (Const.Igmax) 16.4 Summary References
17.
471 475 477 482 488 491 492 494 495 496
Gridfollowing and gridforming PV and wind turbines P. Rodriguez and N.B. Lai 17.1 17.2
17.3
Introduction PV and wind turbine systems 17.2.1 PV structures 17.2.2 Grid converter for wind turbine systems Gridfollowing power converters 17.3.1 Definition 17.3.2 Synchronization strategies 17.3.3 Current controllers
499 500 500 501 502 502 503 506
Contents
17.4
Gridforming power converters 17.4.1 Definition 17.4.2 Control schemes for gridforming power converter 17.4.3 Droop control in gridforming power converters 17.4.4 The synchronous power controller 17.5 Conclusions References
18.
xiii 508 508 509 510 516 519 519
Virtual inertia operation of renewables Rasool Heydari, Mehdi Savaghebi and Frede Blaabjerg 18.1 18.2
Introduction Evolution of green energy transition 18.2.1 Lowinertia grid challenges 18.2.2 Control of power convertereinterfaced renewables 18.3 Virtual inertiabased control 18.3.1 Virtual synchronous machine 18.3.2 Concepts and fundamentals 18.4 VSM implementation 18.4.1 NSG penetration level 18.4.2 VSM performance 18.4.3 Fault right through capability 18.5 Summary and future trend References
19.
523 525 525 527 529 530 531 532 533 534 534 536 538
Virtual inertia emulating in power electronicebased power systems Yousef Khayat, Mobin Naderi, Hassan Bevrani and Frede Blaabjerg 19.1 19.2 19.3
Introduction Inertia concept Inertia challenges of power electronicebased power systems 19.4 Adaptive inertia for gridconnected VSGs 19.4.1 VSG principle 19.4.2 Adaptive inertia 19.5 Simulation and experimental results 19.5.1 Simulation results 19.5.2 Experimental results 19.6 Summary References Further reading
20.
541 542 547 549 550 551 553 553 555 557 558 560
Abnormal operation of wind turbine systems Dao Zhou and Mads Graungaard Taul 20.1
Introduction 20.1.1 Classification of grid faults
561 561
xiv Contents 20.1.2 Grid code requirements on lowvoltage ridethrough 20.1.3 Fundamental wind turbine configurations 20.2 Control of type III wind turbine during grid faults 20.2.1 Existing challenges during grid faults 20.2.2 Control strategies during symmetrical grid faults 20.2.3 Control strategies during asymmetrical grid faults 20.3 Control of type IV wind turbine during grid faults 20.3.1 Modeling and control of gridside converter 20.3.2 Symmetrical grid fault control 20.3.3 Asymmetrical grid fault control 20.4 Summary References
21.
562 566 568 568 573 578 582 583 587 593 604 604
Wind farm control and optimization Weihao Hu, Qi Huang, Xiawei Wu, Jian Li and Zhenyuan Zhang 21.1 21.2
Introduction Wind farm active dispatch 21.2.1 Wake effect 21.2.2 Single MPPT and global MPPT 21.2.3 Noise impact reduction 21.3 Wind farm reactive dispatch 21.3.1 Regular method 21.3.2 Loss minimization 21.3.3 Levelized production cost minimization 21.4 Wind farm layout optimization 21.4.1 Topography and wind direction impact for layout 21.4.2 Layout optimization for minimum levelized production cost 21.5 Conclusion References
22.
609 611 611 612 614 618 618 622 625 636 636 641 643 643
Power converters and control of LEDs Xiaohui Qu and Huai Wang 22.1 22.2
Introduction Characteristics of LEDs and drivers 22.2.1 Physical principle of LEDs 22.2.2 Optoelectrical properties of LEDs 22.2.3 Characteristics of drivers for LEDs 22.3 Color control with a power converter 22.4 Efficiency and lifetime improvement 22.5 Current sharing schemes 22.5.1 Passive current sharing schemes 22.5.2 Active current sharing schemes 22.6 Reliability assessment References Index
645 650 650 651 653 654 658 669 669 673 678 685 689
Contributors Jan Dimon Bendtsen, Aalborg University, Aalborg, Denmark Hassan Bevrani, University of Kurdistan, Sanandaj, Kurdistan, Iran Frede Blaabjerg, Department of Energy Technology, Aalborg University, Aalborg, Denmark Morteza Dabbaghjamanesh, Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, TX, United States Pooya Davari, Aalborg University, Aalborg, Denmark Li Ding, Department of Electrical and Computer Engineering, University of Alberta, Edmonton, AB, Canada Tomislav Dragicevic, Department of Electrical Engineering, Technical University of Denmark, Lyngby, Denmark Hosein GholamiKhesht, Aalborg University, Aalborg, Denmark Ghanshyamsinh Gohil, University of Texas at Dallas, Richardson, TX, United States Yonghao Gui, Aalborg University, Aalborg, Denmark Jinkui He, Aalborg University, Aalborg, Denmark Rasool Heydari, University of Southern Denmark, Odense, Denmark Weihao Hu, University of Electronic Science and Technology of China, Chengdu, Sichuan, China Qi Huang, University of Electronic Science and Technology of China, Chengdu, Sichuan, China Yousef Khayat, University of Kurdistan, Sanandaj, Kurdistan, Iran Jonathan W. Kimball, Electrical and Computer Engineering Department, Missouri University of Science and Technology, Rolla, MO, United States N.B. Lai, Luxembourg Institute of Science & Technology, EschSurAlzette, Luxembourg Minrui Leng, School of Electrical Engineering, Southwest Jiaotong University, Chengdu, China Yunwei Li, Department of Electrical and Computer Engineering, University of Alberta, Edmonton, AB, Canada Yu Li, School of Electrical Engineering, Shandong University, Jinan, China Jian Li, University of Electronic Science and Technology of China, Chengdu, Sichuan, China xv
xvi Contributors Yicheng Liao, Aalborg University, Aalborg, Denmark Bochen Liu, Department of Energy Technology, Aalborg University, Aalborg, Denmark Wenzhao Liu, Aalborg University, Aalborg, Denmark Jianzhe Liu, Argonne National Laboratory, Lemont, IL, United States Dapeng Lu, Aalborg University, Aalborg, Denmark Amirhossein Moeini, Electrical and Computer Engineering Department, Missouri University of Science and Technology, Rolla, MO, United States Mobin Naderi, University of Kurdistan, Sanandaj, Kurdistan, Iran Yiwei Pan, Aalborg University, Aalborg, Denmark Xiaohui Qu, Southeast University, Nanjing, Jiangsu, China P. Rodriguez, Luxembourg Institute of Science & Technology, EschSurAlzette, Luxembourg Jose Rodriguez, Faculty of Engineering, Universidad Andres Bello, Santiago, Chile Subham Sahoo, Department of Energy Technology, Aalborg University, Aalborg East, Nordjylland, Denmark Ariya Sangwongwanich, Aalborg University, Aalborg, Denmark Mehdi Savaghebi, University of Southern Denmark, Odense, Denmark Jakob Stoustrup, Aalborg University, Aalborg, Denmark Zhongting Tang, Central South University, Changsha, China Mads Graungaard Taul, Department of Energy Technology, Aalborg University, Aalborg, Denmark Xiongfei Wang, Aalborg University, Aalborg, Denmark Huai Wang, Aalborg University, Aalborg, North Jutland, Denmark Yongdu Wang, School of Electrical Engineering, Shandong University, Jinan, China Chao Wu, Department of Energy Technology, Aalborg University, Aalborg, Denmark Xiawei Wu, University of Electronic Science and Technology of China, Chengdu, Sichuan, China Qianwen Xu, Electric Power and Energy Systems Division, KTH Royal Institute of Technology, Stockholm, Sweden Yongheng Yang, Aalborg University, Aalborg, Denmark Zhenbin Zhang, School of Electrical Engineering, Shandong University, Jinan, China Jie Zhang, Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, TX, United States Zhenyuan Zhang, University of Electronic Science and Technology of China, Chengdu, Sichuan, China Changming Zheng, School of Electrical and Power Engineering, China University of Mining and Technology, Xuzhou, Jiangsu, China Dao Zhou, Department of Energy Technology, Aalborg University, Aalborg, Denmark Zichao Zhou, Aalborg University, Aalborg, Denmark
Preface In 2018 I published two volumes in my edited book series Control of Power Electronic Converters and Systems, which have been very well received by society. Now, 3 years later, it is very clear that such a broad topic has become more and more important as the world has acknowledged the fact that carbon emission from energy consumption is a major contribution to global warming. Countries all over the world have already specified dedicated goals of 2030 and 2050 for carbon emission reduction. One of the important technologies is renewable generation because this technology leaves only the slightest of carbon footprints when electrical power is generated. Further concerns are that more electricity is needed to “fuel” the heavy transportation sector, to make the energy chain more efficient, and to deal with more seasonalbased power generation. These include different levels and sizes of storage as well as performing energy vector coupling to make modern society operate with a safe energy supply. The outcome of all this is that power electronics technology will be increasingly needed to control energy/power throughout the energy chain and different kinds of control are unavoidable. In general, my preference for control technology is to keep it as simple as possible to solve a necessary problem. However, we are now seeing more advanced control methodologies coming into play in industrial applications, which offer a lot of added value. As a consequence, I decided to edit a Volume 3 in the series Control of Power Electronic Converters and Systems, which explores emerging topics in the control of power electronics and converters in different applications, which are all important in the energy transition we are now facing. The book covers the theory behind control, but also practical implementation and operation, which are always important. What is also evident is that in the power grid, controller interactions exist due to increasing renewable energy penetration in the power system and challenges with stability and power quality are beginning to appear. So, in this book, with contributions from all over the world, the focus is on small scale to large scale renewable generation. Also under scrutiny are terminal behavior at the connection to the grid and how to ensure better performance. I have also decided to cover a few important applications seen from a load perspective, which can also be used to control the grid. I will like to thank all my colleagues who worked with me on this book, as well as the contributions from outside Aalborg University. I would also like
xvii
xviii Preface
thank the Villum Fonden for supporting my research as a Villum investigator through the project Reliable Power Electronicbased Power System. Such support is important to be able to realize a book like this. Frede Blaabjerg Aalborg University March 2021
Chapter 1
Advanced control of power electronic systemsdan overview of methods Chao Wu1, Frede Blaabjerg1 1
Department of Energy Technology, Aalborg University, Aalborg, Denmark
1.1 Introduction With the rapid development of semiconductor devices, power electronic technology has also been expanding rapidly and widely used in modern power system. Therefore, the power system is gradually changing toward a power electronicebased power system, which can be seen from these three aspects: power generation, power transmission and distribution grid, and electrical load. For the power generation, environmental and energy crisis call for the extensive application of renewable energy like wind power and photovoltaics. These power generations are highly dependent on power electronic converters to transform the generated electricity to grid [1,2]. For the power transmission and distribution grid, flexible AC transmission systems and static synchronous compensator are needed to improve grid voltage quality, which are also mostly based on power electronic devices [3e5]. On the electrical load side, the wide application of variable frequency drive further increases the proportion of versatile inverters as well as many others like computer centers, lighting, and so on [6,7]. In summary, the modern power system will be with particularly high penetration rate of power electronic devices, so it is of great importance to study how to efficiently and stably control these power electronic equipment. According to different applications, it may have different control objectives such as position control and speed control of electric motor, active and reactive power control of renewable energies, and voltage and current control of diverse power converters. In order to realize these different control objectives, different and reliable control strategies need to be adopted. This chapter will
Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000202 Copyright © 2021 Elsevier Ltd. All rights reserved.
1
2 Control of Power Electronic Converters and Systems
summarize the advanced control technologies in modern power electronic systems. Based on the characteristics of controllers, it can generally be divided into linear controller and nonlinear controller. When it comes to linear controllers, it means that the structure of the controller is linear and the controller design is based on a linear model of the controlled plant. The most commonly used linear controller is proportional and integral (PI) controller, which can track a DC reference without steadystate error [8e11]. The structure is simple and easy to be applied. However, only the DC reference signal can be controlled through a PI controller. Other kinds of reference signals like the sinusoidal reference cannot be tracked without errors. Hence, the proportional and resonant (PR) controller is proposed to track the sinusoidal reference [12e15]. In essence, the resonant controller is an extension of integral control in the frequency domain, which is not just aimed at the zero frequency but can be used at any desired frequency. Through the PR controller, the sinusoidal current or voltage can be controlled in the stationary frame without complex coordinate transformation. However, when the frequency is various, an adaptive resonant controller should be designed, otherwise the control accuracy cannot be guaranteed [16,17]. When the grid voltage is distorted with multifrequency harmonics, a series of resonant controllers should be employed in the control structure which will make the controller design complex and cumbersome to do the discretization in digital controller [18]. In this context, repetitive controller (RC) is a good choice for controlling multifrequency component simultaneously, since the RC is equivalent to a series of resonant controllers in parallel [19e21]. Nevertheless, the delay time in the RC structure may deteriorate the dynamic performance if the basic fundamental frequency is low. Furthermore, the bandwidth and stability of RC should be paid serious attention to when applying it to the digital controller. When it comes to the dynamic performance design process, the pole placement control is an effective method to set all desired poles of system according to the state space equation [22e26]. Sometimes there are special requirements of the control performance, and the linear quadratic regulator (LQR) is an effective method to achieve an optimized control like finite time to the steady point or least energy consumed during dynamic process [27e30]. The LQR is also based on the linear state space equation of the system, and the key point is to find the weighting matrix of state and input, and then calculate the feedback gain matrix. This procedure is dependent on the system parameters, which will jeopardize the system performance with parameter deviations. It should be noted that the application and design of these linear controllers are dependent on the linear model of the controlled plant. The PI controller, resonant controller, and RC are all based on the linear transfer function, and the pole placement control and the LQR are both based on the linear state space equation. Nevertheless, most of the power electronicebased system are nonlinear when considering the power control, synchronization block, switching process, and so on. In order to achieve an accurate and robust
Advanced control of power electronic systems Chapter  1
3
control during the whole operation range, many nonlinear and advanced control methods are taken into consideration in the power electronic systems. Hysteresis control is a simple nonlinear control method with the advantages of easy implementation, strong robustness, and low parameter dependence [31e38]. However, the switching frequency of the hysteresis control is uncertain, so that the theoretical frequency band of the output current can be as high as twice the operating frequency of its hysteresis comparator, which causes unnecessary difficulties in the design of the output filter. Model predictive control (MPC) is an optimization control method which is not limited to the linear system but dependent on how to choose the cost function. Based on the cost function and the system model, the MPC method can directly get the PWM signals without any synchronization block. However, the switching frequency of the basic MPC method is variable which may make the filter design difficult [39e47]. Backstepping (BS) control technique has received much attention due to its merits of systematic and recursive design methodology [48e55]. The idea of BS control strategy is to select the proper Lyapunov function according to control targets on different design stages and form a final Lyapunov function by summing up the previous ones on every stage. The characteristics of the Lyapunov function can guarantee the stability of the control systems. The parameters needed to be tuned are less than PI controller; however, the control performance might have steadystate error when the system parameter changes. In order to enhance the robustness of the BS control, sliding mode control (SMC) is widely used in power electronic systems because of the immunity to system parameters variation [56e61]. The sliding mode variable structure control strategy can change its own control structure in real time according to the change of the operating state of the system, and finally make the state variables of the system enter the predesigned sliding mode surface. However, due to the discontinuous control input for varying, the structure will lead to chatter problem. In order to deal with the nonlinear block in power electronic systems, feedback linearization (FBL) is a good choice to change the nonlinear to linear system by constructing new intermediate variables. The FBL is particularly useful in power electronics to tackle the nonlinear problem. The technique is widely used while developing the control for electric machines, such as the DC machine, induction machine, switched reluctance machine, and permanentmagnet synchronous machine using power electronic converters, where a nonlinear process is transformed into a linear one by forcing the output to follow the input in a closedloop fashion [62e66]. However, FBL control cannot tackle with parameter variations and disturbance uncertainty. In order to enhance the robustness of the control system, the robust control can also be applied to tackle with the uncertain disturbance. The aforementioned controllers are mainly focusing the selfperformance of converters
4 Control of Power Electronic Converters and Systems
without considering the interaction with the grid. The impedance characteristic of converters with different control strategies changes dramatically in a wide frequency range. When it is connected to weak grid, resonance or unstable divergence is frequently occurred due to the impedance interaction. Thus, the passivity control is proposed for regulating the phase of converter impedance between 90 and 90 degrees, which can avoid the instability caused by the impedance interaction [67e73]. When considering the parameter variations during working period and reducing the negative effect caused by parameter changes, adaptive control is an effective way to estimate the state and change the control parameter real time according to the current working status [74e77]. To deal with this stability problem, an HN controller with explicit robustness in terms of grid impedance variations is proposed to incorporate the desired tracking performance and the stability margin [73,79e83]. By properly selecting the weighting functions, the synthesized HN controller exhibits high gains at the vicinity of the line frequency, similar to the traditional PR controller; meanwhile, it has enough highfrequency attenuation to keep the control loop stable. Furthermore, artificial intelligence (AI) is developing rapidly and is one of the most salient research areas during the last several decades [83,84]. The aim of AI is to facilitate systems with intelligence that is capable of humanlike learning and reasoning. There are various applications, including design optimization of power module heatsink [85], intelligent controller for multicolor lightemitting diode [86], maximum power point tracking control for wind energy conversion systems [87,88], anomaly detection for inverter [89], remaining useful life prediction for supercapacitors [90], etc. By implementing AI, power electronic systems are embedded with capabilities of selflearning and selfadaptation, and therefore the system autonomy can be improved. In conclusion, many advanced control methods can be applied in power electronic systems with different objectives, which can be generally divided into two categories: linear controller and nonlinear controller. The overview of these advanced control methods can be seen in Fig. 1.1. This chapter tries to give a comprehensive and systematic review of these advanced control methods, which will start with the linear controller in the first section. Then, various nonlinear controllers are introduced in the next section. In order to give an intuitive interpretation of different control methods, the simple threephase voltage source converter (VSC) is utilized as the controlled plant. In the last section, the comparison of different advanced control methods is compared through different aspects, such as the tracking performance, disturbance rejection ability and complexity of control algorithm, and so on. According to different performance requirements, different advanced control methods should be chosen. Along with the quick development of digital controller and signal processing ability, the advanced nonlinear control method will be a trend for improving the performance of power electronic systems.
Advanced control of power electronic systems Chapter  1
5
FIGURE 1.1 Overview of advanced control methods in power electronic systems.
1.2 Linear controller implementation and analysis Due to the simple structure and easy to be implemented, linear controllers such as PI controller, resonant controller, and RC are widely used in the power electronic systems. It should be noted that an important characteristic of linear controller is that the application and design of linear controller are all based on a linear system. In this section, all the linear controllers are elaborated in details to better understand the advantages and disadvantages of different methods.
1.2.1 Implement PI controller The PI regulator is a linear controller. The input of PI controller is the deviation of a given reference value and a feedback value. The output is a linear combination of the PI of the deviation. The PI controller is the most widely and extensively used controller in the power electronics, since it is simple and stable. The transfer function of conventional PI controller in the continuous domain can be expressed as ki Gpi ðsÞ ¼ kp þ : s
(1.1)
where kp and ki are the proportional and integral gains, respectively. A threephase twolevel VSC is very often used in the power electronic system and it is taken as the controlled plant. The control diagram of PI controller applied for the current control in VSC is shown in Fig. 1.2, where
6 Control of Power Electronic Converters and Systems VSC Zf vdc
C
PCC Ugabc
Igabc abc
PWM modulator θg abc
Vcabc Vcdq dq
dq
θg
Zg
Grid voltage
PLL
Igdq Igdq_ref
PI
FIGURE 1.2 Control diagram of proportional and integral controller applied in a threephase voltage source converter to control the current.
Ugabc is the grid voltage of point of common coupling, Igabc is the grid current, Zf is the impedance of filter which can be a simple L filter or LCL filter, Zg represents the impedance of weak grid, Vcabc is the converter voltage, vdc is the DC voltage, and C is the DC capacitance. As can be seen from Fig. 1.2, the threephase alternative currents in stationary frame need to be changed to DC signals in the synchronous frame, since the PI controller can only achieve a zero steadystate control of DC signal. Thus, the phaselocked loop and coordinate transformation are necessary in this control structure.
1.2.2 Implement PR controller The transfer function of PR controller can be obtained by shifting the integrator part of the conventional PI controller to both positive and negative fundamental frequency: GPR_ideal ðsÞ ¼ kp þ
ki ki 2ki s þ ¼ kp þ 2 : s þ u2 s ju s þ ju
(1.2)
where kp, ki are the proportional and integral parameters of the PI part, u is the tuned resonant frequency. However, it should be pointed out that the PR controller expression in Eq. (1.2) is an ideal one, which is quite sensitive to the grid frequency variation. Thus, the practical PR controller is adopted by the following equation: GPR ðsÞ ¼ kp þ
s2
k r uc s : þ uc s þ u20
(1.3)
where kr is the resonant gain, uc is the resonant bandwidth introduced to increase the control bandwidth and improve the robustness against the grid frequency variation, u0 is the resonant frequency of 50 Hz.
Advanced control of power electronic systems Chapter  1
7
VSC Zf vdc
C
PCC Igabc
PWM modulator Vcabc Vcαβ abc PR αβ
Ugabc Grid voltage
abc αβ Igαβ
Zg
Igαβ _ref
FIGURE 1.3 Control diagram of proportional and resonant controller applied in a threephase voltage source converter to control the current.
The PR controller can be applied for the current control in VSC as illustrated in Fig. 1.3. Since the resonant controller is capable of regulating the alternating signal, the Park transformation can be avoided and the control structure is implemented in the stationary frame.
1.2.3 Implement repetitive controller In some cases, the grid might be distorted by the nonlinear load like a diode bridge. Hence, the distorted harmonic voltage will cause multifrequency harmonic currents. In such case, a single PI controller or PR controller is not enough for dealing with all the harmonics with different frequencies. Under this situation, it is necessary to apply some controllers which are able to deal with multiple frequencies simultaneously. The conventional repetitive controller (CRC) in the continuous domain can be expressed as follows: GCRC ðsÞ ¼ kRC
esT0 : 1 esT0
(1.4)
where kRC is the gain parameter of CRC regulator, T0 is the period of the fundamental control frequency, and T0 ¼ 1/300 s if considering the harmonics caused by diode bridge [21]. Moreover, based on the nature of exponential function, the conventional RC given in Eq. (1.4) can also be written as GCRC ðsÞ ¼
N kRC kRC 2kRC X s þ þ : 2 T0 s T0 n¼1 s2 þ ðnu0 Þ2
(1.5)
From the above expression, it can be concluded that an RC is equivalent to a parallel combination of a negative proportional term, an integral element,
8 Control of Power Electronic Converters and Systems
and an infinite number of resonant controllers connected in parallel at specific harmonic frequencies. This implies that an RC contains lots of resonant frequencies, while a PR controller contains only one. An ideal RC has an infinite openloop gain at the resonant frequencies and can make the control system to track the reference signals without steadystate error. However, it also results in poor stability inevitably and has poor robustness against the deviation of resonant frequencies. To improve the stability and robustness, a lowpass filter Q(s) or a constant Q less than 1 is added to the internal model. When Q is a constant less than 1, the expression of the RC with a certain bandwidth becomes GBRC ðsÞ ¼
kRC QeT0 s : 1 QeT0 s
(1.6)
Based on the nature of exponential function, Eq. (1.6) can be rewritten as follows: GBRC ðsÞ ¼
z
N kRC QeT0 s kRC kRC 2kRC X s þ uc T0 s ¼ 2 þ T s þ T u þ T 2 2 2 1 Qe 0 0 c 0 n¼1 s þ 2uc s þ uc þ ðnu0 Þ
N kRC kRC 2kRC X s þ uc þ þ : 2 T0 s þ T0 uc T0 n¼1 s2 þ 2uc s þ ðnu0 Þ2
(1.7)
When uc is far smaller than u0, the symbol of approximate equal holds. uc is the resonant bandwidth and uc ¼ lnQ/T0. From Eq. (1.7), it can be seen that the integral element in Eq. (1.5) becomes an inertial element, and the resonant controllers become quasiresonant controllers. Therefore, it can be concluded that the modified RC with Q corresponds to a parallel combination of a negative proportional term, an inertial element, and a series of quasiresonant controllers connected in parallel at harmonic frequencies, which can tackle with multifrequency component simultaneously. The control diagram of RC applied for a threephase VSC is shown in Fig. 1.4, which is aimed to Zf vdc
C
PCC Ugabc
Igabc abc PWM modulator θg
abc
Vcabc Vcdq dq
Zg
dq
θg
Igdq PI RC
PLL
Grid voltage
Igdq_ref 0 Igdq
FIGURE 1.4 Control diagram of repetitive controller applied in a threephase voltage source converter to suppress harmonic currents.
Advanced control of power electronic systems Chapter  1
9
suppress the harmonics in grid current. It should be noted that the input of RC is not only limited to the current error but it can also be chosen as power error when the control objective is to suppress power ripple under distorted grid voltages.
1.2.4 Pole placement control Pole placement is a method used in feedback control system theory which allows all the dominant closedloop poles of a plant to be placed in a desired location in the splane. If all the system state variables are considered to be measurable and controllable, then poles of the closedloop system may be placed at any desired location by means of state feedback through an appropriate state feedback gain matrix [22]. The location of the poles directly corresponds to the roots of the system characteristic equation, which controls the characteristics of the response of the system [23]. In this way, the dynamic and steadystate performance can be designed flexibly by regulating the dominant closedloop poles to have a desired damping ratio and an undamped natural frequency [24]. Based on the system data considered and the operating condition, the state equation of the system under study is given by X_ ¼ AX þ BU, where the state feedback control is used, U ¼ KX. The use of state feedback control modifies the system state equation to X_ ¼ ðA BKÞX which is the state variable presentation of the compensated system. The characteristic equation of the compensated system is jsI A þBKj ¼ 0. Here, the state feedback gain matrix K can be determined using the Ackermann’s formula [25], which is 1 K ¼ ½ 0 0 / 1 B AB / An1 B fðAÞ: (1.8) where fðAÞ ¼ An þ a1 An1 þ . þ an1 A þ an I, ai are the coefficients of the desired characteristic polynomial [26]. However, the pole placement control is based on an accurate model of the system. The control performance will be deteriorated if the system parameters are deviated. Furthermore, the pole placement control is only applicable for a linear plant, which might be a limitation for this method, since most power electronic plants are nonlinear.
1.2.5 Linear quadratic regulator LQR is a linear optimal control scheme based on a state space equation. By establishing a linear model and introducing a quadratic integral function of the control objective as an evaluation function, the feedback matrix is acquired based on solving the minimum value of the quadratic integral function. Compared with the proportional integral PI regulator, the control objectives of LQR are more flexible. At present, this LQR method has been initially used in active power filters, wind turbine pitch control, motor speed control, and other control systems [27e30].
10 Control of Power Electronic Converters and Systems
The control object of LQR is a linear system given in the form of state space in modern control theory, and the objective function is a quadratic function of the object state and control input. The optimal design of LQR refers to the state feedback controller K designed to minimize the quadratic objective function J, and K is uniquely determined by the weight matrices Q and R, so the selection of Q and R is particularly important. The LQR theory is the earliest and the most mature state space design method in modern control theory. Especially valuable is that LQR can obtain the optimal control law of state linear feedback, which is easy to form a closedloop optimal control. Moreover, the application of Matlab provides conditions for theoretical LQR simulations, and it also provides convenience for us to achieve stable, accurate, and fast control goals. The LQR method is based on the state space equation, and the linear state space equation of the threephase VSC can be expressed as X_ ¼ AX þ BU:
(1.9)
where X is the state vector, U is the input vector which normally contains the grid voltage and converter voltage. In the LQR method, the optimal state feedback control law is derived as u ¼ KðX XÞ:
(1.10)
X
where is the desired state vector, X is the actual state vector, and K is the feedback gain matrix. The value of K is obtained by solving the Algebraic Riccati equation AT P þ PA þ Q PBR1 BT P ¼ 0:
(1.11)
where Q is a symmetrical positive semidefinite diagonal matrix, which is the weighing matrix of X. R is a symmetrical positive definite diagonal matrix, which is the weighing matrix of control action U. The weights are selected according to the task to be done in this control method. In this way, the control objectives can be set flexibly in LQR method.
1.3 Nonlinear controller implementation and analysis As commonly known, all the power electronic converters are made up of switching components, which are nonlinear blocks, and the controlled plant belongs to a nonlinear system. Thus, it is often not applicable to achieve the performance by simply using a linear control method when encountering some hostile environment such as unbalanced or distorted grid voltage. Furthermore, when considering the system parameter variation and disturbance uncertainty, a nonlinear controller is more suitable for enhancing the robustness and being adaptive to different occasions. In this context, some advanced nonlinear control methods may be necessary in order to improve the performance of power electronic systems. In recent years, the FBL method, BS method, SMC,
Advanced control of power electronic systems Chapter  1
11
adaptive control, and MPC are all applied in the power electronic system to better satisfy the control requirements of the power electronic system in versatile applications.
1.3.1 Hysteresis control The hysteresis current controller is simple to implement and independent of load parameters. But the switching frequency changes according to the load which leads to high switching losses and resonance problems. This limits the usage of hysteresis controllers to low power levels. The control logic for the hysteresis controller per phase is shown in Fig. 1.5. The reference inverter current and the actual inverter current Igabc are compared and the pulses are generated according to the error. The switching logic is expressed as 1; I gabc I abc ref þ h : (1.12) Sabc ¼ 1; I gabc I abc ref h where Sabc is the switching signal, 1 means upper switch is on, 1 means lower switch is on, and h is the hysteresis band. The actual current is made to follow the reference current by forcing the actual current to stay within the hysteresis band.
1.3.2 Lookup tableebased direct power control Direct power control (DPC) is another kind of highperformance control strategy for PWM rectifier based on instantaneous power theory, which was first proposed in Ref. [33] and more clearly presented in Ref. [34]. The basic principle of DPC is similar to direct torque control in motor drives [35]. It directly selects the desired voltage vector from a predefined switching table, according to the grid voltage position (or virtual flux position) and the errors between the reference active/reactive power and feedback value. The internal current loop existing in voltageoriented control is eliminated in DPC. As a result, DPC features very quick dynamic response with simple structure.
VSC Zf vdc
C
PCC
Zg
Igabc Iabc_ref abc dq
Igdq _ref θg
Grid voltage PLL
Hysteresis controller FIGURE 1.5 Hysteresis control scheme applied in a threephase voltage source converter.
12 Control of Power Electronic Converters and Systems VSC vdc
PCC
Sp Sq
Ugabc Igabc abc abc Grid αβ voltage αβ Igαβ Ugαβ Power Calculation (1.13) P Q Pref Hysteresis Qref controller (1.16)
C
Switch table 1.1 θn
Sp Sq Ugαβ
Zg
Zf
FIGURE 1.6 Control diagram of lookup tableebased direct power control applied in a threephase voltage source converter [34].
The control structure of a typical LUTDPC is shown in Fig. 1.6, where the switch table is shown in Table 1.1. The power calculation block can be expressed as 8 3 3 > > < P ¼ ðU gab $I gab Þ ¼ ðuga iga þ ugb igb Þ 2 2 : (1.13) > 3 > : Q ¼ ðUgab I gab Þ ¼ 3 ðugb iga uga igb Þ 2 2 where Ugab ¼ uga þ jugb and Igab ¼ iga þ jigb. In order to obtain the position of grid voltage vector, the angle of grid voltage can be calculated based on uga and ugb as q ¼ arctanðugb = uga Þ:
(1.14)
The switch table is based on the position of grid voltage vector. In order to optimize the performance of converter, the grid voltage can be divided into 12 sectors. The angle of 12 sectors can be expressed as p p (1.15) ðn 2Þ qn ðn 1Þ ; n ¼ 1; 2; /; 12: 6 6 The power hysteresis controller consists of active power and reactive power hysteresis controller. The inputs are the power error between the power reference and the real power. The output reflects the status of power deviation, which can be expressed as Sp and Sq. Sp and Sq only have two status, which can be expressed as 0 P < Pref Hp 0 Q < Qref Hq Sp ¼ ; Sq ¼ : (1.16) 1 P > Pref þ Hp 1 Q > Qref þ Hq where Hp and Hq are the hysteresis band. Although LUTDPC has been considered as a powerful and robust control scheme for PWM rectifier, highpower ripples and variable switching
Sp
Sq
q1
q2
q3
q4
q5
q6
q7
q8
q9
q10
q11
q12
1
0
101
111
100
000
110
111
010
000
011
111
001
000
1
1
111
111
000
000
111
111
000
000
111
111
000
000
0
0
101
100
100
110
110
010
010
011
011
001
001
101
0
1
100
110
110
010
010
011
011
001
001
101
101
100
Advanced control of power electronic systems Chapter  1
TABLE 1.1 Switch table applied in lookup tableebased direct power control.
13
14 Control of Power Electronic Converters and Systems
frequency are two of the most notable drawbacks of conventional LUTDPC. Furthermore, the required sampling frequency is usually very high in order to achieve relative satisfactory performance, which increases the hardware burden and might limit the application of this method.
1.3.3 Model predictive control As a more advanced control theory, MPC has received extensive attention in the field of power electronic systems in recent years. In MPC, a cost function should first be defined according to an expected control target, and then calculate and compare the cost function when different voltage vectors are applied in each control cycle. Then, the optimal voltage vector is selected for the system control in a specific sample period. Compared with the traditional lookup table method, this vector selection method based on the cost function is more accurate in selecting the voltage vector due to the great flexibility in the selection of the cost function. The objectives that can be achieved by this control strategy are very diverse not only to achieve conventional power or current control but it can also optimize other performances such as reducing switching loss, mitigating power/torque ripple, etc. In addition, by adjusting the weighting factors of different variables in the cost function, different control performances can also be achieved. In Ref. [39], the author designed a control algorithm for the active and reactive power error in the cost function. The weighting factors are adjusted in real time, optimizing the dynamic performance of the entire system. Some studies aimed at the effect of system delay in the control process on the control accuracy of the MPC strategy and proposed a twostep prediction method to compensate for the delay error and thereby effectively reduce the current harmonics and power ripple [40e43]. Other studies have proposed a method to optimize the control accuracy of the MPC strategy by adding a duty cycle modulation. This control strategy adds zero vectors to modulate on the basis of the optimal effective vector selected in each control cycle. The execution time of the effective vector and the zero vector is calculated through the deadbeat principle, so that the control error at the end of each control cycle is effectively suppressed, and the overall control performance of the control system is optimized [44e48]. However, the MPC strategy based on the finite set needs to calculate and compare the objective functions of all voltage vectors in each control cycle, which demands a large amount of calculation and requires high performance of the processor when the sampling frequency is high and also to achieve a high bandwidth. In summary, some open questions for the MPC can be concluded as follows: 1. When the cost function is a combination of multitarget optimization, how to choose the weighting factor? 2. How to ensure the switching frequency is constant of MPC? 3. How to reduce the effect of parameter deviation on the
Advanced control of power electronic systems Chapter  1
15
control accuracy of MPC? Such detailed discussions about MPC methods will be addressed in Chapter 4 of this book.
1.3.4 Backstepping control The BS control technique has received much attention due to its merits of systematic and recursive design methodology. The idea of BS control strategy is to select a proper Lyapunov function according to the control targets in different design stages and form a final Lyapunov function by summing up the previous ones in every stage. The characteristics of the Lyapunov function can guarantee stability of the control systems. The BS control strategy has been widely used in aeronautical and astronautical systems [49e51], and satisfied control performance is achieved. In addition, the BS control technique can also be applied to power electronic system [52e55]. In order to take an intuitive example of BS applied in power electronic system, a BSDPC strategy is designed for the gridconnected AC/DC converter. The main objective for the BSDPC strategy is to maintain the DCbus voltage of the AC/DC converter to be a given value by regulating the active power. The other two objectives are the control of active and reactive powers to the given reference. Since the threephase VSC is working as a rectifier in this case, the reference direction of the grid current is shown in Fig. 1.7. For simplification, the inductance filter is applied. Lf is the inductance of filter and Rf is the resistance of filter. According to Fig. 1.7, the relationship between the grid voltages, converter voltages, and the grid currents in the stationary ab frame can be expressed as U gab ¼ I gab Rf þ Lf
dI gab þ V cab : dt
(1.17)
VSC Zf vdc
C
PWM modulator vcabc abc αβ
vcαβ
PCC
Zg
Ugabc Igabc abc abc Grid αβ voltage αβ Igαβ Ugαβ Power calculation (1.19) pg qg pgref Backstepping qgref Controller (1.39)
FIGURE 1.7 Control scheme of backstepping control applied in a threephase voltage source converter [55].
16 Control of Power Electronic Converters and Systems
In the stationary ab frame and for a balanced threephase system, the instantaneous active and reactive power outputs, seen from the grid side, can be defined as pg þ jqg ¼
3 U gab bI gab : 2
(1.18)
where ˆ denotes the conjugate of the complex vector. In such case, pg and qg can be deduced as 3 pg ¼ ðuga iga þ ugb igb Þ 2 3 qg ¼ ðugb iga uga igb Þ: 2
(1.19)
Differentiating Eq. (1.19) results in the instantaneous active and reactive power variations as dpg 3 diga duga digb dugb ¼ þ iga þ ugb þ igb uga 2 dt dt dt dt dt (1.20) dqg 3 diga dugb digb duga ¼ þ iga uga igb ugb : 2 dt dt dt dt dt Assuming the grid voltage is an ideal sinusoidal wave form, namely, uga ¼ Ug sinðu1 tÞ ugb ¼ Ug cosðu1 tÞ:
(1.21)
The instantaneous grid voltage variation can be obtained as duga ¼ u1 Ug cosðu1 tÞ ¼ u1 ugb dt dugb ¼ u1 Ug sinðu1 tÞ ¼ u1 uga : dt
(1.22)
Based on Eq. (1.17), the instantaneous grid current variations can be expressed as diga 1 ¼ ðuga iga Rf vca Þ Lf dt digb 1 ¼ ðugb igb Rf vcb Þ: Lf dt Substituting Eqs. (1.22) and (1.23) into Eq. (1.20) yields i
dpg 3 h 2 Rf ¼ uga þ u2gb ðuga vca þ ugb vcb Þ u1 qg pg 2Lf dt Lf dqg 3 Rf ¼ ðuga vcb ugb vca Þ þ u1 pg qg : 2Lf dt Lf
(1.23)
(1.24)
Advanced control of power electronic systems Chapter  1
17
The relationship between DCbus voltage and active power can be written as C
dvdc pg ¼ iload : dt vdc
(1.25)
where iload is the load current and C is the value of capacitor. For the whole design procedures of the nonlinear BS controller, there are two steps in order to force the system states to track the desired reference commands. The first step is to design the DCbus voltage BS controller to control the vdc, and the output is regarded as the reference value of active power pref g . The second step is to design the power BS controller to control the ref active and reactive powers according to pref g and qg (which is always set to zero for unity power factor control). The BSDPC design procedures are described as follows: The tracking error of DCbus voltage ev is defined as ev ¼ vref dc vdc : where
vref dc
is the reference value of DCbus voltage. dev dvdc 1 pg ¼ ¼ iload : C vdc dt dt
(1.26)
(1.27)
Then, a Lyapunov function V1 is defined as 1 V1 ¼ e2v : 2 The differentiation of V1 can be calculated as dV1 dev e v pg ¼ ev ¼ iload : dt dt C vdc
(1.28)
(1.29)
The active power tracking error ep is introduced as ep ¼ pref g pg :
(1.30)
The active power reference can be designed as pref g ¼ ðkv Cev þ iload Þvdc :
(1.31)
where kv is the BS parameter for the DCbus voltage control loop. Substituting Eq. (1.31) into Eq. (1.29), the differentiation of V1 can be rewritten as dV1 ev ep ¼ kv e2v þ : dt Cvdc
(1.32)
18 Control of Power Electronic Converters and Systems
As the tracking error ep tends to be zero for an appropriate control, it is obvious that the differentiation of V1 will be negative; thus, the tracking error ev will tend to be zero according to the Lyapunov stability analysis. The tracking error of reactive power is defined as eq ¼ qref g qg :
(1.33)
According to Eq. (1.24), the differentiation of the power tracking errors (ep and eq) can be represented by 8 ref ref
> > dep dpg dpg dpg 3 2 > > ¼ ¼ ðu v þ u v Þ þ u 1 qg U > ga ca gb cb g < dt 2Lg dt dt dt : (1.34) > > de dq 3 > q g > > : dt ¼ dt ¼ 2L ðuga vcb ugb vca Þ u1 pg g where Ug2 ¼ u2ga þ u2gb is used for simplicity. Define a Lyapunov function V2 to realize the global tracking control of DCbus voltage and power as
1 V2 ¼ V1 þ e2p þ e2q : (1.35) 2 The differentiation of V2 can be deduced as dV2 dV1 dep deq ev dep deq ¼ þ ep þ eq ¼ kv e2v þ ep þ þ eq : dt dt dt dt Cvdc dt dt For the purpose of (dV2/dt) < 0, it can be designed as follows 8 ev dep > > > < Cv þ dt ¼ kp ep dc : > deq > > : ¼ kq ep dt
(1.36)
(1.37)
where kp and kq are the BS parameters of the active power control loop and the reactive power control loop. Since the differentiation of V2 is a negative definite function, it implies that ev, ep, and eq go to zero asymptotically. According to the Lyapunov stability theorem, the stability of the proposed BSDPC strategy can be guaranteed. Substituting Eq. (1.37) into Eq. (1.34), we can obtain 8 ref i dpg > 3 h 2 ev > > ðu v þ u v Þ u q U ¼ kp ep > ga ca gb cb 1 g < dt 2Lg g Cvdc : (1.38) > > 3 > > ðuga vcb ugb vca Þ þ u1 pg ¼ kq eq : 2Lg
Advanced control of power electronic systems Chapter  1
19
It can be concluded from Eq. (1.38) that the BSDPC law for the AC/DC converter can be designed as " !
dpref 2Lf ev g ref ref vca ¼ p þ k q p þ q u k u ga p g gb q g g g Cvdc dt 3Ug2
8 > > > > > >
> v > > > > :
# u1 uga qg þ u1 ugb pg þ uga
cb
" !
dpref 2Lf ev g ref ref ¼ p þ k q p þ q u k þ u gb p g ga q g g g Cvdc dt 3Ug2
:
# u1 ugb qg u1 uga pg þ ugb (1.39)
As can be seen from Fig. 1.7, the advantages of BS method are that the control structure is pretty simple and having good steadystate and dynamic performance. Compared with the conventional PI controller, only one parameter needs to be tuned which can simplify the parameter design process. However, the BS control method is highly dependent on the system parameters, which will cause steadystate error if the parameter deviation is large.
1.3.5 Sliding mode control As a nonlinear control strategy, the SMC strategy has excellent dynamic performance and good parameter robustness and has been widely studied in the field of power electronic systems. The main idea of the sliding mode variable structure control strategy is to design the sliding mode surface and the approaching law according to the desired dynamic characteristics of the system, so that the system state moves from outside the sliding mode surface to the sliding mode surface, which is achieved by discontinuous control. Once the system reaches the sliding mode surface, it is not affected by the system parameters and external disturbances and moves along the sliding mode surface, effectively solving the problem of poor robustness of the BS control strategy. But the sliding mode variable structure control strategy might produce chattering problems because of the introduction of the switching function in discontinuous control, which will affect the control accuracy of the system. In Ref. [57], a sliding mode variable structure DPC strategy is applied to a threephase DC/AC converter. The design of the sliding mode surface selects the form of PI of power error. The SMC strategy with a variable control structure is based on the design of discontinuous control signal that drives the system operation states toward
20 Control of Power Electronic Converters and Systems
special manifolds in the state space. These manifolds are chosen in such a way that the control system will have the desired behavior as the states converge to them. In order to understand the SMC principle well, an SMC scheme for regulating the instantaneous active and reactive powers of gridconnected threephase DC/AC converter is elaborated as an example. The first step is to design the sliding surface. The DC voltage of the threephase PWM converters is assumed constant, and the control objectives are to track the predefined active power and reactive power accurately and quickly. Thus, the sliding surface is based on the power error. In order to maintain the enhanced transient response and minimize the steadystate error, the switching surfaces can be in the integral forms as SQ T : Z t 8 ep ðsÞds > < SP ¼ e p þ K p 0 : Z t > : SQ ¼ e q þ K q eq ðsÞds S ¼ ½ SP
(1.40)
(1.41)
0
where ep ¼ pg and eq ¼ qg are the respective errors between the references and the actual active and reactive powers. Kp and Kq are the positive control gains. The manifolds SP ¼ 0 and SQ ¼ 0 represent the precise tracking of converter’s active and reactive powers. The second step is to design the approaching law of SMC, namely, how to ensure the states can come to the sliding with arbitrary initial states. The task is to force the system state trajectory to the interaction of the switching surfaces as aforementioned. In this example, an SMC scheme is suggested to generate the converter output voltage reference as the input to SVM module. Based on Eq. (1.41), the differential of the sliding surface can be expressed as pref g
qref g
dSP dep d ¼ þ Kp ep ¼ pg þ Kp pref g pg dt dt dt
dSQ deq d ¼ þ Kq eq ¼ qg þ Kq qref q g : g dt dt dt
(1.42)
Substituting Eq. (1.24) into Eq. (1.42) leads to dS ¼ F þ DV g : dt where F ¼ ½ FP
FQ T , V g ¼ ½ vga
vgb T
(1.43)
Advanced control of power electronic systems Chapter  1
2 u 3 4 ga D¼ 2Lf u gb
ugb
21
3 5
uga
FP ¼
R
3 2 f p uga þ u2gb þ pg þ u1 qg þ KP pref g g 2Lf Lf
FQ ¼
Rf qg u1 pg þ Kq qref g qg : Lf
(1.44)
In the SMC, a Lyapunov approach is usually used for deriving conditions on the control law that will drive the state orbit to the equilibrium manifold. The quadratic Lyapunov function is selected as W ¼ 0:5ST S 0:
(1.45)
The time derivative of W on the state trajectories of Eq. (1.43) is then given by dW 1 T dS dST ¼ þS (1.46) S ¼ ST ðF þ DV g Þ: dt 2 dt dt The approaching law must be properly chosen so that the time derivative of W is definitely negative with Ss0. Therefore, the following control law is selected as
FP KP1 sgnðSP Þ 0 1 V cab ¼ D þ : (1.47) 0 KQ1 sgnðSQ Þ FQ where KP1 and KQ1 are the positive control gains, sgn(SP) and sgn(SQ) are the respective switch functions for active and reactive powers. The robustness analysis, in practical operation, the sliding surface S will be affected by the parameter variations, AD sample errors and measurement noise, etc. Thus, Eq. (1.43) should be rearranged as dS ¼ S þ DV g þ H: dt where H ¼ ½ HP HQ T represents the system disturbances. In this way, Eq. (1.46) can be rewritten as
HP KP1 sgnðSP Þ 0 dW T dS T ¼S ¼S : dt dt HQ 0 KQ1 sgnðSQ Þ
(1.48)
(1.49)
It is worth noting that if the positive control gains fulfill the following condition, KP1 > HP  and KQ1 > HQ, the time derivative of Lyapunov function dW/dt is still definitely negative. Therefore, the SMC obviously features strong robustness.
22 Control of Power Electronic Converters and Systems
Remedy of the Power Chattering Problem: The SMC scheme developed earlier guarantees a fast tracking of instantaneous active and reactive powers of gridconnected DC/AC converters. However, fast switching may generate unexpected chattering, which may excite unmodeled highfrequency system transients and even result in unforeseen instability. To eliminate this problem, the discontinuous part of the controller is smoothed out by introducing a boundary layer around the sliding surface. As a result, a continuous function around the sliding surface neighborhood is obtained as 8 > 1; if Sj > lj > > > >
l j > > > > : 1; if Sj < lj where lj > 0 is the width of the boundary layer and j represents pg and qg, respectively. According to Eq. (1.47), the converter output voltage reference is obtained in the stationary reference frame and can be directly transferred to the SVM module for generating the required switching voltage vectors and their respective duration times. The control diagram of SMC applied in VSC can be shown in Fig. 1.8. The improved sign function is designed for approaching law, which effectively reduces the chattering problem of the system. This sliding mode variable structure DPC strategy can be realized in a static coordinate system, which has good steadystate and dynamic performance and has strong
VSC Zf vdc
PCC
C
SVM modulator Vcαβ
Vcαβ calculation based on (1.47)
FP FQ
Zg
Ugabc Igabc Grid abc abc αβ voltage αβ Igαβ Ugαβ Power calculation (1.19) qg pg pgref FP and FQ calculation qgref based on (1.44)
SMCbased DPC FIGURE 1.8 Control diagram of sliding mode control applied for power control in a threephase voltage source converter [57].
Advanced control of power electronic systems Chapter  1
23
parameter robustness. However, how to choose the appropriate sliding mode surface and how to design the approaching law are various, which will be thoroughly discussed in Chapter 3 for different applications.
1.3.6 Feedback linearization FBL is a common nonlinear control approach, whose main idea is to transform a nonlinear system into a fully or partially decoupled linear one by means of state feedback and nonlinear transformation, so that linear control strategies can be used. The basic theory of the FBL of nonlinear system is to linearize the nonlinear system in a wide range by choosing a proper nonlinear transformation z ¼ T(x) and a nonlinear state feedback variable v ¼ a(x)þb(x)u. At the same time, for a multivariable nonlinear system, decoupled control can also be realized [62]. Such linearization has higher accuracy, since the linear approximation is not employed, which means that no higher order nonlinear terms are ignored. Recently, the FBL method has been widely used in power electronics, motor control, distributed generation, and other fields [63e66]. In Ref. [63], an adaptive inputeoutput FBLbased torque control of synchronous reluctance motor is proposed without any mechanical sensor. In Ref. [66], the dynamic performance of the system is improved by configuring zeros and poles of the feedbacklinearized system. Although the FBL method can transform the nonlinear system to a linear system and using the linear control method, it is based on the accurate system model without considering the parameter deviations or disturbance uncertainty.
1.3.7 Passivity control The passivitybased control of VSCs recently emerges as a promising way to tackle the instability challenge and achieve robustness. The concept of passivity in the frequency domain implies that the real part of the output impedance/admittance of VSC is nonnegative. To keep the real part of the VSC output admittance always larger than zero, which can maintain the phase of output admittance belong to 90 from 90 degrees. Since the grid impedance is always passive, the interaction between the VSC and the grid would always be stable if the output admittance of VSC is controlled to be passivity. The first step of passivity control is to get the impedance/admittance model of the VSC, then trying to control the real part of the impedance/admittance to be larger than zero, which can be expressed as ReðGVSC ðjuÞÞ > 0:
(1.51)
where GVSC(ju) represents the frequency characteristic of VSC impedance/ admittance, which is a complex variable and the real part denotes
24 Control of Power Electronic Converters and Systems
resistance/conductance. If the real part is above zero, which means the VSC is passive and will be stable even under weak grid conditions. It should be noted that the passivity control of VSC is a sufficient condition but not a necessary condition for the stable operation under weak grid. The detailed discussions of the passivity control can be seen in Chapter 2 of this book.
1.3.8 Adaptive control Most of the aforementioned control methods are based on the model parameters without considering the parameters uncertainty and deviation due to different operation points. Large parameter deviations can be detrimental for the system performance like steadystate error or oscillating dynamic process. In this context, adaptive control methods have attracted much attention, since it is robust to parameter variations or other uncertainties. Adaptive control is a control algorithm with adjustable control gains that are adaptive to variable or initially uncertain parameters of a controlled system [74e76]. Hence, this control method can keep the system performance and stability at the desired or optimum level under different conditions. The model reference adaptive system (MRAS) is the mostly commonly applied control structure which is used for acquiring the rotor speed or position in motor drive field. The classical rotor flux MRAS speed observer is shown in Fig. 1.9, where us and is represent the stator voltage and stator current vector, fr is the actual rotor flux acquired from the reference model, f^r is the estib r is the mated rotor flux vector obtained from the adaptive model, and u estimated rotor speed. The steps of MRAS control can be elaborated as follows: a reference model and an adaptive model are firstly constructed, then the error between these two models is calculated, and an appropriate controller is lastly designed to regulate the error to be zero. There are also other types of adaptive control structures, which can be seen from Chapter 5 in this book.
us is
Reference model
Ir
Adaptive model
Iˆr
Zˆ r
Adaptive mechanism
FIGURE 1.9 Classical rotor flux model reference adaptive system speed observer for an induction motor [77].
Advanced control of power electronic systems Chapter  1
25
1.3.9 HN control The HN control has been introduced in the early 1980s and opened a new direction in robust control design. Recently, this approach has been applied to the control of active power filters [73,78]. The feature is that the quantitative measure (HNnorm) of robustness may be explicitly included in the design goal. But its application to practical design problems was not so popular as the conventional control methods such as PI and pole assignment because the degree of the resulting controller is usually too high to be implemented by analog circuits. Moreover, the synthesis of the optimal controller involves complex manipulation on the system transfer functions, and a general and reliable numerical algorithm was not available. Although the state space solution approach [79] provides an attractive way to synthesize the controller by solving two Riccati equations, it is restricted to the socalled regular case where transfer function of the plant suffers from some inherent limitations. Recently, Gahinet and Apkarian [80] proposed the LMI approach which relaxes the above restrictions and provides a more general and numerically sound way to synthesize the controller. The design of loopshaping controller based on HN control method is displayed in Fig. 1.10, in which P(s) is the augmented plant obtained by appending the weighting function to the output of the transfer function of the desired loop shapes, w is in the original input, u is the additional input to enhance the system performance, v is the output, H(s) is the plant, D(s) represents the system uncertainty, and W(s) is the system weighting function. The design goal is to synthesize the stabilizing controller K(s) so that the HN gain from w to z is less than 1 [82]. The HN method is a robust control method; however, the derived controller and weighting function always are high orders which will increase the computation burden of the digital controllers.
Δ(s) P(s) w
v
H(s)
W(s)
z
u K(s) FIGURE 1.10 Control configuration for generalized HN control [82].
26 Control of Power Electronic Converters and Systems
1.3.10 Artificial intelligence The AI contains many smart algorithms, one of them mostly used is the neural network (NN). The main advantages of NN are parallel processing, learning ability, robustness, and generalization. They can be not only effectively used for current controller but also for other functions in a controller structure. NNbased controller can be named as a blackbox technique, and it can approximate a wide range of nonlinear functions to arbitrary accuracy. An exemplary application of NNbased current vector control for a threephase voltage converter is shown in Fig. 1.11 [91]. The NN is applied for generating the converter voltage reference. The inputs of the NN are the grid current, the error of grid current, and the integration of error. The action network in Ref. [91] had two hidden layers of six nodes each and 2 output nodes, and shortcut connections between all pairs of layers, with hyperbolic tangent functions at all nodes. Even though the NNbased controller possesses several advantages such as robustness, being model free, also adaptive, universal approximation, etc., however, it needs massive training data in order to obtain accurate results. There are also other intelligent control methods, like fuzzy control, reinforcement control [92e95], which are elaborated in detail in more details in Chapter 6 of this book.
VSC Zf vdc
C
Ugabc
Igabc Igdq PWM modulator
θg abc
abc dq
θg
Hidden
Vcabc Vcdq
Zg
PCC
Output
Grid voltage
PLL Igdq
Input Igdq_ref
dq
³ FIGURE 1.11 Control diagram of neural network applied in a threephase voltage source converter for current control [91].
TABLE 1.2 The comparison between different advanced control methods applied in power electronic systems. Algorithm complexity
Parameter dependency
Robustness to disturbance
Easy to design
Linear controller
PI
DC
Low
Low
High
Poor
Easy
Resonant controller
AC
Low
Low
High
Poor
Easy
RC
AC
Low
Medium
High
Poor
Easy
PPC
DC
Medium
Medium
High
Poor
Medium
LQR
DC
High
High
High
Poor
Medium
Nonlinear controller
Hysteresis control
AC and DC
High
Low
Low
Medium
Easy
LUTDPC
AC and DC
Low
Low
Medium
Medium
Easy
MPC
AC and DC
High
Medium
Medium
Medium
Medium
BS
AC and DC
High
High
High
Poor
Medium
SMC
AC and DC
High
Medium
Low
Good
Medium
FBL
AC and DC
High
Medium
Medium
Medium
Medium
Passivity control
AC and DC
Medium
Medium
Medium
Good
Medium
Adaptive control
AC and DC
High
High
Low
Good
Complex
HN control
AC and DC
High
High
Low
Good
Complex
AI
AC and DC
High
High
Low
Good
Complex
27
Control flexibility
Advanced control of power electronic systems Chapter  1
Reference signal
Control methods
28 Control of Power Electronic Converters and Systems
1.4 Summary In conclusion, there are many different advanced control methods applied in power electronic systems. Different control performances can be achieved through different methods under diverse application. In order to make a systematic and comprehensive comparison of these advanced control methods, Table 1.2 compares the performance of each control method from five aspects: reference signal, control flexibility, algorithm complexity, parameter dependency, and robustness to disturbance. There is no universal principle or criterion for choosing each control method. How to choose control methods should be analyzed according to the specific applications. For example, when a VSC is working in grid connection, the conventional linear controller is capable enough for tracking the reference, like the PI controller or PR controller. Nevertheless, for the standalone application like the uninterruptible power supply system, the load is unknown which can change in a wide range and in such situation, the advanced robust control like HN controller is more suitable for operating well under wide range disturbances and uncertainties.
References [1] F. Blaabjerg, Y. Yang, D. Yang, X. Wang, Distributed powergeneration systems and protection, Proc. IEEE 105 (7) (July 2017) 1311e1331. [2] F. Blaabjerg, K. Ma, Future on power electronics for wind turbine systems, IEEE J. Emerg. & Sel. Top. Power Electron. 1 (3) (September 2013) 139e152. [3] F.B. Alhasawi, J.V. Milanovic, Technoeconomic contribution of FACTS devices to the operation of power systems with high level of wind power integration, IEEE Trans. Power Syst. 27 (3) (August 2012) 1414e1421. [4] M. Nagpal, C. Henville, Impact of powerelectronic sources on transmission line ground fault protection, IEEE Trans. Power Deliv. 33 (1) (February 2018) 62e70. [5] L.K. Haw, M.S.A. Dahidah, H.A.F. Almurib, A new reactive current reference algorithm for the STATCOM system based on cascaded multilevel inverters, IEEE Trans. Power Electron. 30 (7) (July 2015) 3577e3588. [6] E.B. Agamloh, Power and efficiency measurement of motorvariablefrequency drive systems, IEEE Trans. Ind. Appl. 53 (1) (JanuaryFebruary 2017) 766e773. [7] M. Malekpour, R. AzizipanahAbarghooee, F. Teng, G. Strbac, V. Terzija, Fast frequency response from smart induction motor variable speed drives, IEEE Trans. Power Syst. 35 (2) (March 2020) 997e1008. [8] J. Dannehl, F.W. Fuchs, P.B. Thogersen, PI state space current control of gridconnected PWM converters with LCL filters, IEEE Trans. Power Electron. 25 (9) (September 2010) 2320e2330. [9] J. Selvaraj, N.A. Rahim, Multilevel inverter for gridconnected PV system employing digital PI controller, IEEE Trans. Ind. Electron. 56 (1) (January 2009) 149e158. [10] F.D. Freijedo, A. Vidal, A.G. Yepes, J.M. Guerrero, O. Lopez, J. Malvar, J. DovalGandoy, Tuning of synchronousframe PI current controllers in gridconnected converters operating at a low sampling rate by MIMO root locus, IEEE Trans. Ind. Electron. 62 (8) (August 2015) 5006e5017.
Advanced control of power electronic systems Chapter  1 [11] [12] [13]
[14]
[15]
[16]
[17]
[18]
[19]
[20] [21]
[22]
[23] [24]
[25] [26]
[27] [28]
29
M.P. Kazmierkowski, L. Malesani, Current control technique for threephase voltagesource PWM converters: a survey, IEEE Trans. Ind. Electron. 45 (5) (October 1998) 691e703. D.N. Zmood, D.G. Holmes, Stationary frame current regulation of PWM inverters with zero steadystate error, IEEE Trans. Power Electron. 18 (3) (May 2003) 814e822. M. Castilla, J. Miret, J. Matas, L.G. Vicun˜a, J.M. Guerrero, Control design guidelines for singlephase gridconnected photovoltaic inverters with damped resonant harmonic compensators, IEEE Trans. Ind. Electron. 56 (11) (November 2009) 4492e4501. F. Wang, M.C. Benhabib, J.L. Duarte, M.A.M. Hendrix, Sequence decoupled resonant controller for threephase gridconnected inverters, in: In Proc. IEEE APEC, Washington, DC, February 2009, pp. 121e127. D.G. Holmes, T.A. Lipo, B.P. McGrath, W.Y. Kong, Optimized design of stationary frame three phase AC current regulators, IEEE Trans. Power Electron. 24 (11) (November 2009) 2417e2426. S. Go´mez Jorge, C.A. Busada, J.A. Solsona, Frequencyadaptive current controller for threephase gridconnected converters, IEEE Trans. Ind. Electron. 60 (10) (October 2013) 4169e4177. S. Golestan, E. Ebrahimzadeh, J.M. Guerrero, J.C. Vasquez, An adaptive resonant regulator for singlephase gridtied VSCs, IEEE Trans. Power Electron. 33 (3) (March 2018) 1867e1873. J.B. Hu, Y.K. He, Multifrequency proportionalresonant (MFPR) current controller for PWM VSC under unbalanced supply conditions, J. Zhejiang Univ.  Sci. 8 (10) (2007) 1527e1531. D. Chen, J. Zhang, Z. Qian, Research on fast transient and 6n 1 harmonics suppressing repetitive control scheme for threephase gridconnected inverters, IET Power Electron. 6 (3) (March 2013) 601e610. M. Abusara, S. Sharkh, P. Zanchetta, Adaptive repetitive control with feedforward scheme for gridconnected inverters, IET Power Electron. 8 (8) (August 2015) 1403e1410. Y. Song, H. Nian, Sinusoidal output current implementation of DFIG using repetitive control under a generalized harmonic power grid with frequency deviation, IEEE Trans. Power Electron. 30 (12) (December 2015) 6751e6762. S. Gubbala, M.R.P. Reddy, Improvement of dynamic stability using conventional, fuzzy controlled and PSO power system stabilizer, Int. J. Ind. Electron. & Electr. Eng. (IJIEEE) 2 (1) (2014) 6e13. P.P. Kumar, M.R. Babu, Saraswathi, Dynamic analysis of single machine infinite bus system using single input and dual input PSS, Int. Electr. Eng. J. (IEEJ) 3 (2) (2012) 632e641. F.M. Serra, A. DoriaCerezo, C.H. De Angelo, L.L. Martı´n Ferna´ndez, M. Bodson, Complex pole placement control for a threephase voltage source converter, in: 2020 IEEE International Conference on Industrial Technology (ICIT), Buenos Aires, Argentina, 2020, pp. 901e906. H. Saadat, Power System Analysis, WCB/McGrawHill, 1999. C.A.B. Karim, M.A. Zamee, Design and analysis of poleplacement controller for dynamic stability improvement of VSCHVDC based power system, in: 2014 9th International Forum on Strategic Technology (IFOST), Cox’s Bazar, 2014, pp. 272e275. C. Olalla, R. Leyva, A. El Aroudi, I. Queinnec, Robust LQR control for PWM converters: an LMI approach, IEEE Trans. Ind. Electron. 56 (7) (July 2009) 2548e2558. B. Kedjar, K. AlHaddad, DSPbased implementation of an LQR with integral action for a threephase threewire shunt active power filter, IEEE Trans. Ind. Electron. 56 (8) (August 2009) 2821e2828.
30 Control of Power Electronic Converters and Systems [29] T. Qoria, et al., Tuning of AC voltagecontrolled VSC based linear quadratic regulation, in: 2019 IEEE Milan PowerTech, Milan, Italy, 2019, pp. 1e6. [30] U. Markovic, Z. Chu, P. Aristidou, G. Hug, LQRbased adaptive virtual synchronous machine for power systems with high inverter penetration, IEEE Trans. Sustain. Energy 10 (3) (July 2019) 1501e1512. [31] M. Mohseni, S.M. Islam, A new vectorbased hysteresis current control scheme for threephase PWM voltagesource inverters, IEEE Trans. Power Electron. 25 (9) (September 2010) 2299e2309. [32] M. Mohseni, S.M. Islam, M.A.S. Masoum, Enhanced hysteresisbased current regulators in vector control of DFIG wind turbines, IEEE Trans. Power Electron. 26 (1) (January 2011) 223e234. [33] T. Ohnishi, Three phase PWM converter/inverter by means of instantaneous active and reactive power control, in: In Proceedings IECON ’91: 1991 International Conference on Industrial Electronics, Control and Instrumentation, 1991, pp. 819e824. [34] T. Noguchi, H. Tomiki, S. Kondo, I. Takahashi, Direct power control of PWM converter without powersource voltage sensors, IEEE Trans. Ind. Appl. 34 (3) (MayJune 1998) 473e479. [35] R. Datta, V.T. Ranganathan, Direct power control of gridconnected wound rotor induction machine without rotor position sensors, IEEE Trans. Power Electron. 16 (3) (May 2001) 390e399. [36] Y. Zhang, Z. Li, Y. Zhang, W. Xie, Z. Piao, C. Hu, Performance improvement of direct power control of PWM rectifier with simple calculation, IEEE Trans. Power Electron. 28 (7) (July 2013) 3428e3437. [37] D. Zhi, L. Xu, B.W. Williams, Improved direct power control of gridconnected DC/AC converters, IEEE Trans. Power Electron. 24 (5) (May 2009) 1280e1292. [38] M. Malinowski, M.P. Kazmierkowski, S. Hansen, F. Blaabjerg, G.D. Marques, Virtualfluxbased direct power control of threephase PWM rectifiers, IEEE Trans. Ind. Appl. 37 (4) (JulyAugust 2001) 1019e1027. [39] D. Choi, K. Lee, Dynamic performance improvement of AC/DC converter using model predictive direct power control with finite control set, IEEE Trans. Ind. Electron. 62 (2) (February 2015) 757e767. [40] P. CortE´s, J. RodrI´guez, P. Antoniewicz, M. Kazmierkowski, Direct power control of an AFE using predictive control, IEEE Trans. Power Electron. 23 (5) (September 2008) 2516e2523. [41] R. Vargas, U. Ammann, J. RodrI´guez, Predictive approach to increase efficiency and reduce switching losses on matrix converters, IEEE Trans. Power Electron. 24 (4) (April 2009) 894e902. [42] W. Xie, X.C. Wang, F.X. Wang, et al., Finitecontrolset model predictive torque control with a deadbeat solution for PMSM drives, IEEE Trans. Ind. Electron. 62 (9) (September 2015) 5402e5410. [43] R. Vargas, P. Cortes, U. Ammann, J. Rodriguez, J. Pontt, Predictive control of a threephase neutralpointclamped inverter, IEEE Trans. Ind. Electron. 54 (5) (October 2007) 2697e2705. [44] Y. Zhang, W. Xie, Z. Li, Y. Zhang, Model predictive direct power control of a PWM rectifier with duty cycle optimization, IEEE Trans. Power Electron. 28 (11) (November 2013) 5343e5351.
Advanced control of power electronic systems Chapter  1 [45]
[46]
[47]
[48] [49] [50]
[51] [52]
[53]
[54]
[55]
[56] [57]
[58]
[59]
[60]
[61]
31
Y. Zhang, W. Xie, Z. Li, Y. Zhang, Lowcomplexity model predictive power control: doublevectorbased approach, IEEE Trans. Ind. Electron. 61 (11) (November 2014) 5871e5880. H. Fang, Z. Zhang, X. Feng, R. Kennel, Ripplereduced model predictive direct power control for active frontend power converters with extended switching vectors and timeoptimised control, IET Power Electron. 9 (9) (September 2016) 1914e1923. Y. Zhang, Y. Peng, H. Yang, Performance improvement of twovectorsbased model predictive control of PWM rectifier, IEEE Trans. Power Electron. 31 (8) (August 2016) 6016e6030. F. Lin, L. Teng, P. Shieh, Intelligent adaptive backstepping control system for magnetic levitation apparatus, IEEE Trans. Magn. 43 (5) (May 2007) 2009e2018. B. Xian, J. Guo, Y. Zhang, Adaptive backstepping tracking control of a 6DOF unmanned helicopter, IEEE/CAA J. Autom. Sinica 2 (1) (January 2015) 19e24. C. Peng, Y. Bai, X. Gong, Q. Gao, C. Zhao, Y. Tian, Modeling and robust backstepping sliding mode control with adaptive RBFNN for a novel coaxial eightrotor UAV, IEEE/CAA J. Autom. Sinica 2 (1) (January 2015) 56e64. K. Kim, Y. Kim, Robust backstepping control for slew maneuver using nonlinear tracking function, IEEE Trans. Contr. Syst. Technol. 11 (6) (2003) 822e829. S.M.J. Rastegar Fatemi, N.R. Abjadi, J. Soltani, S. Abazari, Speed sensorless control of a sixphase induction motor drive using backstepping control, IET Power Electron. 7 (1) (January 2014) 114e123. M. Salimi, J. Soltani, G.A. Markadeh, N.R. Abjadi, Indirect output voltage regulation of DCDC buck/boost converter operating in continuous and discontinuous conduction modes using adaptive backstepping approach, IET Power Electron. 6 (4) (April 2013) 732e741. G. Wang, R. Wai, Y. Liao, Design of backstepping power control for gridside converter of voltage source converterbased highvoltage dc wind power generation system, IET Renew. Power Gener. 7 (2) (March 2013) 118e133. D. Sun, X. Wang, Y. Fang, Backstepping direct power control without phaselocked loop of AC/DC converter under both balanced and unbalanced grid conditions, IET Power Electron. 9 (8) (2016) 1614e1624. V.I. Utkin, Sliding mode control design principles and applications to electric drives, IEEE Trans. Ind. Electron. 40 (1) (February 1993) 23e36. J. Hu, L. Shang, Y. He, Z.Q. Zhu, Direct active and reactive power regulation of gridconnected DC/AC converters using sliding mode control approach, IEEE Trans. Power Electron. 26 (1) (January 2011) 210e222. A. Hemdani, M. Dagbagi, W.M. Naouar, L. Idkhajine, I. Slama Belkhodja, E. Monmasson, Indirect sliding mode power control for three phase grid connected power converter, IET Power Electron. 8 (6) (2015) 977e985, 6. J. Hu, H. Nian, B. Hu, Y. He, Z.Q. Zhu, Direct active and reactive power regulation of DFIG using slidingmode control approach, IEEE Trans. Energy Convers. 25 (4) (December 2010) 1028e1039. R.P. Burgos, E.P. Wiechmann, J. Holtz, Complex statespace modeling and nonlinear control of active frontend converters, IEEE Trans. Ind. Electron. 52 (2) (April 2005) 363e377. K.H. Kim, M.J. Youn, A nonlinear speed control for a PM synchronous motor using a simple disturbance estimation technique, IEEE Trans. Ind. Electron. 49 (3) (June 2002) 524e535.
32 Control of Power Electronic Converters and Systems [62] H. Chen, Z. Xu, F. Zhang, Nonlinear control for VSC based HVDC system, in: 2006 IEEE Power Engineering Society General Meeting, Montreal, Que., 2006, pp. 1e5, p. 5. [63] H. Abootorabi Zarchi, J. Soltani, G. Arab Markadeh, Adaptive inputoutput feedbacklinearizationbased torque control of synchronous reluctance motor without mechanical sensor, IEEE Trans. Ind. Electron. 57 (1) (January 2010) 375e384. [64] J. Matas, M. Castilla, J.M. Guerrero, L. Garcia de Vicuna, J. Miret, Feedback linearization of directdrive synchronous windturbines via a sliding mode approach, IEEE Trans. Power Electron. 23 (3) (May 2008) 1093e1103. [65] L. Yacoubi, K. AlHaddad, F. Fnaiech, L.A. Dessaint, A DSPbased implementation of a new nonlinear control for a threephase neutral point clamped boost rectifier prototype, IEEE Trans. Ind. Electron. 52 (1) (February 2005) 197e205. [66] D.E. Kim, D.C. Lee, Feedback linearization control of threephase UPS inverter systems, IEEE Trans. Ind. Electron. 57 (3) (March 2010) 963e968. [67] W. GilGonza´lez, O.D. Montoya, A. Garces, Direct power control for VSCHVDC systems: an application of the global tracking passivitybased PI approach, Int. J. Electr. Power Energy Syst. 110 (2019) 588e597. [68] L. Harnefors, L. Zhang, M. Bongiorno, Frequencydomain passivity based current controller design, IET Power Electron. 1 (4) (December 2008) 455e465. [69] Y. Gui, Y.O. Lee, Y. Han, C.C. Chung, Novel passivitybased controller design for backtoback STATCOM with asymmetrically structured converters, in: In IEEE Power and Energy Society General Meeting, 2012. [70] Y. Gui, B. Wei, M. Li, J.M. Guerrero, J.C. Vasquez, Passivitybased coordinated control for islanded AC microgrid, Appl. Energy 229 (2018) 551e561. [71] L. Harnefors, A. Yepes, A. Vidal, J. DovalGandoy, Passivitybased controller design of gridconnected VSCs for prevention of electrical resonance instability, IEEE Trans. Ind. Electron. 62 (2) (2014) 702e710. [72] Y. Gui, W. Kim, C.C. Chung, Passivitybased control with nonlinear damping for type 2 STATCOM systems, IEEE Trans. Power Syst. 31 (4) (2016) 2824e2833. [73] Z. Yao, S. Lahaie, V. Rajagopalan, Robust compensator of harmonics and reactive power, In Proc. IEEE PESC (1995) 215e220. [74] A. Vidal, et al., A method for identification of the equivalent inductance and resistance in the plant model of currentcontrolled gridtied converters, IEEE Trans. Power Electron. 30 (12) (December 2015) 7245e7261. ´ . Lo´pez, J. DovalGandoy, A technique to [75] A. Vidal, A.G. Yepes, F.D. Freijedo, J. Malvar, O estimate the equivalent loss resistance of gridtied converters for current control analysis and design, IEEE Trans. Power Electron. 30 (3) (March 2015) 1747e1761. [76] T.D. Do, V.Q. Leu, Y. Choi, H.H. Choi, J. Jung, An adaptive voltage control strategy of threephase inverter for standalone distributed generation systems, IEEE Trans. Ind. Electron. 60 (12) (December 2013) 5660e5672. [77] I. Benlaloui, S. Drid, L. ChrifiAlaoui, M. Ouriagli, Implementation of a new MRAS speed sensorless vector control of induction machine, IEEE Trans. Energy Convers. 30 (2) (June 2015) 588e595. [78] P. Chevrel, F. Auger, M. Machmoum, “HNcontrol for a single phase active power filter: a systematic approach, In Proc. IEEE PESC (1996) 1112e1118. [79] J. Doyle, K. Glover, P.P. Khargonekar, B.A. Francis, “State space solution to standard H2 and HN control problem, IEEE Trans. Autom. Control 34 (August 1989) 831e847. [80] P. Gahinet, P. Apkarian, “Linear matrix inequality approach to HN control, Int. J. Robust Nonlinear Control 4 (1994) 421e448.
Advanced control of power electronic systems Chapter  1 [81] [82] [83] [84] [85]
[86] [87]
[88]
[89]
[90]
[91]
[92] [93]
[94]
[95]
33
D. McFarlane, K. Glover, “A loop shaping desing procedure using HN synthesis, IEEE Trans. Automat. Contr. 37 (June 1992) 759e769. T.S. Lee, S.J. Chiang, J.M. Chang, HN loopshaping controller designs for the singlephase UPS inverters, IEEE Trans. Power Electron. 16 (4) (July 2001) 473e481. C.M. Bishop, Pattern Recognition and Machine Learning, Springer, 2006. I. Goodfellow, Y. Bengio, A. Courville, Deep Learning, MIT press, 2016. T. Wu, Z. Wang, B. Ozpineci, M. Chinthavali, S. Campbell, Automated heatsink optimization for aircooled power semiconductor modules, IEEE Trans. Power Electron. 34 (6) (June 2019) 5027e5031. X. Zhan, W. Wang, H. Chung, A neuralnetworkbased color control method for multicolor LED systems, IEEE Trans. Power Electron. 34 (8) (August 2019) 7900e7913. C. Wei, Z. Zhang, W. Qiao, L. Qu, Reinforcementlearningbased intelligent maximum power point tracking control for wind energy conversion systems, IEEE Trans. Ind. Electron. 62 (10) (October 2015) 6360e6370. C. Wei, Z. Zhang, W. Qiao, L.Y. Qu, An adaptive network based reinforcement learning method for MPPT control of PMSG wind energy conversion systems, IEEE Trans. Power Electron. 31 (11) (November 2016) 7837e7848. I. Bandyopadhyay, P. Purkait, C. Koley, Performance of a classifier based on timedomain features for incipient fault detection in inverter drives, IEEE Trans. Ind. Inf. 15 (1) (January 2019) 3e14. A.E. Mejdoubi, H. Chaoui, J. Sabor, H. Gualous, Remaining useful life prognosis of supercapacitors under temperature and voltage aging conditions, IEEE Trans. Ind. Electron. 65 (5) (May 2018) 4357e4367. S. Li, M. Fairbank, C. Johnson, D.C. Wunsch, E. Alonso, J.L. Proao, Artificial neural networks for control of a gridconnected rectifier/inverter under disturbance, dynamic and power converter switching conditions, IEEE Trans. Neural Netw. & Learn. Syst. 25 (4) (April 2014) 738e750. M.A. Dzieniakowski, M.P. Kazmierkowski, Selftuned fuzzy PI current controller for PWMVSI, in: In Proc. EPE Conf., Seville, Spain, 1995, pp. 1.308e1.313. M.G. Simoes, B.K. Bose, R.J. Spiegel, Design and performance evaluation of a fuzzylogicbased variablespeed wind generation system, IEEE Trans. Ind. Appl. 33 (4) (JulyAugust 1997) 956e965. M. Glavic, R. Fonteneau, D. Ernst, Reinforcement learning for electric power system decision and control: past considerations and perspectives, IFAC Papersonline 50 (1) (July 2017) 6918e6927. P. Kofinas, S. Doltsinis, A.I. Dounis, G.A. Vouros, A reinforcement learning approach for MPPT control method of photovoltaic sources, Renew. Energy 108 (August 2017) 461e473.
This page intentionally left blank
Chapter 2
Robust design and passivity control methods Yonghao Gui1, Jianzhe Liu2, Jan Dimon Bendtsen1, Jakob Stoustrup1 1 Department of Energy Technology, Aalborg University, Aalborg, Denmark; 2Argonne National Laboratory, Lemont, IL, United States
2.1 Introduction Power electronic systems are used in the wide range of applications like flexible AC transmission systems, renewable energy sources (RESs) (e.g., solar and wind), and smart grids [1e4]. Numerous control strategies have been researched for those power electronic systems to improve their stability, performance, robustness, etc. Linear control methods (e.g., proportional and integral (PI), linear quadratic regulator (LQR), linear quadratic Gaussian (LQG), etc.) have been designed to control power electronic systems. Normally, for threephase voltage source converters, a PI control with a decoupling feedforward term is designed separately for controlling deq axes currents in a synchronously rotating reference frame [5]. Though it is easy to do stability analysis when using the PI methods, there are some disadvantages, e.g., the performance will be highly sensitive to the accuracy of controller gains and parameters, the completeness of the current decoupling, the grid voltage conditions, etc. Linear optimal controller based on LQR/LQG control, with better robustness properties than a PI controller, has also been designed for inverters [6,7], and wind turbine (WT) systems [8]. However, due to the fact that these control strategies are designed via a model linearized around the nominal operating point, and inherently nonlinear property of power electronic systems, it is difficult to guarantee a uniform performance in the range of the whole operating points by using the aforementioned techniques. To handle these problems, various nonlinear control strategies have been designed in the literature. One of the efficient control strategies for nonlinear systems to handle trajectory tracking problems is inputeoutput feedback linearization method, since a linearized closedloop system is obtained [9]. The inputeoutput linearization (IOL) method has been applied to rectifiers [10,11], WT systems [12,13], STATCOMs [14,15], modular multilevel Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.00010X Copyright © 2021 Elsevier Ltd. All rights reserved.
35
36 Control of Power Electronic Converters and Systems
converters [16], etc. With the inputeoutput feedback linearization method, it is easy to obtain a satisfactory tracking performance of the output and to guarantee the stability of the output. However, the performance and stability of the internal dynamics are not easy to be guaranteed. In addition, the inputeoutput feedback linearization method is sensitive to the model mismatches, parameter uncertainties, disturbances, etc. To improve the robustness properties, sliding mode control was introduced in Ref. [17]. The sliding mode control method guarantees the trajectory converges to a specified manifold in finite time, and once it reaches the target manifold, it then remains on it even in the presence of parametric uncertainties. Due to such advantages, the sliding mode control method has been applied to various applications such as rectifiers [18e20], WT systems [21,22], STATCOM [23,24], modular multilevel converter [25,26], etc. The sliding mode control method has advantages that it obtains a fasttransient response and has robustness property against parameter uncertainties. However, one disadvantage of the method is the issue of chattering due to the requirement of variable switching frequency. The aforementioned control methods are designed without the consideration of the system’s physical structure such as its passivity feature, which describes energy conservation, acting a central role in control techniques [27]. Employing passivitybased control (PBC), it is able to improve the robustness and simplify the implementation compared with the aforementioned control methods because it prevents the system canceling the nonlinearities that are helpful for its stability [28]. Due to these advantages, PBC is employed into WT system [29], STATCOM [30], modular multilevel converter [31], etc. In this chapter, we introduce how to design robust controllers as well as PBC methods for power electronic systems. At first, a state space model of the power electronic systems, e.g., gridconnected voltage source inverter and STATCOM, is given in Section 2.2. Then, a robust controller is designed for a gridconnected voltage source inverter in Section 2.3. In Section 2.4, a robust smallsignal stabilization control design for DC microgrids is introduced. In Section 2.5, a PBC design for an islanded microgrid and multipulse STATCOM is shown. Finally, the conclusions and future perspectives are summarized in Section 2.6.
2.2 State space model of power electronic systems In this section, a state space model of gridconnected voltage source inverter is described in order to use modern control theory rather than traditional control theory such as a transfer function. Moreover, an active and reactive powere based model is introduced where direct power control (DPC) method is applied.
Robust design and passivity control methods Chapter  2
37
2.2.1 Introduction The calculations of a threephase AC power system are substantially simplified by introducing “Phasor” concept, which is an arrow in the complex plane that has a onetoone relation with a sinusoidal signal as shown in Fig. 2.1. For example, the voltage has the angular frequency, u, then a radius with a length pﬃﬃﬃ 2 times the length of the phasor rotates counterclockwise in the complex plane with the same frequency [32]. The phasor of a signal represents a sinusoidal signal as follows: the length of the voltage phasor equals the effective or the RMS value of the signal. Moreover, the angle of the signal equals to the phase shift of the signal with respect to a voltage reference. For the sake of the simplicity of the analysis for threephase circuits, the Clark transformation is introduced, which transforms a threephase model to one in the stationary reference frame (aeb frame), as shown in Fig. 2.2. The Clark transformation is given as follows: 3 2 1 1 7 6 1 2 2 7 26 7 Tabc2ab ¼ 6 (2.1) 6 p ﬃﬃ ﬃ p ﬃﬃﬃ 7: 34 3 35 0 2 2 Then, in order to obtain a model in the synchronously rotating reference frame, Park transformation is given as follows: # " cosðqÞ sinðqÞ Tab2dq ¼ (2.2) sinðqÞ cosðqÞ where q is a phase angle of the grid voltage, which is obtained via a phaselocked loop (PLL) system in a gridconnected power converter. It should be noted that, the daxis is always coincident with the instantaneous voltage, whereas the qaxis is in quadrature with it, i.e., vgd ¼ Vg and vgq ¼ 0, as it is shown in Fig. 2.3.
FIGURE 2.1 Relationship between a phasor and a sinusoidal of signal.
38 Control of Power Electronic Converters and Systems
FIGURE 2.2 Transformation from abc frame to aeb frame.
FIGURE 2.3 Transformation from aeb frame to deq frame.
2.2.2 Gridconnected voltage source inverter A DPC model of voltage source inverter is briefly introduced where the active and reactive powers are directly controlled without innerloop current controller.
Robust design and passivity control methods Chapter  2
39
FIGURE 2.4 Gridconnected voltage source inverter.
Fig. 2.4 shows a twolevel gridconnected voltage source inverter, which uses an Lfilter to filter out the high frequency of harmonics from the switching operation. At the DC side, RESs (e.g., wind and PV) or energy storage systems (ESSs) can be connected through a capacitor C. For the AC side, the dynamics regarding the grid voltages, the output voltages, and the output currents is able to be formulated as follows: ua ub uc
dia þ vga ; dt dib þ vgb ; ¼ Rib þ L dt dic þ vgc ; ¼ Ric þ L dt ¼ Ria þ L
(2.3)
where vga;b;c , ia;b;c , and ua;b;c indicate the three phases of grid voltages, the output currents, and the output voltages at the inverter, respectively. R and L indicate the filter resistance and inductance, respectively. It should be noted that we assume that Ra ¼ Rb ¼ Rc ¼ R and La ¼ Lb ¼ Lc ¼ L. In this study, a balanced grid voltage condition will be considered. Consequently, the dynamics in Eq. (2.3) is transformed in the stationary reference frame through the Clark transformation introduced in Eq. (2.1) such as ua ub
dia þ vga ; dt dib þ vgb ; ¼ Rib þ L dt
¼ Ria þ L
(2.4)
where ua and ub are the inverter output voltages, ia and ib are the output currents, and vga and vgb are the grid voltages in the aeb frame, respectively. The instantaneous active and reactive powers of the inverter in the aeb frame is given as
40 Control of Power Electronic Converters and Systems
P¼
3 ; 2ðvga ia þ vgb ib Þ
Q¼
3 ; 2ðvgb ia vga ib Þ
(2.5)
where P and Q indicate the instantaneous active and reactive powers of the inverter, respectively. Variations in P and Q are able to be expressed via the grid voltages and output currents variations by differentiating Eq. (2.5) as follows: dP 3 dvga dia dvgb dib ¼ þ vga þ ib þ vgb ia ; dt 2 dt dt dt dt (2.6) dQ 3 dvgb dia dvga dib ¼ þ vgb ib vga ia : dt 2 dt dt dt dt Since only a balanced nondistorted grid voltage is considered, the following relationship can be formulated: vga ¼ Vg cosðutÞ; vgb ¼ Vg sinðutÞ;
(2.7)
where Vg indicates the amplitude of the grid voltage. Furthermore, the grid voltage variations can be formulated by differentiating Eq. (2.7). dvga ¼ uVg sinðutÞ ¼ uvgb ; dt dvgb ¼ uVg cosðutÞ ¼ uvga . dt
(2.8)
Substituting Eqs. (2.4) and (2.8) into Eq. (2.6), a state space model of P and Q can be expressed such as dP R 3 ; ¼ P uQ þ dt L 2Lðvga ua þ vgb ub V 2 g
(2.9)
dQ R 3 ¼ uP Q þ : dt L 2Lðvgb ua vga ub Þ Notice that the dynamics in Eq. (2.9) is a nonlinear multiinputmultiqﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ output (MIMO) system. Notice also that Vg ¼ v2a;g þ v2b;g .
2.2.3 An example of a VSC (STATCOM) In Fig. 2.5, a STATCOM system injects/absorbs the reactive power to the grid to enhance voltage stability and power quality in the transmission line. The inductance, L, denotes the leakage of the power transformer. The resistance, Rs ;denotes the converter and transformer conduction losses, whereas the
Robust design and passivity control methods Chapter  2
41
FIGURE 2.5 Equivalent circuit of an STATCOM.
resistance, Rp , at the DC link denotes the switching losses of the STATCOM where a capacitor, C; connects in parallel. Vða;b;cÞ denote the phasetoneutral bus voltages. eða;b;cÞ denote the voltage of the STATCOM at the AC side [33]. The STATCOM dynamics is formulated as 2 03 2 3 dIa R0s ub 0 0 6 dt 7 6 72 0 3 2 L0 3 6 7 I 7 6 e0a Va0 6 07 6 7 a 0 6 dI 7 6 7 R ub 6 07 6 0 0 7 6 b7¼6 0 (2.10) s0 0 7 6 dt 7 6 74 Ib 5 þ 4 eb Vb 5 L 6 7 0 7 6 0 0 ec Vc 6 7 I 7 6 c 4 05 4 R0s ub 5 dIc 0 0 0 dt L where Iða;b;cÞ represent threephase currents. ub is the angular speed at the nominal frequency of the grid. The state variables and parameters with the apostrophe represent the values in perunit based on the following definitions: zbase ¼
iða;b;cÞ 0 Vða;b;cÞ 0 eða;b;cÞ Vbase 0 : Iða;b;cÞ ¼ ; Vða;b;cÞ ¼ ; eða;b;cÞ ¼ ; Ibase Ibase Vbase Vbase
L0
ub L 0 1 Rs Rp ; C ¼ ; R0 ¼ ; R0 ¼ zbase ub Czbase s zbase p zbase
¼
By applying the transformation Eqs. (2.1) and (2.2) into Eq. (2.10), a model of STATCOM can be formulated on the deq frame as 2 3 R0s ub 0 kub 0 ub 0 0 I þ uI þ V cosa jV j q 7 2 0 3 6 L0 d L0 dc L0 6 7 6 7 I_d 0 7 6 _0 7 6 Rs ub 0 kub 0 6 7 0 6 I 7¼6 uId 0 Iq þ 0 Vdc sina (2.11) 7; 4 q 5 6 L L 7 6 7 0 6 7 V_dc 3 0 ub C 0 0 5 4 3 0 0 0 kC ub Id cosa kC ub Iq sina 0 Vdc 2 2 Rp 0 indicate the active current, reactive current, and DClink where Id0 , Iq0 , and Vdc voltage, respectively. Since the synchronously rotating reference frame is
42 Control of Power Electronic Converters and Systems
FIGURE 2.6 Operation modes of STATCOM.
oriented along the grid voltage, Vq0 is zero. u is the angular velocity of the fundamental frequency of the grid voltage. a is the phaseshift angle by which the converter voltage vectors ea;b;c lead the line voltage vectors Va;b;c . It should be noted that a is only one control input. k is a constant factor relating the DClink voltage to the maximum amplitude of the voltage at the STATCOM AC side. In this study, when the STATCOM voltage is larger than the grid voltage, the STATCOM works in the capacitive mode, i.e., Iq0 is negative and the STATCOM generates the reactive power to the grid as shown in Fig. 2.6A. When the STATCOM voltage is smaller than the grid voltage, the STATCOM operates in the inductive mode, i.e., Iq0 is positive and it absorbs the reactive power from the grid as shown in Fig. 2.6B.
2.3 Robust controller design In this section, robust but simple control law, called a voltagemodulated direct power control (VMDPC) with additional damping scheme is introduced for gridconnected inverter, as shown in Fig. 2.7. The main objective of the VMDPC method is to obtain not only a fasttransient response but also a good steadystate performance, where the power ripples as well as total harmonics distortion (THD) of the output currents are decreased. The VMDPC with additional damping method has four components. The first one is a nonlinear controller, which defines the VM inputs that transform the inverter system into a linear timeinvariant (LTI) one. Then, a feedback controller is designed to control active and reactive powers independently. The VMDPC obtains two separate secondorder error dynamics of active and reactive powers, which guarantee the exponentially stable of the closedloop system in the whole operating range. The third one is the additional damping term, which is designed to enhance the robustness properties against uncertainties in terms of the model and parameter as well as guarantee the exponential stability. Finally, the last one is the inverse of VM, which generates the original control input of the inverter.
Robust design and passivity control methods Chapter  2
43
FIGURE 2.7 Control block diagram of the voltagemodulated (VM) direct power control with addition damping.
2.3.1 Voltagemodulated direct power control It should be noted that the dynamics Eq. (2.9) is a timevarying MIMO system. In addition, the control inputs are coupled with the states P and Q. Let us define the VM inputs as [34]. uP : ¼ vga ua þ vgb ub ; uQ : ¼ vgb ua þ vga ub :
(2.12)
Through the relationship of the grid voltage in Eq. (2.7), the new VM inputs are the new control inputs expressed in the deq frame such as #" # " # " " # ud uP ua cosðutÞ sinðutÞ ¼ Vg ¼ Vg ; (2.13) uq uQ sinðutÞ cosðutÞ ub where ud and uq are the inverter voltages on the deq frame. Vg is the magnitude of the grid voltage. From Eq. (2.13), it can be seen that the VM inputs are the inputs in the deq frame. The system in Eq. (2.9) can be rewritten as
44 Control of Power Electronic Converters and Systems
dP R 3 ; ¼ P uQ þ dt L 2LðuP V 2 g
(2.14)
dQ R 3 ¼ uP Q uQ ; dt L 2L where u is angular frequency of the grid voltage. Notice that the system in Eq. (2.14) has been converted into a simple linear MIMO system.
2.3.2 Tracking controller This subsection introduces a traditional tracking control strategy, which is designed to make P and Q to track their references exponentially. At first, let us define the following errors of P and Q: eP : ¼ Pref P; eQ : ¼ Qref Q;
(2.15)
where Pref and Qref indicate the reference of P and Q, respectively. For the system in Eq. (2.9), we design a tracking control law consisting of a feedforward and feedback such as 2R 2Lu 2L 2Lu 2R 2L Pþ Q þ n P ; uQ ¼ P Q nQ ; uP ¼ Vg2 þ 3 3 3 3 3 ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} ﬄ{zﬄ} ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} ﬄ3ﬄ{zﬄﬄ} feedforward
feedback
feedforward
(2.16)
feedback
where nP and nQ are the new feedback controller that can be designed as follows: nP ¼
P_ref þ KP;P eP þ KI;P
Zt eP ðsÞds; 0
nQ ¼ Q_ref þ KP;Q eQ þ KI;Q
(2.17)
Zt eQ ðsÞds; 0
where KP;P , KI;P , KP;Q , and KI;Q are the controller gains. Substituting Eqs. (2.16) and (2.17) into Eq. (2.14), the following error dynamics can be obtained: P_ref P_
Zt ¼ e_P ¼ KP;P eP KI;P
eP ðsÞds; 0
Q_ref Q_
(2.18)
Zt ¼ e_Q ¼ KP;Q eQ KI;Q
eQ ðsÞds: 0
Robust design and passivity control methods Chapter  2
45
If we define j_ P ¼ eP and j_ Q ¼ eQ , the error dynamics Eq. (2.18) are transformed into the following form: 3 3 32 2 2 eP KP;P KI;P 0 0 e_P 6 7 6 j_ 7 6 1 0 0 0 7 7 6 jP 7 6 P7 6 (2.19) 7 76 6 e_ 7 x_ ¼ 6 4 Q5 4 0 0 KP;Q KI;Q 5 4 eQ 5 j_ Q jQ 0 0 1 0 ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} ﬄﬄﬄ{zﬄﬄﬄ} ﬄﬄﬄ{zﬄﬄﬄ} A
x
From Eq. (2.19), it can be found that if KP;P , KI;P , KP;Q , and KI;Q are larger than zero, then A is Hurwitz matrix, i.e., it has all negative eigenvalues, which guarantee the exponential stability. It should be noted that it is possible to design KP;p , KP;i , KQ;p , and KQ;i with the consideration of the damping ratio and settling time of the closedloop system [11]. Consequently, the original control inputs can be calculated through inverse of Eq. (2.12), which is formulated as follows: ua ¼
vga uP vgb uQ vgb uP þ vga uQ ; ub ¼ Vg2 Vg2
(2.20)
By using the VMDPC method, the system in Eq. (2.9) is converted into an LTI MIMO system. Fig. 2.8 shows a comparison between the inputs of the conventional DPC and the VMDPC when P ¼ 1 kW and Q ¼ 1 kVAr. It can be seen that the VMDPC control inputs, uP and uQ , are constant, i.e., the VMDPC has the same construction to the conventional vector current control designed on the deq frame, which has been proven in Ref. [3]. That means various linear control methods can be designed by using the VMDPC concept to overcome the corresponding issues [35e40].
FIGURE 2.8 Conventional direct power control (DPC) and voltagemodulated direct power control (VMDPC) of inputs.
46 Control of Power Electronic Converters and Systems
2.3.3 Robust control design In a practical system, the stability obtained by Eq. (2.19) will be affected by uncertainties including measurement noises, parameter uncertainties, discretization errors, etc. Consequently, it is assumed that there exist some uncertainties as dP , dQ , and D in the LTI MIMO system Eq. (2.14), which are formulated as follows: x_ ¼ fa ðx; uP :uQ ; dP ; dQ Þ 3 2 R 3 þ dP 7 6 L x1 ux2 þ 7 6 2LðuP Vg2 7 6 7; 6 ¼ 6 7 7 6 5 4 R 3 ux1 x2 þ uQ þ dQ L 2L
(2.21)
It should be noted that the uncertainties, dP and dQ , are bounded such as 0 jdP j D; 0 jdQ j D:
(2.22)
In the gridconnected voltage source inverter, the parameter uncertainties can, e.g., be the frequency, grid voltage, the inductance of filter, or change of gains in measurement. For a practical system with limited operating range, it should estimate an upper bound of uncertainties, D. From Eq. (2.19), it is always guaranteed the exponential stability of the nominal plant with the control law Eq. (2.16) in the operating range. The integral action in Eq. (2.17) is designed to compensate for any DC offset due to model or parameter uncertainties. Moreover, a small perturbation from the operation point is ultimately bounded through the exponential stability property [41]. Consequently, assumptions Eqs. (2.21) and (2.22) are reasonable without loss of generality. When dP ¼ 0 and dQ ¼ 0, it has been proved that the closed loop is exponentially stable. When dP s0 or dQ s0, only dP and dQ terms are considered for the sake of simplicity. Let us define a Lyapunov function candidate as 1 1 V ¼ e2P þ e2Q : 2 2
(2.23)
Then, the derivative of the Lyapunov function candidate Eq. (2.23) in terms of time is calculated such as V_ ¼ eP e_P þ eQ e_Q ¼ eP dP Kd;P sgnðeP Þ þ eQ dQ Kd;Q sgnðeQ Þ : (2.24)
Robust design and passivity control methods Chapter  2
47
If the new feedback controllers are taken as nP;n ¼
P_ref þ KP;P eP þ KI;P
Zt eP ðsÞds þ Kd;P sgnðeP Þ; 0
nQ;n ¼
Q_ref þ KP;Q eQ þ KI;Q
(2.25)
Zt eQ ðsÞds þ Kd;Q sgnðeQ Þ; 0
where Kd;P and Kd;Q are the new controller gains, then the derivative of the Lyapunov function candidate in terms of time is changed into the following one: V_ ¼ eP dP Kd;P sgnðeP Þ þ eQ dQ Kd;Q sgnðeQ Þ :
(2.26)
If Kd;P > D and Kd;Q > D are taken, then V_ KDP jeP j KDQ jeQ j;
(2.27)
where KDP ¼ Kd;P D and KDQ ¼ Kd;Q D. Notice that the VMDPC with additional damping term scheme presents the dynamics of the inverter on the deq frame without a PLL [3].
2.3.4 Simulation results MATLAB/Simulink and PLECS are used to validate the control method introduced in this study. The gridconnected voltage source inverter is built in the PLECS, and the control strategy is built in MATLAB/Simulink. The performance of the VMDPC is compared with the PBCDPC proposed in Ref. [42], which obtains a better performance compared to the SMCDPC designed in Ref. [43]. The parameters of the system are listed in Table 2.1. Figs. 2.9 and Fig. 2.10 show the performance of the VMDPC and PBCDPC, respectively, when P changes from 0 W to 1 kW at 0.03 s and is back
TABLE 2.1 Parameters of system in simulation. DClink voltage
250 V
AC frequency
50 Hz
Linetoline voltage (RMS)
133 V
Switching frequency
10 kHz
Ra;b;c
0.16 U
Sampling frequency
10 kHz
La;b;c
4 mH
Power rate
2 kVA
KP ;P &KP ;Q
200
KI;P &KI;Q
2000
Kd;‘p &Kd;Q
10
48 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 2.9 Simulation results of the PBCDPC when Q is 0 VAr and P changes from 0 W to 1 kW. (A) P and Q, (B) output currents. DPC, direct power control; PBC, passivitybased control.
to 0 W at 0.07 s. Among this period, Q is regulated to 0 VAr. Comparing to P tracking performance of the PBCDPC, the VMDPC significantly reduces ripples both in P and Q as well as in the output current. The VMDPC has no need to consider the switching delay as well as harmonics. Furthermore, the VMDPC obtains two independent decoupled error dynamics of P and Q that guarantee the exponential stability in the operating range. Those features are similar with the vector current controller designed in the deq frame. Figs. 2.11 and Fig. 2.12 show the case when Q changes from 0 VAr to 1 kVAr at 0.03 s and then returns to 0 VAr at 0.07 s. P is regulated to 0 W. As it is similar with the case obtained from P tracking performance, the VMDPC has smaller ripples both in P and Q as well as in output current compared to the PBCDPC. The steadystate performance (i.e., power ripples) is the main disadvantage of DPC methods compared with the vector current controller designed on the deq frame. However, the VMDPC method overcomes the steadystate performance problem since two decoupled LTI error dynamics are obtained and
Robust design and passivity control methods Chapter  2
49
(A)
(B)
FIGURE 2.10 Simulation results of the voltagemodulated direct power control when Q is 0 VAr and P changes from 0 W to 1 kW. (A) P and Q, (B) output currents.
they are guaranteed to be globally exponentially stable by Eqs. (2.19) and (2.27). To compare the steadystate performance of the two methods, the case is tested where P ¼ 1kW and Q ¼ 1 kVAr, as shown in Figs. 2.13 and Fig. 2.14. Fig. 2.15 shows the harmonic spectra of the output currents from 3rd up to 25th harmonics. The VMDPC obtains a THD of the output current as 1.4%, which is lower than 5% required for grid connection. However, the PBCDPC obtains THD ¼ 5.6%. Therefore, it can be concluded that the VMDPC decreases the THD of the output current but maintains the same transient performance as the PBCDPC.
2.4 Robust smallsignal stabilization control design 2.4.1 Introduction From the previous discussions, the linearized system model is dependent on the desired states. Variations in the desired states cause entries of the linearized
50 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 2.11 Simulation results of the PBCDPC when P is 0 W and Q changes from 0 VAr to 1 kVAr. (A) P and Q, (B) output currents. DPC, direct power control; PBC, passivitybased control.
system matrix to change. Such changes can drive the eigenvalues away from the designated positions, which can lead to instability in certain circumstances. Stability analysis of a linearized system requires smallsignal stability analysis methods. However, the conventional smallsignal stability methods could be hard to apply to study a system with strong uncertainty, since the variations in the equilibria are usually hard to precisely characterize for such systems, whereas most existing smallsignal stability analysis results require exact knowledge about the equilibria [44]. To tackle this difficulty, a computationally efficient method is shown in this section to design a robust smallsignal stabilization control. The method does not require knowing the exact system equilibria for a power electronic device
Robust design and passivity control methods Chapter  2
51
(A)
(B)
FIGURE 2.12 Simulation results of the voltagemodulated direct power control when P is 0 W and Q changes from 0 VAr to 1 kVAr. (A) P and Q, (B) output currents.
or a power electronic interfaced system. The equilibria are assumed to be an arbitrary point lying in a polytopic set that represents the operational constraints, such as voltage and current bounds. A stability condition is shown that certifies whether all elements lying in the set are stable. Then, a stabilization control is designed to make sure the condition is always satisfied. The theoretical results are first shown using a general model derived from Lur’e type system. Several examples concerning DC/DC converters and DC microgrids are shown to illustrate the usefulness of the method.
2.4.2 Lur’e type model The technical results are based on a general model. Many power engineering control systems have separate linear and nonlinear parts added together in the system dynamics, thus can be categorized into Lur’e type systems [45].
52 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 2.13 Simulation results of the PBCDPC when P is 1 kW and Q is 1 kVAr. (A) P and Q, (B) output currents. DPC, direct power control; PBC, passivitybased control.
As shown in Fig. 2.16, a Lur’e type system is described by the following model: x_ ¼ Ax þ B1 p þ B2 u; (2.28) y ¼ Cx þ D1 p þ D2 u where p is an additive nonlinear input given by p ¼ fðx; u; wÞ, and w is an unknown variable that represents disturbance or modeling errors, x, u, and yare the system state, input, and output variables. Inserting p ¼ fðx; u; wÞ into Eq. (2.28) yields the classic Lur’e type system model for the state dynamics: x_ ¼ Ax þ B2 u þB1 fðx; u; wÞ . ﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄ} ﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄ} linear part
(2.29)
nonlinear part
It can be seen that the Lur’e type system has two additive parts: the linear part resembles a common LTI system, and the nonlinear part is in a general function form. Throughout the following content in this section, it is assumed
Robust design and passivity control methods Chapter  2
53
(A)
(B)
FIGURE 2.14 Simulation results of the voltagemodulated direct power control when P is 1 kW and Q is 1 kVAr. (A) P and Q, (B) output currents.
FIGURE 2.15 Current harmonic spectra of Figs. 2.13 and Fig. 2.14, where total harmonics distortion of the voltagemodulated direct power control (VMDPC) is 1.4% and PBCDPC is 5.6%. DPC, direct power control; PBC, passivitybased control.
that f is a rational function where both the nominator and denominator are polynomials. It can be seen that AC inverter model Eq. (2.9) follows from this general form. Two examples involving DC/DC converter model and DC microgrid demonstrate the wide applicability of the model.
54 Control of Power Electronic Converters and Systems
FIGURE 2.16 Control diagram for Lur’e type system.
2.4.2.1 Example 1: DC/DC buck converter model A buck converter is widely used in DCbus voltage regulation. Consider a buck converter interfaced voltage source supplying a constant power load P. Let the voltage source be Vi ; it is the input of the buck converter. The dynamics of the output voltage Vo and the current I are given as follows: 8 dI > > > L ¼ Vi D Vo < dt (2.30) > dVo P > > ¼I : C Vo dt where D is the duty ratio of the DC/DC buck converter, L and C are the inductance and capacitance of the converter. Notice that the duty ratio can be determined through controllers like a primary droop controller. It can be clearly seen from the model that there is a nonlinear input P=Vo where the constant power P is often unknown and the reciprocal form makes the term nonlinear in the system state. Since source voltage Vi is usually a constant, the term Vi D can be considered as a control input to the system. Let x ¼ ½I; Vo u . The model can be written in a more compact form as follows:
1
1
0 x_ ¼ (2.31) x þ Vi D þ P=Vo ; 1 0 1 ﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄ} ﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄ} ﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄ} Ax
B2 u
B1 fðx;pÞ
It can be seen clearly that the model follows the general Lur’e type model.
2.4.2.2 Example 2: DC microgrid model In addition, many power electronic interfaced systems can be modeled as Lur’e type system as well. DC microgrid can be used as an example to
Robust design and passivity control methods Chapter  2
55
FIGURE 2.17 Example of a (A) DC microgrid topology and (B) power electronic circuit.
illustrate this. An example DC microgrid is shown in Fig. 2.17. The DC microgrid is composed of multiple DC buses. Each DC bus can be connected to several DC/DC converter interfaced loads and sources. For simplicity, we use one composite source and load to integrate all the sources and loads on each bus. It is well known that many DC sources are controlled under primary droop control. The duty ratio is updated in order to control the terminal voltage following the droop control. For the kbus, the output voltage of the source is given as follows: Vgk ¼ Vgk Igk dk ;
(2.32)
where dk is a droop gain, Vgk is the output voltage of the source, Igk is the is a droop reference. From Eq. (2.32), the desired output current, and Vgk terminal output voltage is an affine function in the output current.
56 Control of Power Electronic Converters and Systems
The time scale for finding the desired duty ratio is often negligible when comparing to that of the primary droop control, hence it is often assumed that the instantaneous terminal voltage output of a source is determined by Eq. (2.32). In addition, the constant power loads are still a nonlinear function in the terminal voltage, and it is assumed that the instantaneous power is always the desired power demand. Bearing this in mind, the circuit model of a DC bus is given as follows: 8 > dIgk > > Lgk ¼ Vgk Igk Rgk Vbk > > dt > > > > > > > > Ck dVbk ¼ Igk Ilk Ikj > > > dt > > < dIlk (2.33) ¼ Vbk Ilk Rlk Vlk Llk > > dt > > > > > > dVlk Plk > > ¼ Ilk Clk > > > dt Vlk > > > > > : Vgk ¼ Vgk Igk dk where Igk and Ilk are the source and load currents, Ijk is the transiting current between bus j and k, Vbk is the DCbus voltage, Vlk is the voltage across the load capacitor, Lgk and Llk are the source and load inductance, Rgk and Rlk are the source and load resistance, Ck is the DCbus capacitance, and Clk is the load capacitance, at last, Rjk is the line resistance between bus j and k. It can be seen that the model has linear part with respect to the source current, DCbus voltage, and load current; meanwhile, it has also a nonlinear part concerning the constant power load. It thus resembles the single DC/DC converter model and further resembles a Lur’e type system. Extending this model to the entire system, one can see that Lur’e type system can be used to describe a DC microgrid system. A salient feature of a Lur’e type system is that the system equilibria depend on the control input, uncertain disturbances, as well as the nonlinear function fðx; u; wÞ. This can be seen from the derivation below. Suppose the equilibrium is xd , so it must satisfy the following steadystate equation: (2.34) 0 ¼ Axd þ B1 f xd ; u; w þ B2 u; which is the steadystate equation, and it verifies that xd is determined by both the linear and the nonlinear parts.
Robust design and passivity control methods Chapter  2
57
If the nonlinear function is not concerned with the system state, i.e., when it is given by fðu; wÞ, and the system matrix is nonsingular, the equilibrium is given as follows: xd ¼ A1 ðB1 fðu; wÞ þ B2 uÞ;
(2.35)
it can be seen that the system equilibrium is dependent on the uncertainty and the control. Nevertheless, when the nonlinear part involves the state variables, it is often difficult to find an analytical expression of the system equilibria. For example, in the DC microgrid system with the reciprocal nonlinear function owing to the constant power loads, the system steadystate equations are multidimensional quadratic equations that are in lack of analytical solutions.
2.4.3 Robust smallsignal stability analysis problem Classic smallsignal stability analysis results cannot be applied when the system equilibria are not known. In the following, it is shown that when xd lies in a known constraint set, robust stability analysis results can be applied to ascertain the smallsignal stability. To simplify the discussion, suppose the system control follows a linear state feedback control as follows: u ¼ Kx;
(2.36)
hence, the closedloop system is now given by x_ ¼ ðA þB2 KÞx þ B1 fðx; wÞ where the dependence of the equilibria on the control variable is removed. This control gain matrix K can be obtained through classic power electronic control design methods and are considered known for now. Let the uncertainty be subject to box constraint arising from capacity limits (physical disturbances are generally bounded) as follows: w ˛ W ; where W ¼ fw : w ˛ ½w ; wþ g. Furthermore, let xd ˛ X d with X d ¼ d d
x : x ˛ ½xd ; xdþ being box constraint sets for the system equilibria. Notice that xd and xdþ are the elementwise lower and upper bounds. This constraint set can represent the operational constraints for a power electronic device or system. For example, it can represent a constraint set for the terminal voltage of an inverter. Any desired system equilibria must lie in the operational constraint set X d . If xd ˛ X d is known, a conventional linearization method can be applied to study the eigenvalues of the system Jacobian matrix. Now that xd could be an arbitrary point in X d , the Jacobian matrix is unknown, and we cannot directly study its eigenvalues. Fortunately, with xd and w lying in a constraint set the Jacobian matrix is bounded as well. Definition 2.1.
lying in X
d
The system is robustly smallsignal stable when any equilibrium is smallsignal stable for any w ˛ W :
58 Control of Power Electronic Converters and Systems
Let the linearized nonlinear function with respect to arbitrary xd ˛ X d e d ; wÞ and the linearized Jacobian matrix is and w ˛ W be fðx e d ; wÞ. J ¼ A þ B2 K þ B1 fðx It is well known that as long as f is a rational function (i.e., the nominator and denominator are polynomials in xd and w), and xd ˛ ½xd ; xdþ , w ˛ þ e h½w :w þ,i a box constraint can always be found for the linearized function f ˛ e ;f e . Both the lower and upper bounds are dependent on xd , xdþ , w f and wþ ; thus are known. As a result, the Jacobian now lies in a known interval matrix constraint, J ˛ ½J ; J þ . So long as any J in this set is smallsignal stable, the system is robustly smallsignal stable. Now the robust stability analysis problem is transformed into determining the stability of a matrix lying in an interval. Several methods exist to solve this problem [44]. A method based on linear matrix inequality testing is shown in the following.
2.4.4 Robust stability analysis method h þi e ;f e˛ f e , each element of this vector lies in a box Knowing that f ei ˛ constraint. lie hin the ifollowing constraint set, f n o h þ i Let the ith element þ e e e e e e e fi ; fi , and define F ¼ f : fi ˛ fi ; fi ; ci .
e e ˛ F: Equivalently, the problem is to determine the stability of Jwith f e Although there are infinitely many elements in F, there are finitely many e is a ndimensional vertices for this set. Let the number of vertices be m. (If f n vector, there are m ¼ 2 vertices). With regard to the kth vertex of F, let it be ek . ek , and let the Jacobian matrix be J k ¼ A þ B2 K þ B1 f f Theorem 2.1. The system is robustly smallsignal stable if there exists a positive semidefinite matrix P that satisfies the following linear matrix inequality constraint: u P _ 0; P J k þ J k P30; k ¼ 1; /; m (2.37)
Notice that the condition derived in Eq. (2.37) is a semidefinite programming problem, that can be solved through offtheshelf algorithms like interior point algorithm [46].
2.4.5 Robust stabilization control design In the robust stability analysis, it is shown how to determine whether the system equilibrium is always smallsignal stable with a given control law u ¼ Kx where the control gain matrix K is known.
Robust design and passivity control methods Chapter  2
59
If an initial design K cannot ensure the robust smallsignal stability, it can be redesigned using the method below. The following robust smallsignal stabilization problem is considered: a control gain matrix K is to be designed such that the system Jacobian J ¼ Aþ e d ; wÞ is robustly smallsignal stable with xd ˛ X d and w ˛ W . B2 K þ B1 fðx The design problem is formulated as follows: min a K;P;a u e k þ A þ B2 K þ B 1 f ek P3 aI; P A þ B2 K þ B1 f
k ¼ 1; /; m a > 0;
P_0:
(2.38)
If problem Eq. (2.38) is feasible, then any solution K ensures that the system is robustly smallsignal stable. As there is no extra constraint on K, an interpolation method can be applied to efficiently solve the problem. Let Y ¼ PK u and substituting it into Eq. (2.38) yields min a P;a;Y
u e k þ A þ B1 f ek P þ YBu þ B2 Y u 3 aI; P A þ B1 f 2 k ¼ 1; /; m
a > 0;
P_0:
(2.39)
Problem Eq. (2.39) is again a linear matrix inequality problem. Often there are constraints on K, for example, with decentralized or distributed control design, there is always structural constraint on K as much system information is not available to the controller. The interpolation method cannot be used in this case. Fortunately, problem Eq. (2.38) is a bilinear matrix inequality constraint problem, and there exist effective methods to solve such a problem as well [47].
2.4.5.1 Example 3: robust smallsignal stability of a DC microgrid Example is based on a ninebus DC microgrid system as shown in Fig. 2.18. Each source uses the VeI droop control and is considered as a DC voltage source [48]. The power electronic circuit of the kth bus is shown in Fig. 2.17. The parameters of the kth bus can be found in Table 2.2. From Eq. (2.33), the nonlinear part is the constant power load Plk = Vlk , and its linearized form is Plk Vlk2 . Suppose that each load power is uncertain and may take values in the range [0, 20 kW] and letthe load voltage be in the range [360 V, 440 V]. Hence, the linearized term Plk Vlk2 is in the range [0, 0.193]. From the results about robust smallsignal stabilization, problem Eq. (2.37) is infeasible when the droop gains are 0.06. This infeasibility indicates that with this design the system is not robustly smallsignal stable as shown in Fig. 2.19. For the simulation we let each load to gradually increase from 5 to 20 kW. During the increase of load, the system becomes unstable.
60 Control of Power Electronic Converters and Systems
FIGURE 2.18 DC microgrid with nine buses for robust smallsignal stability analysis.
TABLE 2.2 Parameters of ninebus DC microgrid. Rgk
0.05 U
Rlk
0.05 U
Lgk
0.9 mH
Llk
0.9 mH
Ck
0.75 mF
Clk
0.75 mF
½Plk ; Plkþ
[0.20 kW]
Rjk
1U
Pkn
15 kW
[360 V,440 V]
Vgk
400 V
dk
0.06
½Vlk ; Vlkþ
Let pnlk be the nominal power, and it is set as 15 kW. The droop gain is redesigned according to Problem Eq. (2.38) and it shows that when the droop gains are set to 0.2, system will be robustly stable. This is verified by Fig. 2.20 as when each load again increases to 20 kW the system remains stable.
Robust design and passivity control methods Chapter  2
61
FIGURE 2.19 Instability with inappropriate control design.
FIGURE 2.20 System is robustly smallsignal stable.
2.5 Passivitybased control design The portcontrolled Hamiltonian (PCH) system is introduced in this section in order to get the passivity properties, which has the advantage that if a group of passive subsystems are in the parallel or feedback connection, the whole system is also stable and passive [49]. This property can be applied to heterogenous RESs in, e.g., a microgrid in order to guarantee the stability of the microgrid, and an STATCOM system in order to guarantee the exponential stability in the whole operating range.
62 Control of Power Electronic Converters and Systems
2.5.1 Introduction Generally, PCH systems can formulate a classification of the variables and the equations into a special model where the interconnection structure related with the exchanges of energy is defined [50]. It is easy to modify the energy function and add dissipation into the system, which are the basic steps of PBC. Moreover, the geometric structure of the state space of PCH systems can profitably be used for PBC. The PCH system has received an increasing amount of interest from the field of control engineering [51,52]. It has been proven effective for a wide range of applications [50]. Many power electronic applications can be modeled as PCH systems where multiple simpler components are interconnected to form a more complicated and complete system.
2.5.2 Portcontrolled Hamiltonian system In this study, at first, a general state and input sets are introduced. Let x ˛ ℝn and u ˛ ℝm indicate the state and input vector, respectively. A system is formulated in the following form: x_ ¼ f ðx; uÞ; x ˛ X3ℝn
(2.40)
u ˛ U3ℝm
where and are the state and the input vector, respectively. f ð $; $Þ is sufficiently smooth in the open connected sets X [53]. Definition 2.2. [54]
If a system in Eq. (2.40) satisfies the following form Eq. (2.41), it can be called as a PCH system. x_ ¼ ðJ > > dt > > > : if ¼ ic þ io
79
(3.1)
where v inv, v o, are output voltage of the inverter and capacitor voltage or voltage at the point of common coupling (PCC), i f, i o, and i c are the inverter or inductor current, capacitor current, and load current. Besides, L and R are filter inductance, and its equivalent resistance and C is the filter capacitance. In practice, the value of parameters L, R, and C is not exactly known, and they could vary from the nominal values. Therefore, the impact of parameter variations must be investigated on the control system performance. By considering v o and ic as system states and io as input disturbance, Eq. (3.1) can be rewritten in standard state space form: dx ¼ Ax þ Bu þ d dt " x¼
x1 x2
#
" ¼
ic vo
2
#
" ;d ¼
d1 d2
#
Rn 6 L n 6 ; u ¼ vinv ; A ¼ 6 6 4 1 Cn
3 2 3 1 1 7 Ln 7 6L 7 7; B ¼ 6 n 7 7 4 5 5 0 0
(3.2) where x and u are system states and control input, d is disturbance inputs, which include all system uncertainties and disturbances, A and B are the state and input matrixes, and parameters with subscript n show the nominal values of the system parameters.
3.3 Introduction to sliding mode control The classic SMC has been applied in power electronic and electric drive applications in Refs. [25e27] for the first time. After that, it has been quickly extended to various implementations of these systems as discussed in the introduction. The primary aim of this section is to review the basic idea of classic SMC and the reachability conditions. After that, the structure of control input in SMC is introduced and calculated. Interested readers can refer to some primary and basic works in this topic for more details in Ref. [25e30]. It is worth to note that the application of classic SMC to threephase UPS will be presented in the next section.
80 Control of Power Electronic Converters and Systems
The dynamic equation of a nonlinear singleinput system can be represented as xn ¼ f ðxÞ þ bðxÞu
(3.3)
where scalar variables x and u are system output and control input. Therefore, the state vector would be T x ¼ x x_ . xðn1Þ (3.4) In Eq. (3.3), f(x) and b(x) are uncertain nonlinear functions, but upper bounded by known, continuous function of x (i.e., jf ðxÞj < F1 ðxÞ; jbðxÞj < B1 ðxÞ, where F1 and B1 are known continuous functions). The sign of b(x) is also known. It is worth to note that Eq. (3.3) is a general form that also includes the studied linear system in Eq. (3.2) by simply assuming f ðxÞ ¼ Ax and bðxÞ ¼ B. The control objective is the convergence of the state vector to the desired or reference state vector in the presence of the system uncertainties and disturbances. Therefore, the tracking error can be defined as h iT ex ¼ x xd ¼ ex e_ x . eðn1Þ (3.5) x T is the reference state vector. where xd ¼ xd x_d . xðn1Þ d Now, let us define the timevariant surface s(t) with the scalar equation sðex ; tÞ ¼ 0 as [30] n1 d þl sðex ; tÞ ¼ ex (3.6) dt here l is a positive constant that determines the convergence rate of the dynamic error to zero. And sðex ; tÞ ¼ 0 shows a differential equation, which has only one solution ex h0. Therefore, if we can bring and remain the system states on the sliding surface ðsðex ; tÞ ¼ 0Þ, the dynamic error Eq. (3.5) converges to zero by the specified dynamics in Eq. (3.6), which is independent of either the plant parameters or the external disturbances. This socalled “invariance” feature looks interesting for designing a controller for the dynamic system operating under plant uncertainties. Another exciting feature is changing the referencetracking problem Eq. (3.5) by keeping the scalar s to zero for all t > 0. In other words, the tracking problem of n dimension vector (xd) is reduced to the stabilizing problem of one scalar variable (s). The sliding surface is attractive,1 and system states move and remain on this surface ðsðex ; tÞ ¼ 0Þ, if the control input fulfills the reachability condition. So, our goals would be 1. Attractive sliding surface means that trajectories outside the surface converge to it.
Sliding mode controllers in power electronic systems Chapter  3
81
1. Defining the reachability condition, which guarantees the moving of the system states to the sliding surface, anywhere in the state plane and under all system uncertainties and disturbances. 2. Calculating the control input to fulfill the reachability condition and maintaining the system state on the sliding surface for all further times.
3.3.1 Reachability condition in the sliding mode control The system state goes and remains on the sliding surface (s(t)) from anywhere in the state plane when the reaching condition is fulfilled. There are different forms of reaching conditions, and generally, they can be classified into three categories of switching functions, Lyapunov function, and dynamic method. The earliest and oldest reaching condition is s s_ < 0
(3.7)
It is worth noting that s can be selected simply as Eq. (3.6) or any desirable differentiable linear or nonlinear dynamic error. The condition in Eq. (3.7) is universal. However, it does not guarantee a finite and limited reaching time. Another choice that belongs to this category is 1 d 2 s hjsj 2 dt
(3.8)
In this condition, h is a positive constant value. If the mentioned condition
[30]. is fulfilled, the reaching time (tr) would be less than tr < jsðt¼0Þj h In the second method, a positive definite Lyapunov function V is considered as follows: 1 Vðx; tÞ ¼ s2 2
(3.9)
To ensure the system stability, the time derivative of V must be negative definite: _ tÞ ¼ s s_ < 0; ss0 Vðx;
(3.10)
To guarantee the finite reaching time, Eq. (3.10) can be modified to _ tÞ ¼ s s_ < hjsj; ss0; h > 0 Vðx;
(3.11)
In the last method, the dynamics of the switching surface are directly defined by the following differential equation: s_ ¼ QsgnðsÞ k f ðsÞ
(3.12)
The above equation is also called the reaching law. Here, Q and k are definite positive gains, and sgn(s) indicates the sign function, which is
82 Control of Power Electronic Converters and Systems
sgnðsÞ ¼
þ1 1
;s > 0 ;s < 0
(3.13)
Different choices for k and Q result in different state trajectories to the sliding plane. Besides, it provides implementing different structures for the reaching law.
3.3.2 Control input calculation The control law in sliding mode usually consists of the following terms: u ¼ ueq þ uun ¼ ueq k sgnðsÞ
(3.14)
where ueq is a linear and continuous part, which is called equivalent control. It is only effective when the system state is in the sliding mode and maintains the system on the sliding surface. It is calculated from s_ ¼ 0/ueq
(3.15)
For calculating ueq it is assumed that uncertainties and disturbances are zeros. These uncertainties and disturbances would be compensated by the second nonlinear and discontinuous part of the control input. uun brings the system state, everywhere on the state plane, to the sliding surface and keeps them on the surface under all bounded disturbances. This term is calculated to accomplish the reachability conditions, which is defined in Section 3.1.
3.4 Classical sliding mode control The control objective in a UPS system is to track the desired or reference output voltage (vod ) under different system uncertainties and disturbances. Therefore, using Eq. (3.2) and doing some manipulations, the tracking error and secondorder derivative of it can be calculated as evo ¼ vo vod 8 > < e€vo ¼ €vo €vod ¼ 1 ðvinv vo Rn ic Þ €vod þ d1 þ d_2 Cn Ln Cn > : e€vo ¼ €vo €vod ¼ 4 þ gvinv þ hðtÞ
(3.16)
(3.17)
where 4 and g are functions based on the system nominal parameters and measured states, and all system uncertainties and unmeasured disturbances are lumped into h(t) as being represented:
Sliding mode controllers in power electronic systems Chapter  3
8 1 > > > g¼ > > C n Ln > > > < 1 4¼ ð vo Rn ic Þ €vod > Cn Ln > > > > > > > hðtÞ ¼ d1 þ d_2 : Cn
83
(3.18)
The system under study is of order 2. Thus, the sliding surface Eq. (3.6) can be defined as sðevo ; tÞ ¼ e_vo þ levo
(3.19)
To have a stable system and an attractive sliding surface (s), the control input must fulfill the following condition based on the Lyapunov stability theorem in Eq. (3.11): ss_ < hjsj; ss0; h > 0 By the time derivative of s, one has _ vo ; tÞ ¼ e€vo þ le_vo ¼ 41 þ gvinv þ hðtÞ sðe 41 ¼ 4 þ le_vo
(3.20)
(3.21)
Based on Eqs. (3.15) and (3.21), the equivalent control law is calculated as 41 _ vo ; tÞ ¼ hðtÞ ¼ 0/41 þ gveq ¼ 0/veq ¼ (3.22) sðe g By considering the nonlinear part of the control law in Eq. (3.14) to guarantee attractivity of s and to compensate for system uncertainties, the final control law can be written vinv ¼ veq þ vun ¼
41 k sgnðsÞ g g
(3.23)
The implementation of the obtained control law in the classic SMC is shown in Fig. 3.3.
FIGURE 3.3 Structure diagram of the classic sliding mode control Eq. (3.23) to control an uninterruptible power supply system.
84 Control of Power Electronic Converters and Systems
By replacing Eqs. (3.21) and (3.23) into Eq. (3.20) and doing some simplifications, one can have ss_ ¼ sð41 þ gvinv þ hðtÞÞ ¼ sðk sgnðsÞ þ hðtÞÞ sðk sgnðsÞ þ HÞ (3.24) where h(t) is bounded and jhðtÞj < H. Choosing k > H þ h: ss_ < hjsj
(3.25)
which completely fulfills Eq. (3.20), and therefore vinv is a stabilizing controller, and s is an attractive surface without any concern about the initial state of system and uncertainties. After that the system reaches to the sliding surface, the error dynamics will be sðevo ; tÞ ¼ e_vo þ levo ¼ 0
(3.26)
evo ðtÞ ¼ evo ð0ÞeðltÞ
(3.27)
Its solution is
which is independent of system uncertainties and system disturbances. However, as it can be seen, the error dynamics on the sliding surface goes exponentially to zero at infinite time, i.e., evo ðtÞ/0 when t/N. It is one major drawback of the classic SMC. It is worth to remark that a sliding mode controller has the two following dynamics, as it is shown in the phase portrait in Fig. 3.4: (1) A reaching phase with limited time ( tr < h1 jsðt ¼ 0Þj ), in which the system states reach the sliding surface in the presence of system uncertainties and disturbances with the help of control input, that is obtained by the reachability condition. (2) A sliding phase with infinite time ðts ¼ NÞ, in which states remain on the sliding surface for all further times and thus converges to the desired state according to the specified dynamic by the sliding surface Eq. (3.26).
FIGURE 3.4 Main phases in sliding mode control of a secondorder system [30].
Sliding mode controllers in power electronic systems Chapter  3
85
3.5 Boundarylayer sliding mode control Although classic SMC is a robust and straightforward control method, however, it suffers from a “chattering” phenomenon due to the unavoidable unmodelled dynamics of the plant and discontinuous sign function in the control law. In many practical applications, chattering makes it impossible to implement the classic SMC. A simple solution to have a continuous/smooth control input is approximating the discontinuous function sgn(s) by some continuous/smooth function like sigmoid function, saturation function, hyperbolic tangent, etc. A good approximation can be achieved by using the saturation function [1]: 8 s < s; jsðevo ; tÞj < ε sat (3.28) ¼ ε : ε sgnðsÞ; jsðevo ; tÞj > ε where ε is a small positive constant. Selecting ε is a compromise between ideal performance and robustness of the ideal SMC and smooth/continuous control action. It is worth to note that under nonideal SMC, the sliding and state variables do not converge to zero at all. However, they go close to the origin ðjevo ðt ¼ NÞj < ε =lÞ, which provides an acceptable performance and accuracy in most applications [30].
3.5.1 Adaptive boundarylayer sliding mode control As discussed in Section 3.5, selection of ε is a tradeoff between maintaining an ideal performance and ensuring a smooth control action. However, tuning of this parameter may be a difficult task in many applications. In Ref. [30], a useful and straightforward adaption law to update this parameter is presented, which ensures a good compromise. The proposed adaption law is ε_ þ lε ¼ kðxd Þ
(3.29)
It is worth to remark that to stabilize system, the control gain must fulfill k > H þ h in Eq. (3.24). Here, H is a bounded, known continuous function of x. To implement the adaption law Eq. (3.29), x in H is replaced by the desired values (xd) [30].
3.6 Adaptive sliding mode control As discussed in Section 3.4, in order to have a stable closedloop system, the control gain k must be greater than the upper bound of the system uncertainties and disturbances, i.e., k > H þ h. If an accurate value of the upper band is not available, the control gain k must be overestimated to have a stable system and an attractive s. However, a high control gain k can lead to too large amplitude and high chattering in the control action, which is not desirable at all. Therefore, an adaption method seems to be necessary to find a proper value of
86 Control of Power Electronic Converters and Systems
control gain that minimizes discontinuity of the control law and consequently reducing the chattering effect. To accomplish this, the Lyapunov function in Eq. (3.9) is modified to include the unknown control gain as [31] 1 1 Vðx; tÞ ¼ s2 þ ke2 ; 2 2r
b r>0 ke ¼ k k;
(3.30)
where k and kb are the accurate and estimated value of the control gain. To ensure the system stability and attractivity of s, the time derivative of V must be negative definite: _ _ tÞ ¼ s s_ þ r1 keke Vðx;
(3.31)
8 > > < k_ ¼ 0 b , and replacing them in Eq. (3.31), By assuming that s_ ¼ mðtÞ ksgnðsÞ > > : jmðtÞj M one has _ b b r1 kb_ k kb _ tÞ ¼ s mðtÞ ksgnðsÞ Vðx; þ r1 kekb ¼ s mðtÞ kjsj (3.32) Considering the following adaption law for the control gain: _ kb ¼ rjsj
(3.33)
Eq. (3.32) can be simplified to _ tÞ Mjsj r1 kb_ k ¼ Mjsj r1 rjsjk ¼ Mjsj kjsj ¼ ðM kÞjsj ¼ hjsj Vðx; (3.34) _ tÞ is a definite negative for all ss 0, where M k ¼ h; h > 0. Therefore Vðx; Eq. (3.30) is a Lyapunov function, and the closed loop is stable, s is attractive, and all signals are bounded under adaption law Eq. (3.33). The implementation of adaptive SMC is shown in Fig. 3.5.
FIGURE 3.5 Structure diagram of the adaptive SMC to control an UPS system.
Sliding mode controllers in power electronic systems Chapter  3
87
3.7 Integral sliding mode control To have better disturbance rejection and zero steadystate error for DC signals, the sliding surface by the integration of an integrator can be modified as ð l2 sðevo ; tÞ ¼ e_vo þ levo þ (3.35) evo dt 4 Thus, the time derivative of the sliding variable results in _ vo ; tÞ ¼ e€vo þ le_vo þ sðe
l2 evo ¼ 42 þ gvinv þ hðtÞ 4
(3.36)
where 42 ¼ 41 þ l4 evo . 2
Based on Eq. (3.36), the equivalent control law is _ vo ; tÞ ¼ hðtÞ ¼ 0/42 þ gveq ¼ 0/veq ¼ sðe
42 g
(3.37)
Again the final control law that must fulfill the attractivity of s ðss_ < hjsj; s s0; h > 0Þ in the presence of uncertainties and disturbances which can be calculated as vinv ¼ veq þ vun ¼
42 k sgnðsÞ; g g
k >Hþh
(3.38)
3.8 Terminal sliding mode control The idea of TSMC was proposed to solve the infinite time convergence of the state variables ðevo Þ when the error dynamics reach and remain on the sliding surface [23,24]. To overcome this issue, the terminal sliding surface is proposed to be p
sðevo ; tÞ ¼ e_ voq þ levo
(3.39)
where p and q must be chosen as two positive odd integers that satisfy the following conditions: 8 >
q > 0 (3.40) p > :1 < q < 2 when the states reach on the sliding surface and they follow the specified dynamics by p
p
sðevo ; tÞ ¼ e_ voq þ levo ¼ 0/ e_ voq ¼ levo
(3.41)
88 Control of Power Electronic Converters and Systems
By integrating both sides of Eq. (3.41) in period (ts ¼ tf tr), and doing some manipulations, one has ðtf
ðtf
p q
q
e_vo ¼ l evo p / tr
ðtf
tr
ðtf
q p
evo devo ¼ l dt/ tr
tr
1 qp qp e ðt Þ e ðt Þ ¼ lðtf tr Þ/ vo f vo r q 1 p
(3.42)
q
evo ðtr Þp Hþh
(3.47)
3.9 Secondorder sliding mode control The primary motivation for the development of SOSMC was to eliminate the chattering. However, it is verified that they cannot provide chatteringfree control action. Although they are able to reduce and adjust chattering effectively [32], it is worth to note that chattering is also reducible in the firstorder SMC (like boundarylayer SMC), however, at the cost of accepting some steadystate output errors. Whereas in SOSMC, the chattering is adjustable, and zero steadystate error is also kept. Another advantage of SOSMC in contrast to the firstorder ones is both sliding (s) and state (evo) variables converge to zero in a finite time, whereas in the firstorder SMC (except TSMC) only the sliding variables converge to zero in a limited time and not the state variables. Before continuing this section, it is appropriate to define the secondorder or higherorder algorithms. In simple words, the SMC algorithm is of degree n, if the control input appears in the nth derivative of s, i.e., 8 > > _ vo ; vinv ; tÞ ¼ 4ðtÞ þ gðtÞvinv þ hðtÞ/First order SMC sðe > > > > > > € vo ; vinv ; tÞ ¼ 4ðtÞ þ gðtÞvinv þ hðtÞ/Second order SMC > < sðe 0 (3.48) sðevo ; vinv ; tÞ ¼ 4ðtÞ þ gðtÞvinv þ hðtÞ/Third order SMC > > > > > > dn ðsðevo ; vinv ; tÞÞ > > ¼ 4ðtÞ þ gðtÞvinv þ hðtÞ/n order SMC > : dtn where 4, g, and h can be linear or nonlinear functions of the system state or disturbances, etc. So far, different algorithms for SOSMC have been proposed as twisting controller, suboptimal algorithm, control algorithm with prescribed convergence law, and quasicontinuous control algorithm [32]. In the following, the twisting controller and its modified versions (supertwisting and adaptive supertwisting controllers) are introduced as the most commonly used SOSMC.
3.9.1 Twisting SMC The twisting algorithm is the first presented SOSMC, which is defined by the following controller equation [32]: _ ; r10 > r20 > 0 (3.49) vinv ¼ veq þ vun ¼ veq r10 sgnðsÞ þ r20 sgnðsÞ
90 Control of Power Electronic Converters and Systems
For this algorithm, the sliding variable can be defined as ( sðevo ; tÞ ¼ evo € sðevo ; tÞ ¼ e€vo ¼ 4 þ gvinv þ hðtÞ;
(3.50)
It can be proved that the twisting controller provides s s_ < 0 and the convergence of s and s_ to zero in a finite time [32]. Since s is equal to evo in Eq. (3.50), therefore, the state variable also converges to zero in a finite time as well. In this algorithm, veq is calculated to guarantee €sðevo ; tÞ ¼ 0, i.e., 4 (3.51) s€ðevo ; tÞ ¼ e€vo ¼ hðtÞ ¼ 0/4 þ gvinv ¼ 0/veq ¼ g And thus the final control law to guarantee a reachability condition can be rewritten as
vinv ¼ veq þ vun ¼ g1 4 þ r1 sgnðsÞ þ r2 sgn s_ ; r1 > r2 > 0 (3.52)
3.9.2 Supertwisting The supertwisting algorithm is not precisely an SOSMC; however, it produces a continuous control action and preserves the accuracy and robustness features of a SOSMC and a classic SMC, respectively. Moreover, unlike the twisting algorithm Eq. (3.49), this method does not require a time derivative of the _ which is another important feature of this method. sliding surface function ðsÞ, The proposed control law for a supertwisting controller is [6,7,32] ð pﬃﬃﬃ vinv ¼ veq þ vun ¼ veq a0 S sgnðsÞ þ b0 sgnðsÞ ; a0 > b0 > 0 (3.53) From Eq. (3.53), it is obvious that both terms of the control input are continuous, and the chattering is attenuated. Considering the following sliding surface of degree one: ( sðevo ; tÞ ¼ e_ vo þ levo (3.54) _ vo ; tÞ ¼ 41 þ gvinv þ hðtÞ sðe The equivalent control law can be calculated as _ vo ; tÞ ¼ hðtÞ ¼ 41 þ gveq /veq ¼ sðe
41 g
And thus the final control law can be rewritten as ð ( pﬃﬃ 1 vinv ¼ veq þ vun ¼ g 41 þ a s sgnðsÞ þ b sgnðsÞ
(3.55)
(3.56)
a > b > 0; In [7,32], some guidelines to tune the control gains ða; bÞ of Eq. (3.56) and stability analysis of the supertwisting controller are investigated. In summary,
Sliding mode controllers in power electronic systems Chapter  3
91
FIGURE 3.6 Structure diagram of the supertwisting algorithm Eq. (3.56) to control an uninterruptible power supply system.
if the upper bound of disturbance is known Eq. (3.24), a good start to choose and tune the control gains would be ( pﬃﬃﬃﬃ a ¼ 1:5 H (3.57) b ¼ 1:1H where H is upper bound of the system disturbance h(t), and is already defined in Eqs. (3.18) and (3.24). The implementation of the obtained control law in the supertwisting controller is shown in Fig. 3.6.
3.9.3 Adaptive supertwisting To avoid trial and error to find proper values of control gains ða; bÞ which have an essential effect on the performance of the supertwisting algorithm, they can be calculated based on the following adaption laws [33]: 8 > C1 > 0; ss0 >
> :b b þ C3 ; C2 &C3 > 0 b ¼ C2 a Consequently, based on estimated control gains, the control input can be updated as ð pﬃﬃ 1 b s sgnðSÞ þ a b sgnðsÞ (3.59) vinv ¼ g 41 þ a
3.10 Comparison of different SMCs The different discussed SMCs are summarized in Table 3.1. The controllers are calculated for the studied UPS system. However, these equations are applicable to any secondorder systems, which have the following form: e€¼ 4 þ gvinv þ h
(3.60)
TABLE 3.1 Comparison of different sliding mode control (SMCs) used for a secondorder system. Stability Controller Classic SMC Boundarylayer SMC Adaptive boundarylayer SMC Adaptive SMC
PI SMC
Terminal SMC
Control input
Chattering
Sliding variables
State variables
s ¼ e_ þ le s_ ¼ v þ h
v ¼ ksignðsÞ
High
Finite time
Asymptotically
s ¼ e_ þ le s_ ¼ v þ h
v ¼ k sat
Low
Lyapunov stable
Bounded
Low
Lyapunov stable
Bounded
Low
Lyapunov stable
Bounded
v ¼ ksignðsÞ
High
Finite time
Asymptotically
v ¼ ksignðsÞ
High
Finite time
Finite time
_ v ¼ asignðsÞ bsignðsÞ
Low
Finite time
Finite time
pﬃﬃﬃﬃﬃ R v ¼ a jsjsignðsÞ b signðsÞ
Low
Finite time
Asymptotically
Sliding variables
s ¼ e_ þ le s_ ¼ v þ h s ¼ e_ þ le s_ ¼ v þ h
8 2Z > < s ¼ e_ þ le þ l edt 4 > : s_ ¼ v þ h 8 > > < s ¼ e_pq þ le
s ε
; ε ¼ constant
8
< v ¼ k sat s ; ε : ε_ þ lε ¼ kðxd Þ 8 s
> < v ¼ k sat ; ε > : k_ ¼ rjsj
> > : s_ ¼ v þ h Twisting SMC Supertwisting SMC
s¼e s€ ¼ e€ ¼ v þ h s ¼ e_ þ le s_ ¼ v þ h
Sliding mode controllers in power electronic systems Chapter  3
93
3.11 Experimental results In order to evaluate the performance of SMCs in practical situations, a laboratory prototype has been built, which includes a threephase 5 kW PWMVSC, which is supplied from a constant DC voltage, threephase resistive load, and LCtype output filter (Fig. 3.7). Moreover, the control methods are realized on a DS1007 dSPACE system. To measure capacitor currents and voltages, the DS2004 highspeed A/D board, and to apply generated switching pulses, the DS5101 digital waveform output board are employed. The system and control parameters of threephase UPS are given in Table 3.2. To evaluate and compare the SMCs features, the most conventional SMC method, i.e., classical SMC with the boundarylayer, and one advanced control method, i.e., supertwisting controller, are chosen and implemented in the following subsection. It is worth to remark that the control gains of both control methods are
FIGURE 3.7 Experimental setup to implement sliding mode controls on a threephase voltage source inverter in uninterruptible power supply application.
94 Control of Power Electronic Converters and Systems
TABLE 3.2 System and control parameters for threephase UPS system. System parameters Nominal power
5 [kW]
Line voltage (rms)
380 [V]
Output frequency (f)
50 [hz]
Inductor and the series resistance (Ln, Rn)
3 [mH], 0.3 [U]
Capacitor (Cn)
15 [mF]
DClink voltage (Vdc)
720 [V]
Sampling frequency
20 [kHz]
Switching frequency
10 [kHz]
Control parameters Classical sliding mode control k
8.6e9
l
1e4
ε
1.2e6
Supertwisting controller a
7e5
b
7e11
calculated based on the guidelines in Sections 3.4 And 3.5 and Section 3.9.2 and then doing some trials and errors, to finally tune the system.
3.11.1 Steadystate performance Steadystate performance of classical SMC and supertwisting controller under nominal load is shown in Fig. 3.8. Both methods have excellent steadystate performance. It is worth to note that, the steadystate error of the supertwisting controller (evo ¼ 2.8 V) is improved to half of the classic SMC (evo ¼ 5.3V), whereas both control methods have approximately the same chattering in the control inputs. As discussed in Section 3.5, the steadystate tracking error of classic SMC can be reduced by reducing the boundary layer. However, it increases the control input chattering and also vulnerability against noise. The effect of the boundarylayer thickness on the steadystate error and also control input chattering is investigated in Fig. 3.9, which confirms the discussion in Section 3.5.
Sliding mode controllers in power electronic systems Chapter  3
95
FIGURE 3.8 Obtained experimental results showing steadystate performance of two control methods under nominal load (5 kW) for uninterruptible power supply system, (A) classical sliding mode control, (B) supertwisting controller.
96 Control of Power Electronic Converters and Systems
FIGURE 3.9 Obtained experimental results showing steadystate performance of classic sliding mode control under two different boundarylayer selection (A) ε ¼ 1e6 , (B) ε ¼ 1:7e6 .
3.11.2 Dynamic performance The dynamic performance of two control methods under a step change in the resistive load is shown in Fig. 3.10. In this figure, the load power has been suddenly changed from zero to the nominal value. It shows good transient response and disturbance rejection of both control methods. From this figure, it can be seen that the classic SMC has a little better transient performance than the supertwisting controller, due to larger gain in the nonlinear part of the control input. However, it is achieved on the cost of more control input chattering and steadystate errors, as discussed in the previous sections.
3.12 Conclusions Thanks to the simple structure and concept, fast dynamic response, and good robustness and disturbance rejection, SMC has attracted considerable attention from power electronic researchers in recent years. In this chapter, the basic concept of different types of SMCs is investigated. The two most important SMCs, i.e., classic and supertwisting, have been implemented to make voltage control of a UPS. Experimental results confirm their excellent performance in both steadystate and transient operation as well as good load current disturbance rejection.
Sliding mode controllers in power electronic systems Chapter  3
FIGURE 3.10 Obtained experimental results showing transient performance of two control methods under step change of load from zero to 5 kW, (A) classic sliding mode control, (B) supertwisting controller.
97
98 Control of Power Electronic Converters and Systems
References [1] M. Rezkallah, S.K. Sharma, A. Chandra, B. Singh, D.R. Rousse, Lyapunov function and sliding mode control approach for the solarPV grid interface system, IEEE Trans. Ind. Electron. 64 (1) (January, 2017) 785e795. [2] A. Merabet, L. Labib, A.M.Y.M. Ghias, C. Ghenai, T. Salameh, Robust feedback linearizing control with sliding mode compensation for a gridconnected photovoltaic inverter system under unbalanced grid voltages, IEEE J. Photovoltaics 7 (3) (2017) 828e838. [3] N. Kumar, T.K. Saha, J. Dey, Slidingmode control of PWM dual inverterbased gridconnected PV system: modeling and performance analysis, IEEE J. Emerg. Sel. Top. Power Electron. 4 (2) (June, 2016) 435e444. [4] D. Sun, X. Wang, H. Nian, Z.Q. Zhu, A slidingmode direct power control strategy for DFIG under both balanced and unbalanced grid conditions using extended active power, IEEE Trans. Power Electron. 33 (2) (2018) 1313e1322. [5] L. Xiong, J. Wang, X. Mi, M.W. Khan, Fractional order sliding mode based direct power control of gridconnected DFIG, IEEE Trans. Power Syst. 33 (3) (May, 2018) 3087e3096. [6] R. Sadeghi, S.M. Madani, M. Ataei, M.R. Agha Kashkooli, S. Ademi, Supertwisting sliding mode direct power control of a brushless doubly fed induction generator, IEEE Trans. Ind. Electron. 65 (11) (2018) 9147e9156. [7] C. Lascu, A. Argeseanu, F. Blaabjerg, Supertwisting slidingmode direct torque and flux control of induction machine drives, IEEE Trans. Power Electron. 35 (5) (May, 2020) 5057e5065. [8] G. Sun, Discretetime fractional order terminal sliding mode tracking control for linear motor, IEEE Trans. Ind. Electron. 65 (4) (2018) 3386e3394. [9] A. Pilloni, A. Pisano, E. Usai, Robust finitetime frequency and voltage restoration of inverterbased microgrids via slidingmode cooperative control, IEEE Trans. Ind. Electron. 65 (1) (2018) 907e917. [10] H. R. Baghaee, M. Mirsalim, G. B. Gharehpetian, and H. A. Talebi, “A decentralized power management and sliding mode control strategy for hybrid AC/DC microgrids including renewable energy resources,” IEEE Trans. Ind. Informatics, https://doi.org/10.1109/TII. 2017.2677943. [11] R.P. Vieira, L.T. Martins, J.R. Massing, M. Stefanello, Sliding mode controller in a multiloop framework for a gridconnected VSI with LCL filter, IEEE Trans. Ind. Electron. 65 (6) (June, 2018) 4714e4723. [12] R. Guzman, L.G. De Vicun˜a, M. Castilla, J. Miret, J. De La Hoz, Variable structure control for threephase LCLfiltered inverters using a reduced converter model, IEEE Trans. Ind. Electron. 65 (1) (2018) 5e15. [13] M. Huang, Q. Tan, H. Li, W. Wu, Improved sliding mode control method of singlephase LCL filtered VSI, in: 2018 9th IEEE Int. Symp. Power Electron. Distrib. Gener. Syst. PEDG vol. 2018, 2018, pp. 1e5, 2. [14] T.G. Converters, I.M. Hassine, M.W. Naouar, N. Mrabetbellaaj, Model predictivesliding mode control for, IEEE Trans. Ind. Electron. 64 (2) (2017) 1341e1349. [15] N.M. Dehkordi, N. Sadati, M. Hamzeh, A robust backstepping highorder sliding mode control strategy for gridconnected DG units with harmonic/interharmonic current compensation capability, IEEE Trans. Sustain. Energy 8 (2) (2017) 561e572. [16] H. Komurcugil, S. Ozdemir, I. Sefa, N. Altin, O. Kukrer, Slidingmode control for singlephase gridconnected LCL filtered VSI with doubleband hysteresis scheme, IEEE Trans. Ind. Electron. 63 (2) (February, 2016) 864e873.
Sliding mode controllers in power electronic systems Chapter  3 [17]
[18]
[19]
[20]
[21]
[22] [23]
[24] [25]
[26] [27] [28] [29] [30] [31]
[32] [33]
99
J. Saeed, J.P. Mishra, L. Wang, X. Yu, A secondorder sliding mode voltage oriented control of threephase Active front end rectifier, in: 2018 IEEE 27th International Symposium on Industrial Electronics (ISIE) vol. 2018, June, 2018, pp. 175e180. J. Liu, S. Vazquez, S. Member, L. Wu, S. Member, Extended state observerbased slidingmode control for threephase power converters, IEEE Trans. Ind. Electron. 64 (1) (2017) 22e31. R. Guzman, L.G. de Vicuna, J. Morales, M. Castilla, J. Matas, Slidingmode control for a threephase unity power factor rectifier operating at fixed switching frequency, IEEE Trans. Power Electron. 31 (1) (January, 2016) 758e769. Z. Li, C. Zang, P. Zeng, H. Yu, S. Li, J. Bian, Control of a gridforming inverter based on slidingmode and mixed H2\Hinf control, IEEE Trans. Ind. Electron. 64 (5) (May, 2017) 3862e3872. J. Morales, L.G. De Vicuna, R. Guzman, M. Castilla, J. Miret, Modeling and sliding mode control for threephase Active power filters using the vector operation technique, IEEE Trans. Ind. Electron. 65 (9) (2018) 6828e6838. Y. Chen, J. Fei, Dynamic sliding mode control of active power filter with integral switching gain, IEEE Access vol. 7 (c) (2019) 21635e21644. W. Le Zhu, X. Yang, F. Duan, Z. Zhu, B.F. Ju, Design and adaptive terminal sliding mode control of a fast tool servo system for diamond machining of freeform surfaces, IEEE Trans. Ind. Electron. 66 (6) (2019) 4912e4922. C. Mu, H. He, Dynamic behavior of terminal sliding mode control, IEEE Trans. Ind. Electron. 65 (4) (2018) 3480e3490. S.L. Jung, Y.Y. Tzou, Discrete slidingmode control of a PWM inverter for sinusoidal output waveform synthesis with optimal sliding curve, IEEE Trans. Power Electron. 11 (4) (July, 1996) 567e577. J.F. Silva, Slidingmode control of boosttype unitypowerfactor PWM rectifiers, IEEE Trans. Ind. Electron. 46 (3) (June, 1999) 594e603. V.I. Utkin, Sliding mode control design principles and applications to electric drives, IEEE Trans. Ind. Electron. 40 (1) (1993) 23e36. J.Y.J.C. Hung, W. Gao, J.Y.J.C. Hung, Variable structure control: a survey, IEEE Trans. Ind. Electron. 40 (1) (1993) 2e22. ¨ zgu¨ner, “A control engineer’s guide to sliding mode control, ¨. O K.D. Young, V.I. Utkin, U IEEE Trans. Contr. Syst. Technol. 7 (3) (1999) 328e342. J. Slotine, W. Li, Applied Nonlinear Control, Pearson, 1990. Y.J. Huang, T.C. Kuo, S.H. Chang, Adaptive slidingmode control for nonlinear systems with uncertain parameters, IEEE Trans. Syst. Man Cybern. B Cybern. 38 (2) (2008) 534e539. Y. Shtessel, C. Edwards, L. Fridman, A. Levant, Sliding Mode Control and Observation, Birkha¨user, 2014. Y.B. Shtessel, J.A. Moreno, F. Plestan, L.M. Fridman, A.S. Poznyak, Supertwisting adaptive sliding mode control: a Lyapunov design, in: 49th IEEE Conference on Decision and Control (CDC), 2010, pp. 5109e5113.
This page intentionally left blank
Chapter 4
Model predictive control of power converters, motor drives, and microgrids Zhenbin Zhang1, Tomislav Dragicevic2, Yu Li1, Yongdu Wang1, Changming Zheng4, Minrui Leng5, Jose Rodriguez3 1
School of Electrical Engineering, Shandong University, Jinan, China; 2Department of Electrical Engineering, Technical University of Denmark, Lyngby, Denmark; 3Faculty of Engineering, Universidad Andres Bello, Santiago, Chile; 4School of Electrical and Power Engineering, China University of Mining and Technology, Xuzhou, Jiangsu, China; 5School of Electrical Engineering, Southwest Jiaotong University, Chengdu, China
4.1 Introduction on MPC In this chapter, the basic concept and classification of model predictive control (MPC) will be introduced first. Afterward, typical applications of MPC in electrical drives, microgrids (MGs), and wind generation will be presented in detail. Finally, conclusions and future trends in power electronic systems are discussed. MPC, also referred to as receding horizon control, was introduced to petrochemical industrial processes in the early 1970s [1,2]. MPC takes the entire system model into consideration and can penalize the (multiple) system control targets with a flexible cost function. The optimum values of the actuating variables are not computed based on the “posterror” between the reference and the feedback signals. Instead, this is done through minimizing a flexiblydesigned cost function (a.k.a. objective function) with penalized “predicted behaviors” of the system, fully using the system model and the past control actions over a receding prediction horizon [2e4]. MPC has more freedom to further improve the system control performance, since the “predicted behaviors” of the system are utilized within the control/decision process. Additionally, using a cost function to define the control targets makes it more straightforward (at least from a concept perspective) for more complicated systems with multiple control targets. The objective of MPC is to ensure that the output tracks the reference with minimal error, given only information about the system model and the state of Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000172 Copyright © 2021 Elsevier Ltd. All rights reserved.
101
102 Control of Power Electronic Converters and Systems
the controlled output [5]. As illustrated in Fig. 4.1, the measured variable ! x ½k ! and predictive model are used to predict the output y ½k . This predicted output is taken through an optimization algorithm to generate the input switching signal, ! u ½k , that will minimize the cost function of the control objectives. The reference output is ! y ½k . The measured variable ! x ½k could be current, voltage, or frequency. Table 4.1 shows the categories of MPC techniques. These strategies are broadly grouped as finite control set MPC (FCSMPC) and continuous control set MPC (CCSMPC), as shown in Fig. 4.2. Generally, FCSMPC does not require a modulator, while CCSMPC utilizes a modulator for its operation. Thus, there is an inherent advantage of CCSMPC, which is the fact that it has constant switching frequency, while FCSMPC has variable and dynamic switching frequency. This latter quality creates some undesirable harmonics in the output signal. As illustrated in Table 4.1, FCSMPC comprises optimal switching vector MPC (OSVMPC) and optimal switching sequence MPC (OSSMPC). CCSMPC includes generalized predictive control (GPC) and explicit model predictive control (EMPC). The MPC concept for controlling power electronics and electrical drives is applied in a variety of ways. As previously mentioned, CCSMPC requires an extra modulator to generate the switching sequences based on the controller outputs, and these are usually continuous values of duty cycles or reference voltages (see Fig. 4.1). In contrast, FCSMPC combines both the cost optimization and modulation into one single process and directly gives the output switching sequence. No extra modulation scheme is required (see Fig. 4.3). FCSMPC classes of MPC are useful to solving both linear and nonlinear problems with multiple constraints and control objectives. EMPC can also be applied to nonlinear and constrained systems. On the other hand, GPC is only useful for linear unconstrained systems. FCSMPC is relatively easier than CCSMPC because it is more intuitive and does not lead to any complex algorithms. The computational cost of EMPC is the lowest of the four categories because there is no need for online optimization at each sampling time. FCSMPC has a shortcoming with its prediction horizon which is relatively short because longer horizon would increase the computational burden excessively.
4.2 MPC for permanentmagnet synchronous motor drives In this section, the model of permanentmagnet synchronous motor (PMSM) is firstly described in Section 4.2.1, and the direct model predictive current control schemes with prediction horizonone for twolevel (2L) power converter of PMSM is discussed in Section 4.2.2. Its performances are evaluated with simulation results in Section 4.2.3. Summary is given in Section 4.2.4.
(A)
(B)
Model predictive control Chapter  4
FIGURE 4.1 Continuous control set MPC concept (R. Kennel et al., 1983). (A) The reference trajectory ! y ½k and predicted trajectory ! y ½k with continuous control set MPC. (B) The block diagram of continuous control set MPC for power electronics and electrical drives.
103
104 Control of Power Electronic Converters and Systems
TABLE 4.1 Classification of model predictive control [5]. Finite control set MPC (FCSMPC)
Continuous control set MPC (CCSMPC)
Optimal switching vector MPC (OSVMPC) [6e9]
Optimal switching sequence MPC (OSSMPC) [10e12]
Generalized predictive control (GPC) [13,14]
Explicit MPC (EMPC) [15e17]
Problem type
Linear and nonlinear
Linear and nonlinear
Linear unconstrained
Nonlinear constrained
Computational cost
High
Highest
Lowest
Low
Modulator
Not required
Not required
SVM and PWM
SVM and PWM
Switching frequency
Variable
Fixed
Fixed
Fixed
Cost function minimization
Online
Online
Online
Offline
Constraints
Allowed
Allowed
Allowed (with higher computational burden)
Allowed
Prediction horizon
Mostly short
Mostly short
Long
Long
Algorithm complexity
Intuitive
Intuitive
High
High
Description
FIGURE 4.2 Categories of MPC schemes.
(A)
(B)
Model predictive control Chapter  4
FIGURE 4.3 Finite control set MPC concept (J. Rodriguez et al., 2004). (A) The reference trajectory ! y ½k and predicted trajectory ! y ½k with finite control set MPC. (B) The block diagram of finite control set MPC for power electronics and electrical drives.
105
106 Control of Power Electronic Converters and Systems
4.2.1 Model of PMSM In many industrial cases, for an electricdrive system, a smooth surface multipole (pole pair defined as Np) PMSM is used [18]. Both the physical (structural) and saturation salience are, in practice, quite small and can be q 1 s neglected, i.e., Lds ¼ Ls ; vL vqe ¼ 0. Moreover, the armature reaction affect is assumed to be negligible. l
PMSM in ab frame
A smooth surface multipole PMSM can be modeled in abreference frame. The dynamics in the current can be described with the following nonlinear model 0 1 B C B !ab jpm ue ðtÞsinðqe ðtÞÞ ! C C d i m ðtÞ Rs !ab 1B ab ! C; ¼ v m ðtÞ i ðtÞ þ B C dt Ls B Ls m B jpm ue ðtÞcosðqe ðtÞÞ C @ ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} A ab ¼:! e m ðtÞ !ab !0 i m ð0Þ ¼ i m ˛ R 2 ;
(4.1)
T ab where Rs [U] is the stator resistance, ! e m ¼ eam ; ebm ½V2 is the backEMF T ab vector, ! v m ¼ vam ; vbm ½V2 is the output voltage vector of the converter, !ab a b T 2 i m ¼ im ; im ½V is the stator current vector: all in ab frame. ue ðtÞ ¼ Np um ðtÞ½rad =s is the electrical angular velocity of the rotor (rotating with um), Np [1] is the pole pair number, jpm [Wb] is the permanentmagnet flux linkage, qe [rad] is the electrical position of the rotor flux. The stator flux and electromagnetic torque in the ab frame can be modeled as Z !ab !0 !ab !ab ab ! j s ðtÞ ¼ v m ðtÞ Rs i m ðtÞ dt; j s ð0Þ ¼ j s ˛ ℝ2 ; (4.2) Te ðtÞNp jas ibm jbs iam
(4.3)
The dynamics of the mechanical system are given by Qm
dum ðtÞ ¼ Te ðtÞ T1 ðtÞ ¼ F ðum ÞðtÞ; um ð0Þ ¼ u0m ˛ ℝ; dt
(4.4)
where Qm [kgm2] is the overall inertia (of the whole motor), Te [Nm] is the electromagnetic torque, T1 [Nm] is the torque from the load, and F ðum Þ
1. For salience based encoderless control, the true property of
vLs vqe s0
is utilized.
Model predictive control Chapter  4
107
models nonlinear, dynamic friction effects; in this work, a constant coefficient B is used to model the friction effects. Applying the Eulerforward method, the following discrete format of the motor in ab frame can be obtained: 0 1 9 > > > > > B C > ! > B C j u sinðq Þ > e½k pm e½k > B C ab ab T s Rs ! Ts B!ab ! > C; > i m½kþ1 ¼ 1 i m½k þ B v m½k > > C > Ls Ls B > C j u cosðq Þ > e½k pm e½k @ A > ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} > > > > ab > ¼:! e e½k > = > !ab !ab !ab ab > > j s½kþ1 ¼ j s½k þ ! v m½k Rs i m½k Ts ; > > > > > > 0 1 > > > > > Ts Np B a b > C > b a > ue½kþ1 ¼ ue½k þ @Np js½k im½k js½k im½k T1½k B,um½k A: > > > Qm ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} > > > ; ¼:Te½k
(4.5)
l
PMSM in dq frame
The mathematical model of a PMSM in directquadrature (dq) reference frame (indicated by superscript dq) is given by
!dq 0 1 !dq !dq d j s ðtÞ !0 !dq dq ! þ ue ðtÞ v m ðtÞ ¼ Rs i s ðtÞ þ j s ; j s ð0Þ ¼ j s ˛ ℝ2 dt 1 0 ﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄ} :¼J
(4.6) T dq where ! v m ðtÞ ¼ vdm ðtÞ; vqm ðtÞ ½V2 is the converter output voltage vector (to T !dq q be specified later), Rs is the stator resistance, i m ðtÞ ¼ idm ðtÞ; im ðtÞ ½A2 is !dq q T the stator current vector, j s ðtÞ ¼ jds ðtÞ; js ðtÞ ½Wb2 is the flux linkage (in the stator of the motor). The flux linkage is assumed linearly related to the !dq current i m ðtÞ, stator inductance Ls [Vs/A] and (constant) permanentmagnet flux linkage jpm as follows: T !dq !dq (4.7) j s ðtÞ ¼ Ls i m ðtÞ þ jpm ; 0
Te ðtÞ ¼ Np jpm iqm :
(4.8)
108 Control of Power Electronic Converters and Systems
Taking equations (4.4), (4.6)e(4.8) into consideration, and applying the Eulerforward method, yields the discrete format of the generator as 0
!dq i m½kþ1
Ts Rs B 1 Ls B ¼B B @ Ts ue½k
2 1 Ts Ts ue½k C 6 Ls 6 C!dq 6 Ci C m½k þ 6 4 Ts Rs A 1 0 Ls
3 0 19 07 0 > > B 7!dq C> > > B 7v C > þ m½k @ Tjpm 7 A> > > Ts 5 ue½k > > > > Ls > ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} > Ls > = ¼:H m½k
T !dq !dq j s½kþ1 ¼ Ls i m½kþ1 þ jpm ; 0 ; ue½kþ1 ¼ ue½k þ
Ts Np Np jpm iqm½k T1½k B,um½k Qm ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ}
> > > > > > > > > > > > > > > > ;
¼:Te½k
(4.9)
4.2.2 FCSMPC for PMSM drives For a twolevel power converter driving PMSM system depicted in Fig. 4.4A, the torque tracking subject to torque per Ampere” a socalled “maximum Lds Lqs d 2 q 2 d im im ¼ 0 ) is desired. In the setup, (MTPA) law (i.e.im þ j pm
q
Lds zLs ¼ Ls . Therefore, id m :¼ 0 can be set in the controller. The current reference is generated by a proper outer control loop (here a PI controller regulating the speed control is used for generating the qaxis current and the daxis current reference is set to be zero for an MTPA control. These references are then transferred into ab frame to assign to the inner predictive
(A)
(B)
FIGURE 4.4 FCSMPC scheme for the twolevel power converterefed PMSM system. (A) PMSM drive system under control with twolevel converter. (B) Control diagram of direct model predictive control.
Model predictive control Chapter  4
109
current control loop). The inner loop itself can be designed in the ab frame to eliminate the otherwise required synchronous frame transformations. This method is also named direct model predictive current control (DMPCC) because the main objective here is the current of the motor. Therefore, the cost m function JDMPCC for predictive current control is defined as 2 2 b m a b ! ! ! i þ i i : (4.10) JDMPCC u m ¼ ia u u m m m m½kþ1 m m½kþ1 is calculated by Eq. (4.5), with ! um˛ The predicted current vector of iab m½kþ1 U 8 . Note that because the currents in both a and baxis are equally important to the system, no extra weightings are required for these targets. After evaluating and minimizing the costs obtained from Eq. (4.10) for w! um ˛ U8 for the ! 1 ! twolevel converter, an optimal gate vector of G ¼ G u will be m
m
obtained and assigned to the motor drive converter. The overall control diagram of DMPCC is given in Fig. 4.4B.
4.2.3 Performance evaluations of MPC with simulations In this section, the control performances of the aforementioned direct model predictive current control with simulation results are presented. In the following, the overall testing scenarios are as follows: It is assumed that the optimal speed reference um is already known. A rated torque is mounted under such (fast) speed changing rate to test the harshest operational situations. The control interval is set to be 50[ms], and no compensation is inserted into the predictions. The overall control performances are shown in Fig. 4.5A. The zoomedin steadystate control performances of the motor currents are illustrated in Fig. 4.5B. As it can be seen, besides its good control dynamics, a big change of the switching frequency is seen. The current tracking performances are also quite good.
4.2.4 Summary In this section, the direct model predictive current control and its application guidelines have been summarized. Based on this, its application has been demonstrated on a twolevel power converter driving PMSM system taking “currents” as the targeting set.2 In comparison with switching tableebased direct torque and power control methods, the direct MPC methods, by using the cost functionebased solution, make both the controller design and tuning process much more straightforward. In particular when the frequency regulation and DClink capacitor voltage balancing control are considered with
2. This concept can also be applied to deal with “flux” and “voltage” control.
110 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 4.5 Simulation of FCSMPC for the twolevel power converterefed PMSM system. From top to bottom for subfigure (A) are motor speed, motor torque, and motor stator current in a phase, respectively. Subfigure (B) illustration of the zoomed performance of the electromagnetic torque, machine phase currents, and corresponding Fourier analysis, respectively.
more level converters. The shown simulation results using PLECS software validates that FCSMPC is a promising solution for motor drive system.
4.3 MPC for microgrids Voltage source converters (VSCs) act as interfaces between the distributed energy resources and the MG [19]. Reliable control of VSCs is a critical function to achieve a high penetration of renewable energy on such systems. Generally, MG can be configured as AC or DC architecture, which can either work in standalone or in gridconnected mode, as it is shown in Fig. 4.6
FIGURE 4.6 Basic structure of an MG which can be either AC or DC configured.
Model predictive control Chapter  4
111
[20e22]. In this chapter, both the applications of MPC in AC and DC MGs are discussed.
4.3.1 Dynamics and predictive model of VSCbased MG In this section, the dynamics of AC and DC MG are modeled. Typically, an MG consists of multiple VSCs using LC filters in the AC and/or DC side (see Fig. 4.6). l
Loadfilter side dynamics: LC filter
A twolevel threephase VSC with an output LC filter connected to a balanced load is described in stationary ab frame as [23] 3 2 2 3 Rf 1 1 7
0 7
6 6 Lf Lf 7 i f d if 7 u 6 Lf 6 ¼6 þ6 (4.11) 7 7 4 1 4 5 vf dt vf 1 5 io ﬄﬄﬄﬄ{zﬄﬄﬄﬄ} ﬄﬄ{zﬄﬄ} ﬄﬄ{zﬄﬄ} 0 0 Cf Cf _ uðtÞ xðtÞ ﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄ} ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} xðtÞ B
A
where Lf and Cf are filter inductance and capacitance. The filtercapacitor voltage vf ¼ vfaþjv fb, inductor current if ¼ ifa þ jifb, converter voltage vector u ¼ ua þ jub, and load current io ¼ ioa þ jiob. Correspondingly, the ZOH discretizationebased predictive model is constructed using Eq. (4.11) as [24]
if ½kþ1 if ½k u½k ¼ Ad þ Bd (4.12) vf ½kþ1 vf ½k io½k RT where Ad ¼ eATs ; Bd ¼ 0 s eAs Bds; a and Ts is the sampling time. l
DClink dynamics
Apart from the dynamics of the AC side, dynamics are also derived for the DC link. Here, only a differential equation describing vdc is used for modeling the DC link dynamics, where the current idc, which flows through the inductor, is treated as an external disturbance Cdc
dvdc ¼ idc ipol dt
(4.13)
where ipol is the current flowing into the inverter, which can be synthesized from the filter current and the gating signals. For the DClink dynamics, the following approximation is used on the DC side to estimate how much the DClink capacitor is charged/discharged during each sample period, i.e., using predictive DCbus voltage model
112 Control of Power Electronic Converters and Systems
vdc½kþ1 ¼ vdc½k þ
1 ipol;i þ ipol;p idc Ts Cdc 2
(4.14)
ipol,i and ipol,p are the initial and final current flowing into the inverter during the following time step, respectively. The above discrete models on the AC side and DC side are used to predict if, vf, and vdc at the end of the next sampling instant in order to control the MG systems.
4.3.2 MPC for robust and fast operation of an islanded AC MG An islanded AC MG usually consists of multiple parallel VSCs, which are connected to a common AC bus (see Fig. 4.6). The typical primary control contains normally of two control loops. Outerloop control (e.g., droop control) aims to share the load power, while the innerloop control (e.g., cascaded linear control) is responsible for voltage and frequency regulation. Generally, MPC is applied in the inner control loop to enhance the reference voltage tracking and system dynamic performance, and the cost function is defined as [25] 2 2 gac ¼ vf a vf a½kþ1 þ vf b vf b½kþ1 þ ld gder (4.15) with a voltage reference derivative tracking term as 2 2 (4.16) gder ¼ Cf u vf b if a½kþ1 þ ioa½k þ Cf u vf a þ if b½kþ1 iob½k where u and vf are the voltage and frequency reference generated by the outer droop control loop. ld is a weighting factor [26]. The predicted vf ½kþ1 and if ½kþ1 are obtained by Eq. (4.12). Note that Eq. (4.15) simultaneously tracks both voltage reference and its derivative, which can achieve an improved static performance compared to the conventional MPC with only voltage reference tracking [25]. After evaluating
FIGURE 4.7 Block diagram of MPC for an islanded AC MG and a droop control for the outer loop.
Model predictive control Chapter  4
113
and minimizing the costs obtained from Eq. (4.15) for each converter voltage vector u, an optimal voltage vector will be achieved and applied to the VSC. The block diagram of the MPC for an islanded AC MG is shown in Fig. 4.7.
4.3.3 Dynamic stabilization of a DC MG using MPC As shown in Fig. 4.6, the control objective is to achieve an excellent voltage regulation performance on the AC side as well as a stable voltage on the DC side. The cost function used for ACside tracking is shown in Eq. (4.15). For the DC side, a simple solution to the instability problem is to introduce a stabilization term gdc ¼ vdc vdc . As a result, the complete cost function for a DC MG is formed as [27] gp ¼ gac þ ldc gdc
(4.17)
where gac is given in Eq. (4.15), and ldc is the weighting factor, which balances the performance between AC and DC side. A good load performance is achieved with a low setting of ldc, while a higher ldc ensures better DClink dynamic response. Hence, an adaptive selection of ldc can be employed as lnð10Þ ldc ¼ 0:1$exp vdc vdc $ (4.18) 5 Similarly, by evaluating the cost function gp, the optimal actuation can be determined.
4.3.4 Performance evaluation with experimental results In this section, the experimental results for validating the MPC schemes for stabilization of AC and DC MGs are evaluated in Figs. 4.8 and 4.9 with the overall experimental parameters as listed in Table 4.2.
(A)
(B)
FIGURE 4.8 [Measured:] Overall control performance with innerloop MPC (Eq. 4.15) for an islanded AC MG. (A) Steadystate loadside voltage (Voltage THD: 1.10%). (B) Dynamic response of loadside voltage under a load step change from open circuit to Rl ¼ 33 U.
114 Control of Power Electronic Converters and Systems
(A)
(B)
(C)
FIGURE 4.9 [Measured:] Overall control performance of stabilization for DC link and step load change for a DC MG using MPC in Eq. (4.17). (A) ldc ¼ 0.1. (B) ldc ¼ 1. (C) Adaptive ldc using Eq. (4.18).
TABLE 4.2 Experimental Parameters of an MG as shown in Fig. 4.6. Parameter description
Symbols and values
DCbus voltage
Vdc ¼ 700 V
Load reference voltage for AC MG
Vref ¼ 347 V, fref ¼ 50 Hz
Load reference voltage for DC MG
Vref ¼ 208 V, fref ¼ 50 Hz
VSC dead time
Td ¼ 4 ms
Sampling time Ts
Ts ¼ 25 ms
ACside LC filter
Lf ¼ 2.4 mH, Cf ¼ 25 mF
DCside LC filter
Ldc ¼ 5 mH, Cdc ¼ 30 mF
Linear resistive load
Rl ¼ 33 U
Model predictive control Chapter  4
115
As it can be seen in Fig. 4.8, for the AC side, a very fast dynamic response of the voltage under a load step change can be achieved. Meanwhile, the steadystate performance is also quite good with a low THD. As is shown in Fig. 4.9, for the DC side, with ldc ¼ 0.1, a significant ringing can be seen on the DC side. On the other hand, for ldc ¼ 1, very good dynamic performance on the DC link can be observed, but there is a drop in the steadystate voltage drop on the AC side. Finally, with the employed adaptive ldc in Eq. (4.18), the DC link is very fast stabilized after a step change in the load, while the loadside voltage has excellent dynamic response and negligible steadystate tracking error.
4.3.5 Summary It can be seen that the MPCbased stabilization method, on the one hand, stabilizes an MG without implementing any additional active or passive components, thus providing higher energy efficiency and better costeffectiveness than methods that rely on such components. On the other hand, this method has a significantly lower influence on ACside voltage regulation performance.
4.4 MPC for renewable energy applications (PMSG wind turbine) Wind energy, in particular high power wind energy installation, has steadily increased over the last decade. Currently, 12 MW systems are available in the market and numerous research activities aim at 10e14 MW level for offshore applications. For such demands, wind turbine systems (WTSs) using full power scale backtoback power converter and permanentmagnet synchronous generator (PMSG) with direct drive structure (without mechanical gear) have attracted much interest over the last decade. Compared with doublefed induction generator (DFIG)based WTSs, which are equipped with partial power rating power converters, PMSG wind turbine configuration shows many advantages [28], such as wider wind speed operating range, higher power energy density and efficiency, better gridside
FIGURE 4.10 Simplified structure of a voltage source backtoback power converter PMSG wind turbine system with active frontend (AFE); MSC, machineside converter; 2L/3L, twolevel/ three level.
116 Control of Power Electronic Converters and Systems
support and fault ridethrough capabilities, and, more importantly, reduced maintenance requirements. These properties make the PMSG WTSs with backtoback power converter and directdrive configuration a more attractive solution in particular for offshore wind energy systems. Fig. 4.10 shows a simplified structure of such a system.
4.4.1 Modeling of PMSG wind turbine system with backtoback power converter In this section, the PMSG WTS is modeled. In Fig. 4.10, the backtoback power converter and PMSG have already been described in Section 4.2, so only the dynamics of the gridside needs to be modeled here. l
Gridside dynamics: grid and filter
Typically, for MW level WTS, an LCL or LC filter is usually used to connect and interface the power converter to the grid. A proper design of an LCL3 or LC filter results in much smaller inductance values to achieve comparable filtering performances as L filter (see Refs. [29,30] and the references therein). However, given a properly designed hardware system, both the LC and LCL filter can be simplified as the same type of an L filter in the fundamental frequency domain for the controller design process. Therefore, in this work, only an (R) L filter is considered and constructed in the laboratory due to its simplicity. The (controllable) gridside power converter (GSC) is also named an active front end (AFE) power converter or (boost) PWM rectifier in many publications in this chapter, it is named as GSC or AFE for consistency. A typical threelevel AFE with RL filter can be described in ab frame. l
AFE with RL filter modeling in ab frame
An AFE with RL filter connected to an ideal (balanced) grid in ab frame is given by (see, e.g. Refs. [31,32]) !ab d i g !ab !ab !0 ab ab ! ! v g ðtÞ ¼ e g ðtÞ þ Rg $ i g ðtÞ þ Lg $ ; i g ð0Þ ¼ i g ˛ ℝ2 . dt Transferring into the discrete format, one obtains Ts Rg !ab TS !ab !ab !ab i g½k þ i g½kþ1 ¼ 1 v g½k e g½k ; Lg Lg {z} ﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄ} :¼b Bg :¼ b Ag
3. For which a damping of the resonant frequency is very important.
(4.19)
(4.20)
117
Model predictive control Chapter  4
where Rg [U] and Lg [Vs/A] are the filter resistance and inductance, T !ab ab respectively, i g ¼ iag ; ibg ½A2 is the current vector to the grid, ! vg ¼ T ab eg ¼ vag ; vbg ½V2 is the output voltage vector of the GSC, and ! T eag ; ebg ½V2 is the grid voltage: all are represented in the ab frame. l
Gridside power dynamics
In the following, the power dynamics are used to model the system. Therefore, the relevant equations of the gridside power dynamics are given. Invoking the instantaneous power theory [33,34], the gridside power can be calculated as T T T ! !ab !ab !ab ab T ! S ¼ ðP; QÞ ¼ eg i g ; eg Jig (4.21) T T T !dq !dg !dq dq ! ¼ eg i g ; eg Jig ;
where P and Q are active and reactive power at the point of common coupling, T T !dq !dq dg dq ! ! d d d respectively. For dt P ¼ e g eg J dtd i g (assuming dt i g and dt Q ¼ T dq d dg ! is constant, i.e., eg ¼ ebg > 0), the dynamics of e g ¼ ebg ; 0 ¼ ebg ; 0 active and reactive power in the dqreference frame can be computed as 9 d 1 g > d > P¼ Rg P þ vg ebg þ ug Lg Q eb2 > > > > dt Lg = d 1 Q¼ Rg Q þ vqg ebg þ ug Lg P : dt Lg
> > > > > > ; (4.22)
For a balanced grid, the gridside source voltage is in the format of ! eg ¼ ab
Aejug t ; where A and ug are the magnitude and frequency, respectively. !ab de ab e g . So the dynamics of the gridside power can be Therefore, dt g ¼ jug !
118 Control of Power Electronic Converters and Systems
obtained in the ab frame in the discrete format invoking Eulerforward method, given as4 !
T dS ! G S½k ¼ ¼ gP½k ; gQ½k dt 1 0 Rg 2 a 30 a b a 1 P þ u Q g ½k C eg½k eg½k eg½k vg½k B Lg ½k C B 16 7B C: A B ¼ 4 [email protected] C B Lg b b b A @ R g a eg½k vg½k eg½k eg½k Q½k ug P½k Lg (4.23) Therefore, the gridside power at k þ 1 can be predicted as ! ! S ½kþ1 ¼ ðP½kþ1 ; Q½kþ1 ÞT ¼ ðP½k ; Q½k ÞT þ Ts $ G S½k
(4.24)
4.4.2 Direct model predictive current control For PMSG WTSs, DMPCC is applied for GSC and MSC. The DMPCC is illustrated in Section 4.2; only DMPCC for GSC is presented in this section. In the analogy to DMPCC for MSC, instead of using the gridside instantaneous power as tracking targets, the gridside current performances are of higher level priority for direct model predictive current control. The inner loop itself can be designed in the ab frame, and the gridside cost function is defined as 2 2 g b a b ! ! ! JDMPC i þ i i . (4.25) u g ¼ ia u g g g½kþ1 g g½kþ1 u g ! ! The predicted current vector iab ˛ U u u g m 8 can be obtained by g½kþ1 Eq. (4.20). The current references are generated/set by a proper outer control loop (here a PI controller for the DClink control is used to generate the daxis current reference and the qaxis current is set to be zero for unity power factor control. These two references are then transferred into ab frame). In the analogy, due to the currents both in a and baxis are equally important to the system, again no extra weightings are required for these targets. After evaluating and minimizing the costs obtained from (4.25) for ! um˛ ! U 8 for the twolevel MSC, an optimal voltage vector of G g will be obtained ! 4. G s ðtÞ can be regarded as a slope/gradient of the power, with the same definition, which can also !T ! q d ! d S ðtÞ d dig dig be applied in the dq, frame, as G s ðtÞ : ¼ dt ¼ ebg dt ; dt .
Model predictive control Chapter  4
119
FIGURE 4.11 Control structure of direct model predictive current control for threelevel NPC backtoback power converter PMSG wind turbine system.
and assigned to the machineside converter. An overview of the predictive current control method for both the grid and machineside control is shown in Fig. 4.11.
4.4.3 FCSMPC for threelevel PMSG systems In this section, the direct FCSMPC control method for 3L backtoback power converter PMSG WTS is evaluated. Both the overall control performances and the steady state of the machine and gridside control performances are given. The simulation data of direct model predictive current control method (for both MSC and GSC) of the 3L backtoback power converter PMSG WTS are depicted in Table 4.3. The simulations of this chapter are based on a perunit model. The control interval is set to be 20 [ms], without compensation applied to the predictions. The DClink voltage reference is set at 5200 [V], while the reactive power reference is set to be 0 [Var]. The overall control performances are shown in Fig. 4.12A. The steadystate control performances of both the generator and gridside currents are illustrated in Fig. 4.12B. As it can be seen, besides its good control dynamics, a big change of the switching frequency is seen. The current tracking performances are also (quite) good. Besides, it is a nonfixed switching frequency control strategy. At the sampling frequency of 50 kHz, its real switching frequency is about 10 kHz. By adding penalty terms
120 Control of Power Electronic Converters and Systems
TABLE 4.3 Parameters of the PMSG system in Fig. 4.11. Parameter
Symbol
Value
Generator stator resistance
Rs
0.05362 [U]
Generator stator inductance
Ls
17.4 [mH]
Permanentmagnet flux linkage
jpm
42 [Wb]
Generator rotor inertia
Qm
40,000 [kg/m2]
Number of pairs
Np
40 [1]
Grid (phase) voltage
eg
1905 [V]
Grid frequency
ug
100p [rad/s]
Gridside resistance
Rg
0 [U]
Gridside inductance
Lg
1 [mH]
DClink capacitance
C1(C2)
8.4 [mF]
DClink voltage
Vdc
5200 [V]
Sampling time
Ts
20 [ms]
Power level
P
5 [MW]
(A)
(B)
FIGURE 4.12 Simulated Control performance of FCSMPC for the threelevel backtoback power converter PMSG wind turbine system shown in Fig. 4.11. From top to bottom are generator speed, torque, generator stator current in dq frame, DClink voltage, gridside current, active and reactive power, and grid and machineside switching frequency, respectively.
Model predictive control Chapter  4
121
to the cost function, the switching frequency can be further reduced at the expense of current control performance. This section has elaborated and verified the application of MPC in wind power generation system. According to the analysis using simulations, it can be seen that MPC has excellent transient and steadystate performance on such systems, and the algorithm is straightforward and easy to implement. It is foreseeable that MPC will be more widely used in wind power generation systems and other renewable systems.
4.5 Conclusions and future trends This chapter has overviewed the application of MPC methods in power electronic systems. This chapter was started with introduction of the basic definitions of the MPC as a control methodology and then with categorization of various methods belonging to the broad family of MPC, which have been applied in the general power electronic area. Afterward, important practical application details have been presented for several selected MPC applications, i.e., in electrical drives, MGs, and wind generation systems. Each of these applications has been discussed in a dedicated section. From the presented results, it can be concluded that MPC offers competitive performance advantages over conventional cascaded linear control strategies that are commonly used for controlling power electronic converters in these applications. In particular, it has been discussed that these advantages arise either from the avoidance of the use of modulator stage or from avoidance of the cascaded structure. While such measures enable better dynamic performance, as well as handling large disturbances and nonlinearities, they also introduce some applicationdependent disadvantages. The most prominent disadvantages are increased computational burden, variable switching frequency (in case of nonmodulated MPC such as FCSMPC), and increased sensitivity to parameter variations due to reliance of the control strategy on the model of the plant. However, recent results have shown that some MPC strategies are very robust to parameter variations despite that fact they rely on the model of the plant. A clear future trend for the MPC in power electronic systems is the development of new MPC methods that operate with constant switching frequencies using a modulator. The target is then to maximize the dynamic performance of power converters, while having the predictable switching pattern at the same time. This facilitates the output filter design and provides low levels of EMI and acoustic noise. Another prominent research area is development of MPC methods that provide stateoftheart performance but reduce the number of sensors. Low number of sensors reduces the cost of power electronic systems and increases their reliability at the same time (i.e., by reducing the number of failure modes). Hence, this area of research is very important from an industrial standpoint. Finally, another very promising future
122 Control of Power Electronic Converters and Systems
research approach is the combination of MPC with databased methods where advantages of both can be combined. There are numerous implementation possibilities in this area, from using databased methods as surrogate models of the parts of the plant that are difficult to model to training the overall databased controller using the data generated by the MPC.
References [1] S. Kouro, M.A. Perez, J. Rodriguez, A.M. Llor, H.A. Young, Model predictive control: Mpc’s role in the evolution of power electronics, IEEE Trans. Ind. Electron. 9 (4) (Dec. 2015) 8e21. [2] R. Kennel, a. Linder, Predictive control of inverter supplied electrical drives, in: 2000 IEEE 31st Annual Power Electron. Specialists Conference. Conference Proceedings (Cat. No.00CH37018), vol. 2, 2000, pp. 761e766. [3] P. Cortes, M. Kazmierkowski, R. Kennel, D. Quevedo, J. Rodriguez, Predictive control in power electronics and drives, IEEE Trans. Ind. Electron. 55 (12) (2008) 4312e4324. [4] R. Kennel, A. Linder, M. Linke, Generalized predictive control (GPC)ready for use in drive applications?, in: Power Electron. Spec. Conf. 2001. PESC. 2001 IEEE 32nd Annu., 4, 2001, pp. 1839e1844, vol. 4. [5] S. Vazquez, J. Rodriguez, M. Rivera, L.G. Franquelo, M. Norambuena, Model predictive control for power converters and drives: advances and trends, IEEE Trans. Ind. Electron. 64 (2) (Feb. 2017) 935e947. [6] T. Geyer, N. Oikonomou, F.D. Kieferndorf, Direct model predictive control: a review of strategies that achieve long prediction intervals for power electronics, IEEE Trans. Ind. Electron. 8 (2014) 32e43. [7] L.M. Caseiro, A.M. Mendes, S.M. Cruz, Dynamically weighted optimal switching vector model predictive control of power converters, IEEE Trans. Ind. Electron. 66 (2) (Feb. 2019) 1235e1245. [8] Z. Zhang, Z. Li, M.P. Kazmierkowski, J. Rodriguez, R. Kennel, Robust predictive control of threelevel NPC backtoback power converter PMSG wind turbine systems with revised predictions, IEEE Trans. Power Electron. 33 (11) (Nov. 2018) 9588e9598. [9] Y. Li, F. Grimm, Z. Zhang, Model predictive control for threephase fourwire voltage source converters in standalone mode, in: 2019 21st European Conference on Power Electronics and Applications, EPE 2019 ECCE Europe, IEEE, Sept. 2019. [10] S. Vazquez, A. Marquez, R. Aguilera, D. Quevedo, J.I. Leon, L.G. Franquelo, Predictive optimal switching sequence direct power control for gridconnected power converters, IEEE Trans. Ind. Electron. 62 (4) (2015) 2010e2020. [11] S. Vazquez, R.P. Aguilera, P. Acuna, J. Pou, J.I. Leon, L.G. Franquelo, V.G. Agelidis, Model predictive control for singlephase NPC converters based on optimal switching sequences, IEEE Trans. Ind. Electron. 63 (12) (Dec. 2016) 7533e7541. [12] A. Mora, R. Ca´rdenasDobson, R.P. Aguilera, A. Angulo, F. Donoso, J. Rodriguez, Computationally efficient cascaded optimal switching sequence MPC for gridconnected threelevel NPC converters, IEEE Trans. Power Electron. 34 (12) (Dec. 2019) 12464e12475. [13] M.G. Judewicz, S.A. Gonzalez, N.I. Echeverria, J.R. Fischer, D.O. Carrica, Generalized predictive current control (GPCC) for gridtie threephase inverters, IEEE Trans. Ind. Electron. 63 (7) (Jul. 2016) 4475e4484.
Model predictive control Chapter  4 [14]
[15] [16]
[17]
[18] [19] [20]
[21]
[22]
[23]
[24]
[25] [26]
[27] [28]
[29]
123
S. Vazquez, C. Montero, C. Bordons, L.G. Franquelo, Design and experimental validation of a Model Predictive Control strategy for a VSI with long prediction horizon, in: IECON Proceedings (Industrial Electronics Conference), 2013, pp. 5788e5793. S. Mariethoz, M. Morari, Explicit modelpredictive control of a PWM inverter with an LCL filter, IEEE Trans. Ind. Electron. 56 (2) (2009) 389e399. S. Alme´r, S. Marie´thoz, M. Morari, Sampled data model predictive control of a voltage source inverter for reduced harmonic distortion, IEEE Trans. Control Syst. Technol. 21 (5) (Sept. 2013) 1907e1915. A. Mora, M. Urrutia, R. Cardenas, A. Angulo, M. Espinoza, M. Diaz, P. Lezana, Modelpredictive controlbased capacitor voltage balancing strategies for modular multilevel converters, IEEE Trans. Ind. Electron. 66 (3) (Mar. 2019) 2432e2443. Y. Zhang, W. Xie, Z. Li, Y. Zhang, Model predictive direct power control of a PWM rectifier with duty cycle optimization, IEEE Trans. Power Electron. 28 (11) (2013) 5343e5351. J. Rocabert, A. Luna, F. Blaabjerg, P. Rodrı´guez, Control of power converters in ac microgrids, IEEE Trans. Power Electron. 27 (11) (Nov. 2012) 4734e4749. F. Blaabjerg, R. Teodorescu, M. Liserre, A. Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (Oct. 2006) 1398e1409. T. Dragicevic, X. Lu, J.C. Vasquez, J.M. Guerrero, DC microgridspart I: a review of control strategies and stabilization techniques, IEEE Trans. Power Electron. 31 (7) (Jul. 2016) 4876e4891. T. Dragicevic, X. Lu, J.C. Vasquez, J.M. Guerrero, DC microgrids  part II: a review of power architectures, applications, and standardization issues, IEEE Trans. Power Electron. 31 (5) (May 2016) 3528e3549. P. Corte´s, G. Ortiz, J.I. Yuz, J. Rodrı´guez, S. Vazquez, L.G. Franquelo, Model predictive control of an inverter with output LC filter for UPS applications, IEEE Trans. Ind. Electron. 56 (6) (Jun. 2009) 1875e1883. C. Zheng, T. Dragicevic, F. Blaabjerg, Currentsensorless finiteset model predictive control for LCfiltered voltage source inverters, IEEE Trans. Power Electron. 35 (1) (Jan. 2020) 1086e1095. T. Dragicevic, Model predictive control of power converters for robust and fast operation of ac microgrids, IEEE Trans. Power Electron. 33 (7) (Jul. 2018) 6304e6317. T. Dragicevic, M. Novak, Weighting factor design in model predictive control of power electronic converters: an artificial neural network approach, IEEE Trans. Ind. Electron. 66 (11) (Nov. 2019) 8870e8880. T. Dragicevic, Dynamic stabilization of DC microgrids with predictive control of pointofload converters, IEEE Trans. Power Electron. 33 (12) (Dec. 2018) 10872e10884. J.L. Li, S.J. Hu, M. Li, Y. Zhu, D.G. Kong, H.H. Xu, Research on the application of parallel backtoback PWM converter on directdrive wind power system, in: 3rd International Conference on Deregulation and Restructuring and Power Technologies, DRPT 2008, 2008, pp. 2504e2508. Z. Zhang, R. Kennel, Novel ripple reduced direct model predictive control of three level NPC active front end with reduced computational effort, in: Predictive Control of Electrical Drives and Power Electronics (PRECEDE 2015), Valparaiso, Chile, 2015.
124 Control of Power Electronic Converters and Systems [30] Z. Zhang, F. Wang, M. Acikgoz, X. Cai, R. Kennel, FGPA HiL simulation of backtoback converter PMSG wind turbine systems, in: Power Electronics and ECCE Asia (ICPEECCE Asia), 2015 9th International Conference on, Jun. 2015, pp. 99e106. [31] Z. Zhang, F. Wang, T. Sun, J. Rodriguez, R. Kennel, FPGA based experimental investigation of a quasicentralized model predictive control for backtoback converters, IEEE Trans. Power Electron. 31 (1) (2016) 662e672. [32] Z. Zhang, R. Kennel, Direct model predictive control of threelevel NPC backtoback power converter PMSG wind turbine systems under unbalanced grid. Predictive Control of Electrical Drives and Power Electronics (PRECEDE 2015), Valparaiso, Chile, 2015, pp. 97e102. [33] H. Akagi, Y. Kanazawa, A. Nabae, Instantaneous reactive power compensators comprising switching devices without energy storage components, IEEE Trans. Ind. Appl. IA20 (3) (1984) 625e630. [34] X. Dai, G. Liu, R. Gretsch, Generalized theory of instantaneous reactive quantity for multiphase power system, IEEE Trans. Power Deliv. 19 (3) (2004) 965e972.
Chapter 5
Adaptive control in power electronic systems Hosein GholamiKhesht, Pooya Davari, Frede Blaabjerg Department of Energy Technology, Aalborg University, Aalborg, Denmark
5.1 Introduction Nowadays, the control systems are known as the main intelligence of the power converter systems, and their major roles are to keep the stability and high performance of the power electronic system. Therefore, so far, many control strategies have been examined and applied in power electronic applications such as using linear control methods (proportional integral), robust control methods (m analysis), nonlinear control methods (sliding mode control), adaptive control methods, fuzzy control methods, digital and predictive control methods, and also identification techniques can be applied in any control in order online to improve the control. Among them, adaptive control methods have attracted much attention for decades. Adaptive control is a control method with adjustable control gains that are changeable due to variable or initially uncertain parameters in the controlled system. Therefore, this control method keeps the system performance and stability at the desired or optimum level under different conditions. In addition to the ability to consider timevariant parameters and the presence of learning parts like nonlinear and smart control methods, another advantage of the adaptive control methods is the existence of powerful mathematical stability and convergence proofs like in classic and linear control methods. Consequently, the adaptive control method is a bridge between the classic and the more intelligent directions of control engineering and includes all advantages of both areas. In this respect, the first aim of this chapter is to describe and investigate the concept, advantages and disadvantages, and new advances in adaptive control methods and also their application in the power electronic systems are discussed. Generally, adaptive control methods can be classified into two categories [1]: indirect adaptive control (IAC) and direct adaptive control (DAC). In IAC methods, firstly, an appropriate online identification is used for system Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000081 Copyright © 2021 Elsevier Ltd. All rights reserved.
125
126 Control of Power Electronic Converters and Systems
parameters’ estimation; afterward, control gains are updated based on the estimated values [2e18]. So far, many different IAC methods are based on various online identification techniques, where the most important are the Luenberger observer [2e6], the sliding mode observer [7], the Kalman filter [8], the neural network [9], the steepest descent [10], recursive leastsquare (RLS) estimator [11], estimator based on Immersion and Invariance (I&I) theory [12], and the Lyapunov stability theory [13]. Although these identification parts make the control system immune against the system parameter variations and uncertainties, it will increase the computational burden in the control system. In contrary to the IAC, the DAC directly updates the controller gains based on a proper adaption mechanism, which brings simplicity and reduces the computational demand [19e26]. Besides, mathematical stability analysis is more straightforward and simple in DAC compared to IAC. One of the most practical and accessible DAC is known as model reference adaptive control (MRAC) [20,21,24e26]. In this control method, the desired characteristics of the closedloop system are expressed in a reference model. There, the reference model generates the desired closedloop response for the controller, and the controller takes the proper action concerning error between real and reference outputs. It is the main difference between MRAC and other control methods when the controller follows the reference output instead of the reference input. To better clarify, the general structure of DAC and IAC methods is shown in Fig. 5.1. Moreover, the block diagram of the MRAC, which is known as one of the essential adaptive control methods, is shown in Fig. 5.2. In the following sections, firstly, the dynamics of an uninterruptible power supply (UPS) system are described in Section 5.2. Section 5.3 presents system identification techniques, which includes both parameter and state estimation methods. The IAC based on the combination of model predictive control (MPC) and an adaptive observer, is presented in Section 5.4. In this structure,
FIGURE 5.1 Block diagram of (A) indirect adaptive control and (B) direct adaptive control to be applied in power electronic system.
Adaptive control in power electronic systems Chapter  5
127
FIGURE 5.2 Block diagram of direct adaptive control based on the model reference adaptive concept.
the adaptive observer improves performance of the conventional MPC by estimating system uncertainties and load current disturbances. In Section 5.5, DAC based on the MRAC techniques is discussed. Both presented IAC and DAC methods in Sections 5.4 and 5.5 are demonstrated on a threephase voltage source inverter (VSI) in UPS applications experimentally. Finally, conclusions are drawn in Section 5.6.
5.2 System dynamics on a UPS system Singleline diagram of the studied threephase UPS is shown in Fig. 5.3, which consists of a threephase VSI connected to loads through an LC filter. Based on Fig. 5.3 and Kirchhoff’s Laws, the system equation can be written as follows: 8 > dif > > ¼ vinv vo
dv o > >C ¼ if io : dt where vinv, vo, are output voltage of the inverter and capacitor voltage/voltage at the point of common coupling (PCC), if and io are the inverter/inductor
FIGURE 5.3 Singlephase diagram of threephase voltage source inverter in uninterruptible power supply application.
128 Control of Power Electronic Converters and Systems
current and load current. L and C are the filter inductance and capacitance to filter the PWM (pulsewidth modulation) harmonics. It is worth to note that the value of parameters L and C are not precisely known, and they could vary from the nominal amounts in practice. Eq. (5.1) can be converted into standard state space form like the following: dx ¼ Ax þ Bvinv þ Dio ; dt 3 2 2 3 2 3 1 " # 1 0 0 7 6 if L 7 6 6 7 7 6 L7 x¼ ;A¼6 7; B ¼ 6 4 5; D ¼ 4 1 5 5 4 vo 1 0 0 C C
(5.2)
where x, vinv , and io are system states, control, and disturbance inputs, and A, B, and D are the state and input matrixes. The digital implementation of the control algorithm can be based on the discrete state space model of the plant dynamics. Discretizing (5.2) with sampling period of TS yields the following discrete state space equations: 8 > xðk þ 1Þ ¼ Ad xðkÞ þ Bd vinv ðkÞ þ Dd io ðkÞ > > > " # > > 1 0 > > > > yðkÞ ¼ Cd xðkÞ; Cd ¼ > > > > 0 1 > > > n o > < Ad ¼ eATS ¼ L1 ðsI AÞ1 (5.3) t¼TS > > Z > TS > > > > B ¼ eAd ðTS sÞ Bds > d > > 0 > > > Z TS > > > >D ¼ > eAd ðTS sÞ Dds : d 0
5.3 System identification The identification part is an essential part of any adaptive control methods (i.e., IAC and DAC), as discussed in the introduction and shown in Fig. 5.1. When this part is appropriately added to any control methods, in order to compensate for uncertain and variable parameters and disturbances inputs, then a strong adaptive control method can be obtained. Due to the importance of the identification part, the most practical identification methods include both parameter estimators and state observers, which are discussed in this section.
Adaptive control in power electronic systems Chapter  5
129
5.3.1 Parameters identification 5.3.1.1 Recursive leastsquare estimation The performance of many control methods depends on the accuracy of the system parameters. Consequently, uncertainty in these parameters causes large output ripples as well as nonzero tracking errors and even system instability. As yet, various identification methods have been proposed for system parameter estimation. Among them, the RLS estimator is known as the best linear unbiased estimator and it is commonly used in adaptive control systems [1]. The general algorithm to implement RLS is given by the following equations: qest ðkÞ ¼ qest ðk 1Þ þ KðkÞðyðkÞ FT ðkÞqest ðk 1ÞÞ KðkÞ ¼ Pðk 1ÞFðkÞðlI þ FT ðkÞPðk 1ÞFðkÞÞ PðkÞ ¼ ðI KðkÞFT ðkÞÞPðk 1Þ=l
1
(5.4)
where qest, FT, and P(k) are the estimated parameter vector, the regressor vector, and the covariance matrix, respectively. Also, l is the forgetting factor and is selected in the range [0e1], small values of l ensure fast convergence with the cost of larger oscillations in the estimated parameters and vice versa. Moreover, once the system parameters approach the real values, P(k) approaches zero, and the RLS does not work correctly and eventually cannot track the upcoming parameter variations. So, periodic resetting of P(k) to aI (where a is a large constant value) would be the right solution and it is well discussed in the adaptive and identification books. To use RLS algorithm and estimate filter inductance and capacitance of the studied UPS system, the system equation can be written as follows: yðkÞ ¼ FT ðkÞq 8 yðkÞ ¼ ½ ðif ðkÞ if ðk 1ÞÞ ðvo ðkÞ vo ðk 1ÞÞ T > > > > " # > < vinv ðk 1Þ vo ðk 1Þ 0 T F ðkÞ ¼ > > 0 if ðk 1Þ io ðk 1Þ > > > : T q ¼ ½a b
(5.5)
This Eq. (5.5) is achieved by discretizing (5.2) with the firstorder Euler method and considering the system parameters as being a ¼ TLS and b ¼ TCS . Consequently, using the system model (5.5) and the RLS algorithm (5.4), the system parameters can be estimated as qest ¼ [aest$best]T. In our system, since the filter parameters change slowly with time, it is desirable to select the value of l very close to one like 0.999. With this choice, a smooth estimation of the parameters is realized.
130 Control of Power Electronic Converters and Systems
5.3.2 State identification In many power electronic applications, it is necessary to measure all system states to shape the closedloop transfer function, minimize the steadystate errors, and to do better disturbance rejection. However, measuring all signals with related sensors increases system cost, volume, and weight. Besides, physical sensors and related communication lines may worsen the system reliability due to additional physical components, which might be subject to damages and thereby failures. A great solution to maintain the system performance and simultaneously to reduce the number of sensors is to employ an observer. In simple words, an observer is a closedloop estimator that uses the state space model of the system to predict the states of the system from the measured inputs and outputs based on the minimization of the difference between the measured and the estimated outputs. High accuracy and reliability, cost and size reduction, and noise immunity are important advantages of determining signals by the observer. Moreover, an observer allows us to predict the desired quantities one sample ahead and therefore helps the control system to compensate for delays introduced by the digital implementation of power converters. There are different types of observers. The most commonly used is a Luenberger observer, which is normally used for deterministic systems, while Kalman filter is developed for stochastic and noisy systems, and finally adaptive observers for robust estimation of system states under strong uncertainties.
5.3.2.1 Observability condition Before implementing any observers, the observability condition of the system must be investigated. The whole system state will be estimated, if and only if the observability matrix of the system (5.6) has a full column rank. T (5.6) Cd Cd Ad . Cd An1 d For the used system in this chapter, two different cases exist: 1. if and io are measured signals and vo is nonmeasured variable, which can be estimated by the observer In such case the observability matrix can be constructed as follows: 1 0 T Cd ¼ ½ 1 0 /obv ¼ ½ Cd Cd Ad ¼ (5.7) Ad ð1; 1Þ Ad ð1; 2Þ 2. vo and io are measured signals and if is the nonmeasured variable, which can be estimated by the observer Therefore, in the second case, the observability matrix is 0 1 T Cd ¼ ½ 0 1 /obv ¼ ½ Cd Cd Ad ¼ (5.8) Ad ð2; 1Þ Ad ð2; 2Þ
Adaptive control in power electronic systems Chapter  5
131
It is evident from the observability matrixes (5.7) and (5.8) that two systems are observable, and employing an observer is possible. It is worth to note that there is also third case when vo and if are measured signals, and it is preferable to eliminate the sensors for the disturbance input (io), which will also be investigated in this section as a disturbance estimator.
5.3.2.2 Luenberger observer A fullorder Luenberger observer can be constructed as follows: xbðk þ 1Þ ¼ Ad xbðkÞ þ Bd vinv ðtÞ þ Dd io ðkÞ þ GCd xðkÞ b x ðkÞ
(5.9)
where symbol “ˆ” denotes the estimated values, and G is the observer gain. The estimation error (e(k)) dynamics using (5.3) and (5.9) is ( eðk þ 1Þ ¼ ðAd GCd ÞeðkÞ (5.10) eðkÞ ¼ xðkÞ xbðkÞ The observer gain is chosen such that the estimation error dynamics of (5.10) is asymptotically stable and the eigenvalues of the observer are placed in desired locations, which is also discussed in Refs. [27e31]. To make the observer dynamically faster than the system, observable poles must be chosen proportional to the system poles (the proportionality constant is k and k < 1). However, a very small proportionality constant causes more sensitivity to the noise. Therefore, selecting the proportionality constant is a tradeoff between the observer dynamics and the noise immunity.
5.3.2.3 Kalman filter Kalman filter, also known as linear quadratic estimation technique, is an optimal recursive estimator, which can estimate system states in the presence of the measurement noises and model uncertainties [8,32e34]. The implementation of the Kalman filter includes two main steps: 1. Prediction step (or time update equations) In this step, based on the dynamic model (physical model) and estimated states xbðkÞ in the previous sample, an initial estimation of the system states is produced. 8 < xbðk þ 1jkÞ ¼ A xbðkÞ þ B uðkÞ þ D i ðkÞ d d d o (5.11) T : Pðk þ 1jkÞ ¼ Ad PðkjkÞAd þ Q where xbðk þ1jkÞ and Pðk þ1jkÞ are predicted states and error covariance, respectively.
132 Control of Power Electronic Converters and Systems
2. Update step (update equations based on measurements) Once the outcome of the next measurement is achieved, these estimated variables are updated using a weighted estimation error between the measured and estimated signals given as 8 1 T T > < Kðk þ 1Þ ¼ Pðk þ 1jkÞC ðCPðk þ 1jkÞC þ RÞ (5.12) xðk þ 1jk þ 1Þ ¼ xðk þ 1jkÞ þ Kðk þ 1Þðy Cxðk þ 1jkÞÞ > : Pðk þ 1jk þ 1Þ ¼ ðI Kðk þ 1ÞCÞPðk þ 1jkÞ where K(kþ1) is the filter gain, xðk þ1jk þ1Þ and Pðk þ1jk þ1Þ are updated and filtered state and error covariance, based on the last measured data. Moreover, Q and R are the covariance of the process noise (model uncertainties) and the measurement noise. Proper estimations of Q and R have a significant impact on the performance of the Kalman filter. However, there is no straightforward algorithm to find appropriate values of them, and they are usually tuned by trial and error using simulations and experiments.
5.3.2.4 Disturbance estimation based on an adaptive observer In practice, the certainty of the system model is subjected to unmodeled dynamics, parameter uncertainties, and external disturbances. In this subsection, to cope with the problem, an augmented discrete state space model is presented, which includes all system parameters uncertainties and unmodeled dynamics [6,7,35]. To do so, by using the nominal system parameters ðAn ; Bn ; Dn Þ and considering the load current as a dynamic disturbance, the system dynamics (5.3) can be rewritten as xðk þ 1Þ ¼ ðAn þ DAÞxðkÞ þ ðBn þ DBÞvinv ðkÞ þ Dd io ðkÞ xðk þ 1Þ ¼ An xðkÞ þ Bn vinv ðkÞ þ MwðkÞ wðkÞ ¼ M 1 Dd io ðkÞ þ M 1 DA xðkÞ þ M 1 DB vinv ðkÞ
(5.13)
where ðDA; DB; DDÞ are the differences between the real and the nominal system matrixes, and w(k) shows new disturbance input, which represents lumped uncertainties and disturbances. Moreover, M is defined as 2 3 TS 0 6 L 7 6 n 7 M¼6 (5.14) 7 4 TS 5 0 Cn Here the parameters with subscript n (Ln, Cn) show the nominal values of system parameters (L, C). The augmented state space model can be obtained if the disturbance input w(k) is known. This subsection presents a simple with low computational burden technique to online estimate the disturbance input
Adaptive control in power electronic systems Chapter  5
133
by using an adaptive observer. Based on the augmented model, a simple adaptive observer can be realized as follows: b xbðk þ 1Þ ¼ An xðkÞ þ Bn vinv ðkÞ þ M wðkÞ
(5.15)
b where xbðk þ1Þ is the output of the adaptive observer, and wðkÞ is the estimated disturbance input. By comparing (5.15) with (5.13), it can be perceived when the estimation error eðkÞ ¼ xðkÞ xbðkÞ converges to zero, the estimated disturbance input approaches the actual disturbance input vector. Therefore, a steepest descent method is included in the observer system to estimate the disturbance input. This method searches for the minimum of a cost function of many variables. An appropriate cost function to formulate the steepest descent algorithm mathematically is the quadratic error function, which is described as follows: EðkÞ ¼ 0:5eðkÞT eðkÞ The steepest descent adaptation law can be expressed as 8 > b þ 1Þ ¼ wðkÞ b b wðk þ D wðkÞ > > > < 0 1 vE > b D wðkÞ ¼ [email protected] A > > > b vw :
Here,
vE ^ vw
(5.16)
(5.17)
is the gradient of the cost function to the disturbance input,
and l is a positive adaptation gain, which controls the rate of convergence and system stability. 8 > > > vE vE ve vb x > > ¼ ¼ Me > > < vw b ve vb b x vw # " (5.18) l1 0 > > > l¼ > > > > 0 l2 : Too large value for l causes an overshoot or in the worst case instability and also too small value gives a low convergence rate and higher computational burden. Therefore, an adaptive gain selection has an essential impact on the adaptive observer stability and efficiency. Therefore, the presented adaptive disturbance observer can be summarized as 8 > > b vinv ðkÞ þ M wðkÞ < xbðk þ 1Þ ¼ An xðkÞ þ Bn (5.19) b þ 1Þ ¼ wðkÞ b wðk þ lM xðkÞ xbðkÞ > > :
134 Control of Power Electronic Converters and Systems
In the following, the stability analysis of the adaptive observer in the sense of the Lyapunov functions is presented. The Lyapunov function is selected as VT ðeðkÞ; kÞ ¼ 0:5eðkÞT eðkÞ
(5.20)
The closedloop stability of the observer is obtained if the following requirements are satisfied:
ð1Þ: VT ðkÞ > 0 (5.21) ð2Þ: DVT ðkÞ < 0 where DVT (k) is the change in the Lyapunov function. The first stability condition in (5.21) is satisfied, because the Lyapunov function in (5.20) is a squaring function. Therefore, Lyapunov’s convergence criterion is completely fulfilled if the negative change in the Lyapunov function is to be verified. The difference in the Lyapunov function is defined as DVT ðkÞ ¼ VT ðeðk þ 1ÞÞ VT ðeðkÞÞ < 0
(5.22)
Eq. (5.22) can be rewritten as DVT ðkÞ ¼ eðkÞT DeðkÞ þ 0:5DeðkÞT DeðkÞ
(5.23)
where De(k) is the change in the estimation error due to adaption law and disturbance input can be given by DeðkÞ ¼ eðk þ 1Þ eðkÞ ¼
veðkÞ veðkÞ vb x ðkÞ b b D wðkÞ ¼ D wðkÞ ¼ MlMeðkÞ b b v wðkÞ vb x ðkÞ v wðkÞ (5.24)
Consequently, DVT (k) by substituting (5.24) in (5.23) can be represented as DVT ðkÞ ¼ eðkÞT ½MlMðI22 0:5MlMÞeðkÞ
(5.25)
To satisfy the second stability condition in (5.21), the adaptation gains are chosen as 8 2Ln 2 > > > 0 < l1 < > > Tsamp2 < (5.26) > 2 > 2C > n > > : 0 < l2 < T 2 samp
Adaptive control in power electronic systems Chapter  5
135
5.4 Indirect adaptive predictive control The performance of many control methods, such as deadbeat and predictive control methods and also classical linear controllers of UPS, depends on the accuracy of the system parameters. Consequently, uncertainty in these parameters may give large output disturbances as well as nonzero tracking errors and even system instability. In this section, for realtime estimation of system uncertainties and disturbances, an identification part is added to the MPC method. So, a combination of the MPC and an identification part realizes an indirect adaptive predictive voltage control, as shown in Fig. 5.4 [6,28]. The proposed control strategy is based on twostep MPC. In this method, in each sampling period and based on the predicted converter quantities and augmented model, optimal converter voltage that minimizes a cost function is selected. The cost function usually consists of the voltage tracking errors (5.27), but others could be included too. Then the optimal voltage vector is saved and applied at the beginning of the next sampling period. So in the twostep MPC, a whole sampling period is available for computations, and consequently, delays due to calculations and digital implementation can be eliminated Fig. 5.4. A conventional choice for cost function of the MPC is: 2 g ¼ vo ðk þ 2Þ vo;ref ðk þ 1Þ (5.27)
FIGURE 5.4 Block diagram of indirect adaptive control method based on predictive controller and adaptive observer.
136 Control of Power Electronic Converters and Systems
where vo ðk þ2Þ is the output voltage at the beginning of the (k þ 2)th period, which is calculated based on the augmented model (5.13) for different available converter output voltage vectors given as vo ðk þ 2Þ ¼ An ð2; 1Þif ðkÞþ An ð2; 2Þvo ðk þ 1Þ þ Bn ð2; 1Þvinv ðk þ 1Þþ Mð2; 2Þw2 ðkþ1Þ (5.28)
In the above equation, the system states at sample (k þ 1) and disturbance dynamic w are estimated via the proposed adaptive observer in (5.19).
5.4.1 Experimental results To evaluate the performance of adaptive MPC, an experimental setup has been prepared, which is shown in Fig. 5.5. The laboratory setup includes threephase 5 kW PWMVSC, which is supplied from a constant DC voltage, threephase resistive load, and LCtype output filter. Moreover, the control method is realized on a DS1007 dSPACE system platform. Finally, for current and voltage measurements and switching pulses generation, the DS2004 highspeed A/D board and the DS5101 digital waveform output board are employed, respectively. The parameters of threephase UPS are given in Table 5.1. The steadystate performance of indirect adaptive MPC is shown in Fig. 5.6, which includes output voltage, output voltage tracking error, output current, and outputs of adaptive disturbance observer. Results show excellent steadystate performance of control method under unmeasured load currents and disturbances, unlike in the conventional MPC, which needs the load currents measurements. Moreover, the transient performance of adaptive MPC under a step change of resistive load is shown in Fig. 5.7. In this figure, the load power has been suddenly changed from zero to nominal ones. This figure shows very good transient response and also disturbance rejection using the indirect adaptive MPC control method.
5.5 Model reference direct adaptive control of UPS In this section, a twoloop adaptive control of UPS is proposed. Both the outer capacitor voltage control and the inner inductor current control loops are based on the MRAC theorem. The block diagram of the proposed MRAC scheme is shown in Fig. 5.8, which includes a reference model that defines the desired closedloop response for both the inner and outer control loops. Moreover, an adaption mechanism to force the system outputs to track the reference outputs and update the control gains in the presence of parameter uncertainties is designed based on the Lyapunov stability theorem. In the following, more details of the proposed control method are discussed.
Adaptive control in power electronic systems Chapter  5
137
FIGURE 5.5 Experimental setup to implement adaptive model predictive control on a threephase voltage source inverter in uninterruptible power supply application.
5.5.1 The outer voltage control loop The capacitor voltage dynamics are given as v_o ¼ a2 ðif io Þ; a2 ¼
1 C
(5.29)
For this equation and the outer control loop, the inductor current is control input and can be defined as if ðtÞ ¼ k1 vo ðtÞ þ k2 vo;ref ðtÞ þ io k1 and k2 are feedback control gains.
(5.30)
138 Control of Power Electronic Converters and Systems
TABLE 5.1 System parameters for uninterruptible power supply system under the adaptive MPC. Nominal power
5 [kW]
Line voltage (rms)
380 [V]
Output frequency (f )
50 [Hz]
Inductor (L)
4.5 [mH]
Capacitor (C)
15 [mF]
DClink voltage (Vdc)
720 [V]
Sampling time
22 [ms]
FIGURE 5.6 Obtained experimental results showing steadystate performance of the adaptive model predictive control under nominal load (5 kW). (A) Output voltage, (B) steadystate error, (C) output current, (D) and (E) adaptive observer outputs.
Adaptive control in power electronic systems Chapter  5
139
FIGURE 5.7 Obtained experimental results showing transient performance of the adaptive model predictive control under step change of the load from zero to 5 kW of the uninterruptible power supply.
FIGURE 5.8 Block diagram of model reference adaptive control of uninterruptible power supply.
140 Control of Power Electronic Converters and Systems
It is worth to note that (5.30) generates the inductor reference current for the inner loop as well as if ;ref ðtÞ x if ðtÞ . The closedloop system under the proposed control law (5.30) is (5.31) v_o ¼ a2 k1 vo ðtÞ þ k2 vo;ref ðtÞ Based on the system state space model of (5.31), the following reference model is proposed: v_ o;m ðtÞ ¼ a2m vo;m ðtÞ þ b2m vo;ref ðtÞ
(5.32)
where vo;m and vo;ref are the output of the reference system and the reference input, and am is positive constant, which assigns the desired closedloop pole in the reference model. The voltage tracking error is evo ðtÞ ¼ vo ðtÞ vo;m ðtÞ
(5.33)
Therefore, the tracking error dynamics can be calculated as evo ðtÞ ¼ a2m evo ða2 k1 a2m Þvo þ ð a2 k2 þ b2m Þvo;ref
(5.34)
The above can be rewritten as
evo ðtÞ ¼ a2m evo a2 k1 k1 vo þ a2 k2 k2 vo;ref
where k1 ; k2 are the nominal values of the control parameters. ( k1 ¼ a1 2 a2m k2 ¼ a1 2 b2m
(5.35)
(5.36)
If the controller parameters are selected from (5.36), then the inputeoutput relation of the real system and the reference model is identical, which is called a perfectmatching of the reference model (5.32) and the closedloop system (5.31). Due to nonlinearities and uncertainties of the plant parameters, (5.36) cannot be directly used for the calculation of the controller parameters. Thus, to finetune the control gains, a proper adaption law must be employed. In the following, a parameter updating law to tune the control gains based on the Lyapunov stability theory is proposed. The voltage tracking error by defining the control parameters errors is recalculated again as
e_vo ðtÞ ¼ a2m evo þ a2 vo Dk1 a2 vo;ref Dk2 (5.37) ek1 ¼ k1 k1 ; ek2 ¼ k2 k2 The differential equation (5.37) contains adjustable parameters, k1 and k2. So, the main goal is to select a proper Lyapunov function and afterward determine an adjustment mechanism, such that the voltage tracking error,
Adaptive control in power electronic systems Chapter  5
141
defined by (5.37), converges to zero. One solution for the positivedefinite radially unbounded normal and conventional Lyapunov function is 1 g1 Vðevo ; kÞ ¼ e2vo þ 2 e2k1 þ e2k2 2 2
(5.38)
For the Lyapunov function to ensure asymptotic stability, the time derivative of (5.38), defined in (5.39), must be negativedefinite. _ vo ; kÞ ¼ e_vo evo þ g1 Vðe 2 ð e_ k1 ek1 þ e_ k2 ek2 Þ
(5.39)
Considering the following relationship: e_ K1 ¼ k_1 ; e_ K2 ¼ k_2
(5.40)
(5.39) is simplified as
_ _ _ vo ; kÞ ¼ a2m e2v þ evo a2 vo;ref ek2 vo ek1 þ g1 Vðe 2 ð k 1 ek1 þ k2 ek2 Þ o 1 _ _ _ vo ; kÞ ¼ a2m e2v þ g1 Vðe 2 k1 a2 evo vo ek1 þ g2 k 2 þ a2 evo vo;ref ek2 o (5.41)
So the first term of (5.41) is negativedefinite if the second and third term at the righthand side equals to zero, i.e., 1 _ 2 _ _ a2 evo vo þ g1 2 k 1 ¼ 0a2 evo vo;ref þ g2 k2 ¼ 0 0 Vðevo ; kÞ ¼ a2m evo (5.42) Therefore, the proposed adaption laws are given as
k_1 ¼ g2 a2 evo vo k_2 ¼ g2 a2 evo vo;ref
(5.43)
Hence, with evo being nonzero, the time derivative of the Lyapunov function is negativedefinite, which translates to V(t) V(0) for t > 0; therefore, evo , and k in (5.38) are bounded. To ensure global exponential stability, the vector T is described as T ¼ ½ e vo
ek1
ek2
(5.44)
with respect to (5.38) lim VðTÞ ¼ N
T/N
(5.45)
Consequently, the closedloop system is globally exponentially stable, and the only equilibrium point of the system has occurred at T ¼ 0. Therefore, (5.38) is a Lyapunov function, and the overall control system under the adaption law of (5.43) is globally exponentially stable. Also, evo ðtÞ/0 and all signals are bounded.
142 Control of Power Electronic Converters and Systems
5.5.2 The inner current control loop The same procedure as done for the outer voltage control loop design can be employed to design the model reference current control. The inductor current dynamics is given as 1 i_f ¼ a1 ðvinv vo Þ; a1 ¼ L
(5.46)
The converter reference voltage (control input) can be selected as vinv ðtÞ ¼ k3 if þ k4 ifref ðtÞ þ vo ðtÞ k3 and k4 are adaption parameters, which can be updated as
k_3 ¼ g1 a1 eif if k_4 ¼ g1 a1 eif if ;ref
(5.47)
(5.48)
where g1 is a positive adaption gain, if ;ref is the reference input, and eif is the current tracking error that is defined as eif ðtÞ ¼ if ðtÞ if ;m ðtÞ
(5.49)
here if ;m is the output of the following reference system: i_f ;m ðtÞ ¼ a1m if ;m ðtÞ þ b1m if ;ref ðtÞ
(5.50)
a1m and b1m are positive constants, and a1m defines the desired closedloop pole.
5.5.3 Experimental results To evaluate the performance of direct MRAC, several experimental tests have been done on the threephase 5 kW UPS. The system parameters are the same as Section 5.4.1 and Table 5.1 and given in Table 5.2. The only difference is TABLE 5.2 System parameters for uninterruptible power supply system under the MRAC. Nominal power
5 [kW]
Line voltage (rms)
380 [V]
Output frequency (f )
50 [Hz]
Inductor (L)
3 [mH]
Capacitor (C)
15 [mF]
DClink voltage (Vdc)
720 [V]
Sampling and switching frequencies
100 [ms]
Adaptive control in power electronic systems Chapter  5
143
the value of the inductor filter. MRAC, unlike MPC, uses a modulator and fixed switching frequency at 10 kHz in this case, and therefore it is possible to implement this control method with a lower inductor filter. Thus, the inductor filter is chosen L ¼ 3 mH. The steadystate performance of MRAC is shown in Fig. 5.9, which includes output voltage, output voltage tracking error, output current, and adaptive control gains. As it can be seen in Fig. 5.9, the control gains converge to desired and constant values in the steady state. The output of the real plant follows the desired output of the reference plant very well. Results confirm
FIGURE 5.9 Obtained experimental results showing steadystate performance of model reference adaptive control under nominal load (5 kW). (A) Output voltage, (B) output current, (C) steadystate error of output voltage, (D) steadystate error of inductor current, (E) control gains (voltage outer loop), and (F) control gains (current inner loop).
144 Control of Power Electronic Converters and Systems
FIGURE 5.10 Obtained experimental results showing the startup performance of model reference adaptive control. (A) Control gains convergence of voltage outer loop and (B) control gains convergence of current inner loop.
good steadystate performance of the control method under unknown system parameters. It is worth to note that these results have been obtained by considering more than 50% error of initial values of the controller gains, which is shown in Fig. 5.10. It shows a smooth and fast convergence of the control gains in the startup process. Moreover, the transient performance of the adaptive MPC under a step change of resistive load (from zero power to nominal ones) is shown in Fig. 5.11, and it demonstrates a good transient response and disturbance rejection when using the MRAC.
FIGURE 5.11 Obtained experimental results showing transient performance of model reference adaptive control under step change of load from zero to 5 kW.
Adaptive control in power electronic systems Chapter  5
145
5.6 Conclusion Power electronic systems are subject to uncertain and timevariant parameters and also disturbances, e.g., due to aging, thermal effects, load changes, etc. Therefore, a fixed and linear control structure may not be able to present and give the desired performance under all conditions. Adaptive control methods have been proposed by power electronic researchers to overcome these issues. These control methods have the capability to work with unknown and uncertain parameters based on a rich mathematical background and using stability analysis. Thus, in this chapter, different types of adaptive control methods, including direct and IAC methods and also identification techniques (which is the essential part of adaptive control system), are presented. Moreover, their application on a threephase UPS system is examined.
References [1] [2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
K.J. Astrom, B. Wittenmark, Adaptive Control, second ed., PrenticeHall, 1994. J.M. Espi, J. Castello, R. GarciaGil, G. Garcera, E. Figueres, An adaptive robust predictive current control for threephase gridconnected inverters, IEEE Trans. Ind. Electron. 58 (8) (August 2011) 3537e3546. C. Xia, M. Wang, Z. Song, T. Liu, Robust model predictive current control of threephase voltage source PWM rectifier with online disturbance observation, IEEE Trans. Ind. Inf. 8 (3) (August 2012) 459e471. K.J. Lee, B.G. Park, R.Y. Kim, D.S. Hyun, Robust predictive current controller based on a disturbance estimator in a threephase gridconnected inverter, IEEE Trans. Power Electron. 27 (1) (2012) 276e283. Y.A.R.I.A.I. Mohamed, M. ARahman, R. Seethapathy, Robust linevoltage sensorless control and synchronization of LCL filtered distributed generation inverters for high power quality grid connection, IEEE Trans. Power Electron. 27 (1) (January 2012) 87e98. H. GholamiKhesht, M. Monfared, Adaptive predictive voltage control of threephase PWMVSCs in UPS applications, in: 2016 International Siberian Conference on Control and Communications (SIBCON), 2016, pp. 1e6. Y.A.R.I. Mohamed, E.F. ElSaadany, Robust high bandwidth discretetime predictive current control with predictive internal model  a unified approach for voltagesource PWM converters, IEEE Trans. Power Electron. 23 (1) (January 2008) 126e136. N. Hoffmann, F.W. Fuchs, Minimal invasive equivalent grid impedance estimation in inductiveeresistive power networks using extended kalman filter, IEEE Trans. Power Electron. 29 (2) (Febuary 2014) 631e641. Y.A.R.I.R.I. Mohamed, E.F. ElSaadany, Adaptive discretetime gridvoltage sensorless interfacing scheme for gridconnected DGinverters based on neuralnetwork identification and deadbeat current regulation, IEEE Trans. Power Electron. 23 (1) (January 2008) 308e321. Y.A.R. Ibrahim Mohamed, et al., An improved deadbeat current control scheme with a novel adaptive selftuning load model for a threephase PWM voltagesource inverter, IEEE Trans. Ind. Electron. 54 (2) (April 2007) 747e759. Q. Liu, K. Hameyer, Highperformance adaptive torque control for an IPMSM with realtime MTPA operation, IEEE Trans. Energy Convers. 32 (2) (June 2017) 571e581.
146 Control of Power Electronic Converters and Systems [12] Y.A. ZunigaVentura, D. LangaricaCordoba, J. LeyvaRamos, L.H. DiazSaldierna, V.M. RamirezRivera, Adaptive backstepping control for a fuel cell/boost converter system, IEEE J. Emerging Sel. Top. Power Electron. 6 (2) (June 2018) 686e695. [13] R.M. Milasi, A.F. Lynch, Y.W. Li, Adaptive vector control for voltage source converters, IET Control Theory Appl. 7 (8) (May 2013) 1110e1119. [14] Z. Chen, J. Qiu, M. Jin, Adaptive finitecontrolset model predictive current control for IPMSM drives with inductance variation, IET Electr. Power Appl. 11 (5) (May 2017) 874e884. [15] J. Liu, W. Wu, H.S.H. Chung, F. Blaabjerg, Disturbance observerbased adaptive current control with selflearning ability to improve the gridinjected current for LCLfiltered gridconnected inverter, IEEE Access 7 (July 2019) 105376e105390. [16] A. Vidal, et al., A method for identification of the equivalent inductance and resistance in the plant model of currentcontrolled gridtied converters, IEEE Trans. Power Electron. 30 (12) (December 2015) 7245e7261. ´ Lo´pez, J. DovalGandoy, A technique to [17] A. Vidal, A.G. Yepes, F.D. Freijedo, J. Malvar, O estimate the equivalent loss resistance of gridtied converters for current control analysis and design, IEEE Trans. Power Electron. 30 (3) (2015) 1747e1761. [18] C. Zheng, T. Dragicevic, F. Blaabjerg, Currentsensorless finiteset model predictive control for LCfiltered voltage source inverters, IEEE Trans. Power Electron. 35 (1) (2019), 1e1. [19] R.M. Milasi, A.F. Lynch, Y.W. Li, Adaptive control of a voltage source converter for power factor correction, IEEE Trans. Power Electron. 28 (10) (2013) 4767e4779. [20] J.R. Massing, M. Stefanello, H.A. Grundling, H. Pinheiro, Adaptive current control for gridconnected converters with LCLfilter, in: 2009 35th Annual Conference of IEEE Industrial Electronics, vol. 59, 2009, pp. 166e172, no. 12. [21] J.R. Massing, et al., Adaptive current control for gridconnected converters with LCL filter, IEEE Trans. Ind. Electron. 59 (12) (2012) 4681e4693. [22] D. Xuan Ba, H. Yeom, J. Kim, J. Bae, Gainadaptive robust backstepping position control of a BLDC motor system, IEEE/ASME Trans. Mechatron. 23 (5) (October 2018) 2470e2481. [23] X. Fu, S. Li, Control of singlephase gridconnected converters with LCL filters using recurrent neural network and conventional control methods, IEEE Trans. Power Electron. 31 (7) (July 2016) 5354e5364. [24] J. Kim, H. ho Choi, J.W. Jung, MRACbased voltage controller for threephase CVCF inverters to attenuate parameter uncertainties under critical load conditions, IEEE Trans. Power Electron. 35 (1) (2019), 1e1. [25] K. Shyu, M. Yang, Y. Chen, Y. Lin, Model reference adaptive control design for a shunt activepowerfilter system, IEEE Trans. Ind. Electron. 55 (1) (January 2008) 97e106. [26] T.D. Do, V.Q. Leu, Y.S. Choi, H.H. Choi, J.W. Jung, An adaptive voltage control strategy of threephase inverter for standalone distributed generation systems, IEEE Trans. Ind. Electron. 60 (12) (2013) 5660e5672. [27] T.S. Kwon, M.H. Shin, D.S. Hyun, Speed sensorless stator fluxoriented control of induction motor in the field weakening region using Luenberger observer, IEEE Trans. Power Electron. 20 (4) (July 2005) 864e869. [28] P. Corte´s, et al., Model predictive control of an inverter with output LC filter for UPS applications, IEEE Trans. Ind. Electron. 56 (6) (June 2009) 1875e1883. [29] J.M. Espi Huerta, J. CastelloMoreno, J.R. Fischer, R. GarciaGil, A synchronous reference frame robust predictive current control for threephase gridconnected inverters, IEEE Trans. Ind. Electron. 57 (3) (March 2010) 954e962.
Adaptive control in power electronic systems Chapter  5 [30]
[31]
[32]
[33] [34]
[35]
147
V. Miskovic, V. Blasko, T.M. Jahns, A.H.C.C. Smith, C. Romenesko, Observerbased active damping of LCL resonance in gridconnected voltage source converters, IEEE Trans. Ind. Appl. 50 (6) (November 2014) 3977e3985. H. GholamiKhesht, M. Monfared, Deadbeat direct power control for grid connected inverters using a fullorder observer, in: 2015 4th International Conference on Electric Power and Energy Conversion Systems, EPECS 2015, 2015, pp. 0e4. F. Auger, M. Hilairet, J.M. Guerrero, E. Monmasson, T. OrlowskaKowalska, S. Katsura, Industrial applications of the kalman filter: a review, IEEE Trans. Ind. Electron. 60 (12) (2013) 5458e5471. R. Guzman, et al., Model based control for a threephase shunt active power filter, IEEE Trans. Ind. Electron. 63 (7) (July 2016) 3998e4007. R. Guzman, L.G. De Vicuna, A. Camacho, J. Miret, J.M. Rey, Recedinghorizon modelpredictive control for a threephase VSI with an LCL filter, IEEE Trans. Ind. Electron. 66 (9) (September 2019) 6671e6680. A.T. Nguyen, M.S. Rafaq, H.H. Choi, J.W. Jung, A model reference adaptive control based speed controller for a surfacemounted permanent magnet synchronous motor drive, IEEE Trans. Ind. Electron. 65 (12) (December 2018) 9399e9409.
This page intentionally left blank
Chapter 6
Machine learning technique for lowfrequency modulation techniques in power converters Amirhossein Moeini1, Morteza Dabbaghjamanesh2, Tomislav Dragicevic3, Jonathan W. Kimball1, Jie Zhang2 1
Electrical and Computer Engineering Department, Missouri University of Science and Technology, Rolla, MO, United States; 2Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, TX, United States; 3Department of Electrical Engineering, Technical University of Denmark, Lyngby, Denmark
6.1 Introduction Nowadays power systems have many new challenges due to the fast growth of electric vehicle charging stations, renewable energy sources, smart grid technologies, and other power electronicebased loads [1e8]. Active power filters (APFs) have been used more and more for compensating the harmonics of nonlinear loads in power systems at the point of common coupling (PCC) [9]. Moreover, both reactive and active power of the power grid at the PCC can be managed by using an APF [9]. Different ACeDC converters have been proposed in the literature for APF applications. Between all available topology, for highpower applications, multilevel converters are growing in popularity due to their low stress on the semiconductor switches, low total harmonic distortion, and modular structure [10,11]. Based on the switching frequency of the converter, the modulation techniques of multilevel converters can be categorized. Highswitching frequency modulation techniques such as space vector modulation (SVM) and phase shiftepulse width modulation (PSPWM) are commonly used due to realtime control on the fundamental and harmonics of the converter and their simple implementation [12]. However, the switching losses of the highswitching frequency modulation techniques are high and undesirable. Moreover, the baseband and sideband harmonics of highswitching frequency modulation techniques are uncontrollable, which are undesirable for the power quality Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000093 Copyright © 2021 Elsevier Ltd. All rights reserved.
149
150 Control of Power Electronic Converters and Systems
requirements of gridtied converters. On the other hand, lowfrequency modulation techniques, e.g., selective harmonic eliminationPWM (SHEPWM) [13,14], selective harmonic mitigationPWM (SHMPWM) [15,16], and selective harmonic current mitigationPWM (SHCMPWM) [9,10,17e19], have low switching losses. Contrarily, implementation of these techniques requires huge memory storage to save all possible solutions. By solving transcendental (trigonometric) Fourier series equations, these offline solutions are often obtained. Different techniques (such as mathematical approaches and linearization techniques) have been proposed in the literature to implement the lowfrequency modulation techniques in real time [20]. Contrarily, they are difficult to be implemented for a highswitching frequency. For example, in Ref. [20], an approach was proposed to linearize the trigonometric equations of a lowfrequency modulation technique to control harmonic phases and magnitudes of a cascaded Hbridge (CHB) converter. However, the number of harmonics that can be controlled by using [20] is limited. Artificial intelligence, especially machine learning, has become popular in different applications such as health science, meteorology, military, and education [21]. Machine learning can be used for regression or classification by using different training techniques [21]. One of the most commonly used machine learning techniques is artificial neural network (ANN) [22e30]. In ANN, the learning behavior of the human brain is modeled by using mathematical equations. The ANN has also been used for the gridtied converter lowfrequency modulation technique in Refs. [22,31], to utilize the DClink voltages of the gridtied converter for controlling the switching angles of the SHEPWM. Contrarily, existing work in the literature have not extensively investigated how to apply ANN for a realtime implementation of the lowfrequency modulation technique with a high number of switching transitions, how to generate training dataset, or how to control both magnitudes and phases of the voltage harmonics of gridtied converters for the APF application. As discussed in Ref. [9], the quarterwave symmetric modulation techniques such as SHEPWM, SHCMPWM, and SHMPWM cannot control both magnitude and phase of the voltage harmonic of the gridtied converters. To solve this issue, when the lowfrequency modulation technique is applied for the gridtied converter, a halfwave symmetric modulation technique, e.g., asymmetric selective harmonic current mitigationPWM (ASHCMPWM) [18] can be used as discussed in Ref. [9]. As a result, when the lowfrequency modulation technique is applied for the gridtied converter, the ASHCMPWM is the only option for controlling the APF current harmonics. The proposed technique has not been investigated so far by using ASHCMPWM technique for the gridtied converter. In this chapter, an asymmetric selective harmonic current mitigationPWM (ASHCMPWM) realtime implementation is investigated by using the ANN technique. As it will be shown in this chapter, the conventional lookup tables can be replaced by a trained ANN for controlling harmonics and the fundamental of the APF. Furthermore, when
151
Machine learning technique Chapter  6
nonlinear loads are connected at the PCC, the current harmonic requirements of the IEEE Std. 519 [32] can be met by using the proposed technique at the PCC. To reach this goal, a 3cell 7level CHB gridtied converter is employed in both simulations and experiments of this chapter. The remaining of this chapter is organized as follows. Section 6.2 briefly discusses the APFs fundamental principles. Section 6.3 explains the proposed ANN technique for the lowfrequency modulation technique. The proposed ANN technique simulation and experimental results are shown and discussed in Section 6.4. Finally, Section 6.5 concludes the work.
6.2 Cascaded Hbridge active power filter configuration Fig. 6.1 shows a CHB converter configuration for the APF application. As shown in Fig. 6.1, the CHB converter N cells are connected to the grid by using the coupling inductance (LF). The nonlinear loads are also connected to the PCC and inject the nonlinear load current (i(NLL)). The converter generates the current (i(acCHB)) as shown in Fig. 6.1 to compensate for the nonlinear load current. By applying KCL, iin ðtÞ ¼ iðacCHBÞ ðtÞ þ iðNLLÞ ðtÞ.
(6.1)
where iin is the injected current to the power grid. Rgrid and Lgrid are the resistance and inductance of the power grid, respectively. In this chapter, the power grid parameters are ignored. v(acGrid)(t), v(acCHB)(t), and v(pcc)(t) are the AC voltages of the grid, CHB, and PCC in Fig. 6.1, respectively. CELL 1 iacCHB
S1
LF
S3
S2
Lgrid
QacCHB
QPcc
Nonlinear loads
Vb1
CELL 2
iNLL S1
QacGrid
S4
S3 C2
S2
+
Vb2
–
Rgrid
+ –
C1
S4
CELL N S1
S3 CN
S2
S4
FIGURE 6.1 Configuration of the CHB for APF application.
+ –
iin
VbN
152 Control of Power Electronic Converters and Systems
6.3 ANN for the asymmetric selective harmonic current mitigationPWM One of the key challenges in lowfrequency modulation techniques is to find realtime solutions of the Fourier equations for different phases and magnitudes of fundamental and harmonics. The proposed ANNbased technique can be used to reach this goal. The CHB voltage Fourier series equations in Fig. 6.2, with halfwave symmetry for the CHB voltage, is XN 8 yðacCHBÞ ðtÞ ¼ ða cosðhutÞ þ bh sinðhutÞÞ > > h¼1 h > > > > 0 1 > > sinðhq11 Þ þ sinðhq12 Þ / sinðhqin1 Þ > > > > ah ¼ 2Vdc @ A > > > hp > þsinðhq Þ / þ sinðhq Þ > iðni þ1Þ 1ð2n1Þ > > < 0 1 (6.2) cosðhq11 Þ cosðhq12 Þ þ / þ cosðhqini Þ > 2V > dc @ > A > > bh ¼ hp > > > Þ Þ cosðhq þ / cosðhq iðn þ1Þ 1ð2n1Þ > i > > > > > > > 0 q11 q12 / > > > : . qini qiðni þ1Þ .q1ð2n1 Þ p where bh and ah are the hth order sine and cosine components of the CHB voltage, respectively; Vdc is the DClink voltage of the converter (Vdc ¼ Vb1 ¼ Vb2 ¼ . ¼ VbN); qjk is the jth cell kth switching transition of the converter. The KVL equation of Fig. 6.1, when the effects of the impedance of the grid and the resistance of LF are ignored, is written as yðacCHBÞ ðtÞ ¼ yðpccÞ ðtÞ þ LF
diðacCHBÞ . dt
(6.3)
FIGURE 6.2 The timedomain waveform of the active power filter generating the voltage v(acCHB)(t).
Machine learning technique Chapter  6
153
TABLE 6.1 Odd order current harmonic requirements of the IEEE Std. 519 [32]. Harmonic order
Current limit (%)
1 < h < 11
4
11 < h < 17
2
17 h < 23
1.5
23 h < 35
0.6
35 h < 50
0.3
TDD
5
The IEEE Std. 519 [32] requirements are then used to impose restrictions on the frequencydomain components of Iin. In ASHCMPWM, where Ch is the hth order current harmonic requirement of IEEE Std. 519 [32] as shown in Table 6.1, the hard equalities of SHEPWM are replaced by inequalities given in (6.4), when IIscL 20. IL is the maximum demand load at the PCC. Isc is the converter short circuit current at the PCC. K is the switching transitions total number of the ASHCMPWM. In order to meet the current harmonic requirements, the ASHCMPWM technique must generate the CHB voltage. In conventional approaches, offline solutions of (6.4) are obtained for a wide range of phases and magnitudes for the harmonics and fundamental, which requires an impractically large memory for embedded implementation. In the present work, an ANN structure is used instead. In order to meet the current harmonic requirements listed in Table 6.1, the injected current harmonic IacCHBh should have equal magnitude as, and 180 degrees phase shift from, INLLh. When the magnitude is changed from 0 to INLLh and the phase is changed from 0 to 360 degrees, a low number of switching angles in the ASHCMPWM technique cannot mitigate the hth nonlinear load current harmonic. Therefore, the maximum circle whole range of nonlinear load current harmonics can be divided into several sections (24 sections in Fig. 6.3). This helps to have all solutions that are close to each other in a same section of Fig. 6.3. This helps the proposed learningbased technique to better divide all solutions (training data) based on the section into which each order of harmonics is placed.
154 Control of Power Electronic Converters and Systems
FIGURE 6.3 The proposed ANN technique for controlling current harmonics of APF.
8 2Vdc > > a1 ¼ ðsinðq11 Þ þ sinðq12 Þ . þ sinðqK ÞÞ; > > > p > > > > 2Vdc > > > ðcosðq11 Þ cosðq12 Þ þ . cosðqK ÞÞ; b1 ¼ > > p > > > > b1 þ ja1 Vð pcc1Þ cos :Vð pcc1Þ þ j sin :Vð pcc1Þ > > > > > > juL > F > > > > ε1 ; > > .þIðNLL1Þ ðcosð:IðNLL1Þ Þþj sinð:IðNLL1Þ ÞÞ > > > > > > > > .Iðin1Þ ðcosð:Iðin1Þ Þþj sinð:Iðin1Þ ÞÞ > > > > > < b3 þ ja3 Vðpcc3Þ cos :Vðpcc3Þ þ j sin :Vð pcc3Þ I > ð j3uLF IL Þ ðin3Þ ¼ C3 ; > > > > IL > > > .þI ðcos q þj sin q Þ NLL3 NLL3 NLL3 > > > > > > > > b5 þ ja5 Vðpcc5Þ cos :Vðpcc5Þ þ j sin :Vðpcc5Þ > > > I ðin5Þ > ð j5uLF IL Þ C5 ; ¼ > > > IL > > > > .þINLL5 ðcos qNLL5 þj sin qNLL5 Þ > > > > > > bh þ jah VðpcchÞ cos :VðpcchÞ þ j sin :VðpcchÞ > > > > IðinhÞ > > I Þ ð jhuL Ch ; F L > ¼ > > IL > > : .þINLLh ðcos qNLLh þj sin qNLLh Þ (6.4)
Machine learning technique Chapter  6
155
In Fig. 6.3, there are 24 sections for different magnitudes and phases of INLLh. Using a high number of sections can improve the ANN performance to meet the power quality standards in the ASHCMPWM technique. Contrarily, this can also significantly increase the size of the dataset. Thus, the proposed technique computational burden is significantly increased. In this technique, the total demand distortion (TDD) is not controlled and just the current harmonics are controlled, given by sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ﬃ Iin3 2 Iin5 2 Iinh 2 TDD ¼ þ þ/þ IL IL IL
(6.5)
However, controlling the individual harmonics will generally achieve the desired TDD as well. In Fig. 6.3, the IacCHBh can be determined as 8 b1 þ ja1 Vðpcc1Þ cos :Vðpcc1Þ þ j sin :Vðpcc1Þ > > IðacCHB1Þ ¼ ; > > ðjuLF Þ > > > > > > b3 þ ja3 Vðpcc3Þ cos :Vðpcc3Þ þ j sin :Vðpcc3Þ > > > IðacCHB3Þ ¼ ; > > ðj3uLF Þ < b5 þ ja5 Vðpcc5Þ cos :Vðpcc5Þ þ j sin :Vðpcc5Þ > > > IðacCHB5Þ ¼ > > ðj5uLF Þ > > > > . > > > > > bh þ jah VðpcchÞ cos :VðpcchÞ þ j sin :VðpcchÞ > :I ðacCHBhÞ ¼ ðjhuLF Þ (6.6) The solutions of (6.6) are solved for various phases and magnitudes of VPCCh. By checking the phase and magnitude of (6.6), one of the sectors in Fig. 6.3 is selected. In this paper, it is assumed that the PCC voltage harmonic magnitudes are close to zero. Contrarily, the PCC voltage harmonics can be obtained from sLF VðacGridÞ ðsÞ VðPCCÞ ðsÞ ¼
þZGrid ðsÞðVðacCHBÞ ðsÞ þ sLF INLL ðsÞÞ ; ðsLF þ ZGrid ðsÞÞ
(6.7)
where ZGrid(s) is the grid impedance. From (6.7), a high value of ZGrid(s) increases the V(acCHB) effect on the VPCC. However, a low value of ZGrid(s) increases the effect of V(acGrid)(s) on the VPCC(s). From (6.6) and (6.7), I(acCHB) is a function of V(acCHB) and VPCC. Furthermore, VPCC is also function of V(acCHB). Thus, changing the V(acCHB) affects both VPCC
156 Control of Power Electronic Converters and Systems
FIGURE 6.4 The ANN block diagram to be applied in the active power filter.
and I(acCHB). This can complicate the controller design when the grid has an impedance, and is beyond the scope of the present work. Eq. (6.6) data for various switching angles of the CHB converter are obtained. A general block diagram of the ANN [21] is shown in Fig. 6.4. As it can be seen, three main layers are used for the ANN technique: an input layer (X), an output layer (Y), and two hidden layers (H). Several nodes are used as shown in Fig. 6.4. Among consecutive layers (i (previous layer) and j (current layer)), a line which has a weight (uij) is used. Each node output can be calculated using ! nl1 X l l l1 l Oi uij þ bj O j ¼ sj (6.8) i¼0
where
Olj
is the output of the lth layer jth node, Ojl1 is the output of the (l 1)
th layer jth node, Olj is the jth node activation function in the lth layer, blj is the bias of the jth node in the lth layer of the ANN, and nl1 is the number of nodes in the (l 1)th layer. In the proposed method, values of q11, q12, ., and qK that satisfy (6.4) are determined for a PCC voltage harmonic magnitudes and phases range. Given the large number of variables and the range of inputs, the search space is very large. Thus, the technique given in Algorithm 6.1 is used to randomly sample the search space.
Machine learning technique Chapter  6
157
Algorithm 6.1 is used to generate random training data. First, based on the accuracy that is required for the harmonics and the available computational power, a time step is chosen. Switching angles that cover the whole range of search space are produced using random numbers. For the present work, the time step is set to 8 degrees; a smaller time step corresponds to an increase in the dataset size, an increase in the time required for training, and an increase in accuracy. The proposed algorithm first sorts the angles produced. Next, the angles are checked to all lie within the range of [0 degree; 180 degrees] to enforce halfwave symmetry. By using (6.4) and the obtained switching angles, current harmonics are checked to determine whether they meet the IEEE Std. 519 limits [32]. Variables O3, O5, ., Oh, OTDD are logical indicators of whether the results satisfy the given limits; additional constraints can be included for other power quality standards. Finally, O1 assigns the fundamental voltage (b1).
6.4 The proposed technique simulation and experimental results To validate the advantages and effectiveness of the proposed ASHCMPWM with the ANN technique, simulation and experimental results are obtained for a 3cell 7level CHB converter. The parameters (i.e., the circuit parameters and the number of hidden layers of the ANN) of the gridtied converter during the simulations and experiments are shown in Tables 6.2 and 6.3. The opensource KERAS [33] software is used to train the ANN, which is written in Python. The main objective in both simulations and experiments is to prove that the ANN technique can control harmonic phases and magnitudes by using the ASHCMPWM technique.
6.4.1 Simulation results In addition to the parameters that are mentioned in Table 6.2, a diode bridge connected to a parallel combination of a 20 resistor and a 50 mF capacitor is
TABLE 6.2 The number of nodes in ANN hidden layers. Technique
1st hid.
2nd hid.
3rd hid.
4th hid.
ANN
50
50
50
50
158 Control of Power Electronic Converters and Systems
TABLE 6.3 Multilevel converter parameters in simulations and experiments. Parameter
Symbol
Value
AC grid voltage
VacGrid1
110 V
Fundamental frequency
F
60 Hz
Coupling inductance
LF
0.49 pu
DClink voltage
Vdc
65 V
Maximum output demand current
IL
20 A
Number of cells in CHB
i
3
Number of switching transitions in each cell
n
1
Number of switching transitions
K
3
Number of mitigated harmonics
h
49
Decoupling DC capacitance
C
600 mF
used as a nonlinear load in the first simulation. Grid impedance is neglected here so the rectifier does not affect the PCC voltage. MATLAB/Simulink1 is used to simulate the nonlinear load combination and the CHB gridtied converter. Fig. 6.5A illustrates the timedomain waveforms of iin(t), v(acCHB)(t), v(acGrid)(t), and iNLL(t). The ANN technique increases the iin(t) fundamental current from 7:77 to 19:2 A at the PCC. Fig. 6.5B shows the current harmonic spectra of the nonlinear load (iNLL(t)), which cannot meet the current harmonic requirements (the red (gray in print version) line in Fig. 6.5B) of the IEEE Std. 519 [32] for the third and fifth harmonics. Moreover, the nonlinear load current TDD cannot meet the 5% limit of the IEEE Std. 519. Fig. 6.5C shows the current harmonic spectra of iin(t) at the PCC, when the nonlinear load (iNLL(t)) is injected to the grid. Now, the harmonics do meet the requirements due to the compensation provided by the CHB. The ANN achieves harmonic control with only six switching transitions per halfperiod.
1. MATLAB and Simulink are registered trademarks of The MathWorks, Inc.
Machine learning technique Chapter  6
159
(A)
(B)
Fundamental (60Hz) = 7.77 , TDD= 7.18%
8 7
Mag (% of IL)
6 5 4 3 2 1 0 0
Mag (% of IL)
(C)
10
20
30
Harmonic order
40
50
Fundamental (60Hz) = 19.2 , TDD= 4.54%
8
6
4
2
0 0
10
20
30
Harmonic order
40
50
FIGURE 6.5 Simulation results using ASHCMPWM technique for the gridtied CHB converter with a nonlinear load, (A) timedomain waveforms of the v(acCHB)(t), v(acGrid)(t), iin(t), and i(NLL)(t); (B) current harmonic spectra of i( NLL)(t); (C) current harmonic spectra of iin(t).
160 Control of Power Electronic Converters and Systems Algorithm 6.1. (The proposed ASHCMPWM algorithm with ANN.)
Machine learning technique Chapter  6
161
Fig. 6.6 shows the ANN technique second simulation result when the modulation index of the converter with the ASHCMPWM is 2.45. The second simulation result objective is to prove that the ANN can control the current harmonic for the ASHCMPWM without injecting the nonlinear load to the grid. After training the ANN technique, the proposed technique switching angles are 3, 26, 48, 121, 147, and 165 degrees. The gridtied converter current magnitude is 12:9 A with a phase of 104 degrees. Thus, the proposed technique can be applied to the gridtied converter for any reactive and active power. Fig. 6.6A shows the timedomain waveforms of iin(t), vacCHB(t), and
VacCHB(t), VacGrid(t), Iin(t)
(A)
200
VacCHB(50V/div) VacGrid(50V/div) Iin(10A/div)
100
0
100
200 0.08
0.09
0.1
0.11
0.12
0.13
0.14
Time(s)
(B)
Fundamental (60Hz) = 12.90 , THD= 3.59%
5
Mag (% of IL)
4 3 2 1 0 5
10
15
20
25
30
Harmonic order
35
40
45
50
FIGURE 6.6 ASHCMPWM technique simulation results for the gridtied CHB converter without the nonlinear load, (A) timedomain waveforms of v(acCHB)(t), v(acGrid)(t), and iin(t), (B) current harmonic spectra of i(in)(t).
162 Control of Power Electronic Converters and Systems
vacGrid(t). In this figure, a sinusoidal waveform is generated for the AC current (iin(t)) by using the lowfrequency modulation technique. Moreover, there is no variation in vacCHB(t) timedomain waveform due to neglecting the parasitic resistance at the DC link. On the contrary, the proposed technique can meet the power quality standard, when there is a parasitic resistance at the DC link of the CHB converter (internal resistance of the battery) as proven in Ref. [34]. The AC current timedomain waveform ((iin(t))) is shown in Fig. 6.6B with the requirements of the IEEE Std. 519 indicated. From this figure, the proposed ANN can control all 25 odd loworder current harmonics. Also, as shown in Fig. 6.6B, the TDD 5% limit in IEEE Std. 519 can be met by using the ANN.
6.4.2 Experimental results An ANN technique is further validated with experimental results using the same parameters, a 3cell 7level CHB converter. The second simulation in Fig. 6.6 is experimentally repeated using the same parameters. The gridtied CHB converter hardware prototype that is used in experiments is shown in Fig. 6.7. Texas Instruments TMS320F28335 is used for applying the switching angles to the CHB converter. In each Hbridge of the CHB converter, an intelligent power module (IPM) (rated 30 A and 600 V) that uses a 3leg IGBT is used as shown in Fig. 6.7. The block diagram in Fig. 6.8 illustrates the proposed technique implementation during the experiments, which is an openloop control that applies the ASHCMPWM switching angles technique with the ANN. The grid impedance is small and may be ignored as shown in Fig. 6.8. A phaselocked loop is used in this figure to detect the phase and frequency of the grid voltage. Moreover, qCHB is the CHB converter initial phase. In both simulation and experimental results,
FIGURE 6.7 3cell gridtied CHB converter hardware prototype.
Machine learning technique Chapter  6
163
FIGURE 6.8 Gridtied converter block diagram during the experiments and simulations.
the qCHB is assumed to be 0 degree. The optimal solutions (switching angles) (q11, q21, q31, q32, q22, q12) in Fig. 6.8 from the ANN are used in the gridtied CHB converter. The switching transition values that are employed in the experiment are the same as in the second simulation result shown in Fig. 6.6. SW1; SW2; and; SW3 are the converter switchings for the first, second, and third Hbridge cells, respectively. The converter has the lowest possible switching frequency, equal to the line frequency, 60 Hz. As a result, the proposed technique has the lowest possible switching losses for hard switching ACeDC converters. The experimental results of the proposed ANN technique for ASHCMPWM are shown in Fig. 6.9, when the CHB converter modulation index is 2.45, the same as the simulation result in Fig. 6.6. The gridtied converter AC current is 13:4 A with a phase of 102. The vacCHB(t), vacGrid(t), and iin(t) timedomain waveforms are shown in Fig. 6.9A. The AC input current (iin(t)) has a pure sinusoidal current waveform similar to the simulation result. The experimental current harmonic spectrum is shown in Fig. 6.9B. The IEEE Std. 519 [32] current requirements are shown by the red (gray in print version) line in Fig. 6.9B. As illustrated in the experimental results, all loworder current harmonics meet the IEEE Std. 519 requirements at the PCC. Moreover, using the proposed ANN technique for the ASHCMPWM in the experiment in Fig. 6.9B, the 5% limit specified in the IEEE Std. 519 for the TDD is met.
164 Control of Power Electronic Converters and Systems
(A)
(B)
Fundamental (60Hz) = 13.4 , TDD= 4.24%
5
Mag (% of IL)
4 3 2 1 0 5
10
15
20
25
30
Harmonic order
35
40
45
50
FIGURE 6.9 ASHCMPWM technique experimental results for the gridtied CHB converter, (A) timedomain waveforms of v(acCHB)(t), v(acGrid)(t), and iin(t), (B) current harmonic spectra of i(in)(t).
6.5 Conclusion In this chapter, an ANN technique to implement asymmetric selective harmonic current mitigationPWM was proposed to control the current harmonics at the PCC in order to meet the power quality standards. The proposed technique does not need to save all phases and magnitudes of the fundamental and harmonics of the APF application AC current. The lowfrequency ASHCMPWM technique was implemented in both simulations and experiments to prove the advantages of the proposed technique for controlling the APF
Machine learning technique Chapter  6
165
current harmonic. To reach this goal, a technique was proposed to categorize the voltage harmonic vectors of the gridtied converter. As demonstrated in the simulation and experimental results, the proposed technique that uses an ANN could meet the power quality standard loworder current harmonics such as the IEEE Std. 519. Furthermore, in this chapter, a guideline for generating the ANN technique training data was proposed. Using the guidelines in this chapter, the ANN could completely search the solutions whole search space of the lowfrequency modulation techniques.
References [1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
M. Dabbaghjamanesh, A. KavousiFard, Z. Dong, A novel distributed cloudfog based framework for energy management of networked microgrids, IEEE Trans. Power Syst. 35 (4) (2020) 2847e2862. M. Dabbaghjamanesh, A. KavousiFard, S. Mehraeen, J. Zhang, Z.Y. Dong, Sensitivity analysis of renewable energy integration on stochastic energy management of automated reconfigurable hybrid acedc microgrid considering DLR security constraint, IEEE Trans. Industr. Inform. 16 (1) (January 2020) 120e131. M. Dabbaghjamanesh, B. Wang, A. KavousiFard, S. Mehraeen, N.D. Hatziargyriou, D. Trakas, F. Ferdowsi, A novel twostage multilayer constrained spectral clustering strategy for intentional islanding of power grids, IEEE Trans. Power Deliv. 35 (2) (2019) 560e570. M. Dabbaghjamanesh, A. KavousiFard, S. Mehraeen, Effective scheduling of reconfigurable microgrids with dynamic thermal line rating, IEEE Trans. Ind. Electron. 66 (2) (February 2019) 1552e1564. E. Taherzadeh, M. Dabbaghjamanesh, M. Gitizadeh, A. Rahideh, A new efficient fuel optimization in blended charge depletion/charge sustenance control strategy for plugin hybrid electric vehicles, IEEE Trans. Intell. Veh. 3 (3) (September 2018) 374e383. M. Dabbaghjamanesh, B. Wang, S. Mehraeen, J. Zhang, A. KavousiFard, Networked microgrid security and privacy enhancement by the blockchainenabled internet of things approach, in: 2019 IEEE Green Technologies Conference (GreenTech), April 2019, pp. 1e5. A. Kavousi Fard, B. Wang, O. Avatefipour, M. Dabbaghjamanesh, R. Sahba, A. Sahba, Superconducting fault current limiter allocation in reconfigurable smart grids, in: 2019 3rd International Conference on Smart Grid and Smart Cities (ICSGSC), June 2019, pp. 76e80. M. Dabbaghjamanesh, S. Mehraeen, A. KavousiFard, F. Ferdowsi, A new efficient stochastic energy management technique for interconnected ac microgrids, in: 2018 IEEE Power Energy Society General Meeting (PESGM), August 2018, pp. 1e5. A. Moeini, S. Wang, An asymmetric selective harmonic current and voltage modulationPWM technique for electric vehicle charging stations with cascaded hbridge converters to meet power quality standards, in: 2019 IEEE Energy Conversion Congress and Exposition (ECCE), September 2019, pp. 4395e4402. A. Moeini, H. Zhao, S. Wang, A currentreferencebased selective harmonic current mitigation PWM technique to improve the performance of cascaded hbridge multilevel active rectifiers, IEEE Trans. Ind. Electron. 65 (1) (2016) 727e737.
166 Control of Power Electronic Converters and Systems [11] A. Moeini, H. ImanEini, M. Bakhshizadeh, Selective harmonic mitigationpulsewidth modulation technique with variable dclink voltages in single and threephase cascaded hbridge inverters, IET Power Electron. 7 (4) (2013) 924e932. [12] M.S.A. Dahidah, G. Konstantinou, V.G. Agelidis, A review of multilevel selective harmonic elimination PWM: formulations, solving algorithms, implementation and applications, IEEE Trans. Power Electron. 30 (8) (August 2015) 4091e4106. [13] S.R. Pulikanti, V.G. Agelidis, Hybrid flyingcapacitorbased activeneutralpointclamped fivelevel converter operated with shePWM, IEEE Trans. Ind. Electron. 58 (10) (October 2011) 4643e4653. [14] H. Zhao, S. Wang, A. Moeini, Critical parameter design for a cascaded hbridge with selective harmonic elimination/compensation based on harmonic envelope analysis for singlephase systems, IEEE Trans. Ind. Electron. 66 (4) (April 2019) 2914e2925. [15] L.G. Franquelo, J. Napoles, R.C.P. Guisado, J.I. Leon, M.A. Aguirre, A flexible selective harmonic mitigation technique to meet grid codes in threelevel PWM converters, IEEE Trans. Ind. Electron. 54 (6) (December 2007) 3022e3029. [16] M. Najjar, A. Moeini, M.K. Bakhshizadeh, F. Blaabjerg, S. Farhangi, Optimal selective harmonic mitigation technique on variable dc link cascaded hbridge converter to meet power quality standards, IEEE J. Emerging Sel. Top. Power Electron. 4 (3) (September 2016) 1107e1116. [17] A. Moeini, H. Zhao, S. Wang, Improve control to output dynamic response and extend modulation index range with hybrid selective harmonic current mitigationPWM and phaseshift PWM for fourquadrant cascaded hbridge converters, IEEE Trans. Ind. Electron. 64 (9) (2017) 6854e6863. [18] A. Moeini, S. Wang, B. Zhang, L. Yang, A hybrid phase shiftPWM and asymmetric selective harmonic current mitigationPWM modulation technique to reduce harmonics and inductance of singlephase gridtied cascaded multilevel converters, IEEE Trans. Ind. Electron. (2019), 1e1. [19] A. Moeini, S. Wang, Analyzing and reducing current harmonics of ac and dc sides of cascaded hbridge converters for electric vehicle charging stations, in: 2019 IEEE Energy Conversion Congress and Exposition (ECCE), September 2019, pp. 193e200. [20] H. Zhao, T. Jin, S. Wang, L. Sun, A realtime selective harmonic elimination based on a transientfree inner closedloop control for cascaded multilevel inverters, IEEE Trans. Power Electron. 31 (2) (February 2016) 1000e1014. [21] J.M. Zurada, Introduction to Artificial Neural Systems, vol. 8, West Publishing Company St. Paul, 1992. [22] F. Filho, L.M. Tolbert, Y. Cao, B. Ozpineci, Realtime selective harmonic minimization for multilevel inverters connected to solar panels using artificial neural network angle generation, IEEE Trans. Ind. Appl. 47 (5) (September 2011) 2117e2124. [23] M. Balasubramonian, V. Rajamani, Design and realtime implementation of shepwm in singlephase inverter using generalized hopfield neural network, IEEE Trans. Ind. Electron. 61 (11) (November 2014) 6327e6336. [24] K. Yang, J. Hao, Y. Wang, Switching angles generation for selective harmonic elimination by using artificial neural networks and quasinewton algorithm, in: 2016 IEEE Energy Conversion Congress and Exposition (ECCE), September 2016, pp. 1e5.
Machine learning technique Chapter  6 [25]
[26]
[27]
[28]
[29]
[30]
[31]
[32] [33] [34]
167
P. Farhadi, M. Navidi, M. Gheydi, M. Pazhoohesh, H. Bevrani, Online selective harmonic minimization for cascaded halfbridge multilevel inverter using artificial neural network, in: 2015 Intl Aegean Conference on Electrical Machines Power Electronics (ACEMP), 2015 Intl Conference on Optimization of Electrical Electronic Equipment (OPTIM) 2015 Intl Symposium on Advanced Electromechanical Motion Systems (ELECTROMOTION), September 2015, pp. 331e335. P. Shanmuga Aravind, S. Albert Alexander, Harmonic minimization of a solar fed cascaded H bridge inverter using artificial neural network, in: 2013 International Conference on Energy Efficient Technologies for Sustainability, April 2013, pp. 163e167. I. Khoukha, C. Hachemi, B. El Madjid, Ann control of nine level NPC voltage inverter based on selective harmonics elimination, in: 2007 International Aegean Conference on Electrical Machines and Power Electronics, September 2007, pp. 587e591. M.G. Fakhry, A. Massoud, S. Ahmed, Quasi sevenlevel operation of multilevel converters with selective harmonic elimination, in: 2014 26th International Conference on Microelectronics (ICM), December 2014, pp. 216e219. J. Hao, G. Zhang, Y. Zheng, W. Hu, K. Yang, Solution for selective harmonic elimination in asymmetric multilevel inverter based on stochastic configuration network and levenbergmarquardt algorithm, in: 2019 IEEE Applied Power Electronics Conference and Exposition (APEC), March 2019, pp. 2855e2858. M. Mohaddes, A.M. Gole, P.G. McLaren, Application of an artificial neural network to harmonic reduction in PWM acdc converter, in: IEEE WESCANEX 95. Communications, Power, and Computing. Conference Proceedings, vol. 1, May 1995, pp. 123e128. F. Filho, H.Z. Maia, T.H.A. Mateus, B. Ozpineci, L.M. Tolbert, J.O.P. Pinto, Adaptive selective harmonic minimization based on ANNs for cascade multilevel inverters with varying dc sources, IEEE Trans. Ind. Electron. 60 (5) (May 2013) 1955e1962. Ieee Recommended Practice and Requirements for Harmonic Control in Electric Power Systems, IEEE Std 5192014, June 2014, pp. 1e29 (Revision of IEEE Std 5191992). F. Chollet, et al., Keras, 2015. https://keras.io. A. Moeini, S. Wang, A dc link sensorless voltage balancing technique for cascaded hbridge multilevel converters with asymmetric selective harmonic current mitigationPWM, IEEE Trans. Power Electron. 33 (9) (2017) 7571e7581.
This page intentionally left blank
Chapter 7
Overview of stability analysis methods in power electronics Qianwen Xu Electric Power and Energy Systems Division, KTH Royal Institute of Technology, Stockholm, Sweden
7.1 Introduction Power electronic converters are widely adopted as the interface for the integration of various distributed generation systems (e.g., fuel cell, photovoltaic (PV), wind turbine, and energy storage) into the power grid [1], resulting in the emergence of power electronicebased power systems, like PV power plants [2], wind power plants [3], microgrids [4], electric railway systems [5], highvoltage DC transmission (HVDC), flexible AC transmission [6], etc. These interface converters enable full controllability, sustainability, and improved efficiency of the systems [7]. However, due to their negligible physical inertia, they also make the system susceptible to oscillations resulting from the network disturbances and the interactions with the systems [8,9]. Power system stability has been a significant issue since the 19th century. In conventional power system, stability analysis is well established: there are standard models of synchronous machines, governors, and excitation systems of varying orders to capture the important modes for particular problems; the electromechanical stability of synchronous generators, which is normally caused by slow control and dynamics that below the fundamental frequency, is the major concern [10]. However, in the modern power electronicebased power systems, the wide timescale control dynamics of power converters can cause interactions with both electromechanical dynamics in electrical machines and electromagnetic transients in power networks, resulting in wide timescale stability issues, e.g., lowfrequency oscillations associated with the outer power control and grid synchronization loops, and highfrequency oscillations (typically from hundreds of hertz to several kilohertz) driven by mutual interactions between the fast inner control loops of the gridconnected converters [9]. Another significant concern is the high complexity of the power electronicebased power system caused by the large number of power Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000184 Copyright © 2021 Elsevier Ltd. All rights reserved.
169
170 Control of Power Electronic Converters and Systems
electronic converters with dedicated control loops, which complicates the system level modeling and analysis, and also increases difficulties in locating the instabilities present in power electronicebased power system [11,12]. Therefore, it is of significant importance to perform accurate stability analysis for design and operation of power electronicebased power systems. Many research works have been conducted for power system smallsignal stability analysis, either using eigenvalue method or impedancebased method. In eigenvalue method, a system state space model is derived and linearized at the operating point, then system stability is assessed by examining eigenvalues of the system state space matrix. The advantages of the eigenvalue method are the identification of oscillation modes and instability roots. The eigenvalue method has been widely applied for stability analysis of wind farms [3], inverterbased microgrid [4], gridconnected PV systems [13], HVDC system [6], etc. For a largescale power system with high penetration of power electronic converters, there would be a lot of state variables from the detailed models of power converter dynamics, sources, lads, cables, etc. To make the eigenvalue method with modularity and scalability, some integration techniques like component connection method (CCM) or modulebased approach are adopted for the system level integration [14,15]. The drawback of the eigenvalue approach is that it needs to know the full system information which may be kept privacy by different vendors of components to the power system [8]. In impedancebased approach, a system is divided as a source equivalent and a load equivalent, and system stability can be determined by the impedance ratio of the load and source using Nyquist criterion [8,16]. The advantages of impedancebased method are its modularity and its black box feature, where detailed knowledge of the parameters and properties of the system is not required as long as measurements can be obtained. The impedance can be extracted based on the measured signals with frequency scanning [17]. This also provides a promising technique for the online stability assessment. The impedancebased method is widely used in cascaded converter systems and gridconnected converter systems [18], gridconnected converter systems [19], and wind power systems [20]. The drawbacks of the impedancebased method include the conservative results and difficulty in identifying the oscillation modes and instability roots. Smallsignal stability can only ensure system stability at small disturbance. Due to the nonlinearity and complexity of the power electronicebased power system, it is significant to use largesignal analysis methods to ensure system stability at large disturbances, which may be caused by faults, protection, sudden load connections or disconnections, etc [21,22]. The stability behavior under such conditions differs from the smallsignal case. Largesignal analysis provides characterization of stability boundary and transient behavior of system following a large disturbance. The objective is to estimate the domain of attraction of the system operation point, i.e., how large deviations from the operating point can be tolerated by the system. Currently, there are two kinds
Overview of stability analysis methods in power electronics Chapter  7
171
of largesignal analysis methods, one is timedomainebased simulation method and the other is Lyapunovbased analytical methods. Timedomain simulations can provide high accuracy and validity but numerous simulations should be conducted over a wide operating point under various conditions [23]. There are several Lyapunovbased analytical methods, including TakagieSugeno (TS) multimodeling approach [24], BraytoneMoser’s Mixed Potential method [25], and genetic algorithmebased Lyapunov function searching method [26]. Lyapunovbased analytical methods can provide the estimation of domain of attractions for stability, but it is a significant challenge to find a proper Lyapunov function. In this chapter, the stability analysis methods for power electronicebased power systems are introduced, which is summarized in Fig. 7.1. Then Section 7.2 presents smallsignal stability analysis methods and Section 7.3 presents largesignal stability analysis tools. Then two case studies are presented in Section 7.4 with smallsignal stability analysis and largesignal stability analysis. Finally, conclusion is drawn.
7.2 Smallsignal stability analysis methods Smallsignal stability is the system stability when subject to smallsignal disturbances. The wide timescale oscillation issues caused by the interactions of power electronic converters with each other and with the grid are smallsignal stability issues [10]. In this section, the modeling approaches of power converters are introduced. Then two widely used smallsignal analysis methods, i.e., the eigenvalue method and impedancebased method, are introduced to perform smallsignal stability analysis.
FIGURE 7.1 Stability analysis for power electronicebased power systems.
172 Control of Power Electronic Converters and Systems
7.2.1 Modeling of power converter Before stability analysis, it is necessary to build linearized models of power electronic converters due to their switching dynamics and nonlinearities. Several approaches can be adopted considering different level of accuracy requirement and the available information, i.e., state space averaging (SSA) approach, generalized SSA model method, harmonic state space (HSS) modeling method, and black box method. The first three methods require detailed system information and the black box method is based on measured data.
7.2.1.1 State space averaging method The SSA method is a classical method for modeling power converters [27]. The instantaneous value of a state variable can be approximately represented by its one cycle average, expressed as Z 1 t CxD0 ðtÞ ¼ xðsÞds (7.1) T tT where T ¼ 2p/us, us is the switching frequency of the converter. The SSA method can be directly applied to get linearized models of DC/DC converters as the obtained models are time invariant with steadystate operating points. But for AC/DC or AC/DC converters, the linearized models cannot be directly obtained due to timevarying variables of threephase system with no steadystate operating points. To get linearized models of AC systems, dq transformation method [14] or harmonic linearization method [8] is applied with the SSA method. Dq transformation method is to do Park transformation for the threephase variables, then the linearized model can be obtained in dq domain with steadystate operating points [14]. Harmonic linearization method is to superimpose the system with two sinusoidal perturbations, one in positive sequence and the other in negative sequence [8]. Then the linearized model can be obtained in sequence domain with steadystate operating points. The SSA method is widely adopted. But it only considers fundamental frequency, when frequency coupling dynamics have large effects, the following two methods will be adopted, which provide more detailed models.
7.2.1.2 Generalized state space averaging method The generalized state space averaging (GSSA) method [28] is based on Fourier transform of nonperiodic signals. A signal x(t) can be approximated with arbitrary precision in the interval s ˛ ðt T; t by a Fourier series: xðtÞ ¼
N X
Xk ðtÞe jkus s
k¼N
where Fourier coefficients Xk(t) are given by
(7.2)
Overview of stability analysis methods in power electronics Chapter  7
Xk ðtÞ ¼
1 T
Z
t
xðsÞe jkus s ds
173
(7.3)
tT
It should be noted that SSA model is a special case for the GSSA model where k ¼ 0.
7.2.1.3 Harmonic state space method HSS model is a powerful technique to analyze linear time periodic systems [10]. It establishes an analogy to the state space model by introducing an exponentially modulated periodic signal, expressed as N X
xðtÞ ¼
Xk ðtÞest e jkus t
(7.4)
ðs þ jkus ÞXk ðtÞeðsþjkus Þt
(7.5)
k¼N
_ ¼ xðtÞ
N X
k¼N
Both GSSA and HSS methods can capture the frequency coupling of power converters. However, they lead to higher order with complicated expressions.
7.2.1.4 Black box method The previous three modeling methods require detailed information of models. However, in many situations, detailed information of components may be kept confidential by different vendors of components. Then databased black box method provides a good solution for this case, where models can be built based on measured output dynamics of components. The impedance measurement approach is a widely used black box modeling method for power electronic systems and is briefly introduced here [8,16,18]. The main concept of impedance measurement method is to inject small disturbance signals (voltages or currents) at various frequencies in the operating system and then appropriate currents and voltages can be extracted to obtain the desired impedance or admittance [8,16]. This method can be directly applied to extract impedance of DC/DC converters or singlephase systems. For AC/DC and DC/AC converters, measurement of impedances in dq frame can be obtained by the twostep injection procedure: the first injection is on the daxis while keeping qaxis perturbation null and then converted into abc frame to inject into the system; the corresponding voltage and current responses will be measured and converted back to dq frame; then the second injection is implemented qaxis with daxis being zero to get the system responses. Then impedances in dq frame can be calculated [18]. The impedance measurement techniques make it possible to monitor the operational region of the system and assessing the system stability during
174 Control of Power Electronic Converters and Systems
operation. For online applications, the ideal impedance measurement technique should complete the measurement in a short time, in order not to disturb normal operation for too long time, allowing a fast response to system variations. Wideband identification methods are proposed which inject a highfrequency content signal to excite all frequencies of interest simultaneously [19]. As a summary of the modeling methods, the first three modeling methods, i.e., SSA, GSSA, and HSS, are analytical solutions based on detailed information of models, and the black box method provides a powerful solution when detailed information is not available. Among the three analytical methods, the SSA method is the most commonly used approach, which approximates the instantaneous value of a state variable by its one cycle average. But it can only predict the dynamics below half the switching frequency. To further capture the frequency coupling dynamics caused by power electronic converters and provide more accurate assessment, the HSS method and GSSA method can be selected, which are based on Fourier transform of nonperiodic signals. But they also lead to higher orders with complicated expressions, resulting in increased computational burden and are difficult to use in very complex systems. When the information of models is not available, we will use black box method (e.g., impedance measurement approach) to build power converter models for systematic assessment; but the drawback is that the mechanisms inside models cannot be analyzed. These modeling methods can be selected for modeling power converters considering the available information and different levels of accuracy requirements and model complexity.
7.2.2 Eigenvalue method Eigenvalue method is widely adopted for power system stability analysis. The detailed procedure of the smallsignal stability analysis using eigenvalue method is presented in this part, including modeling and integration, stability analysis and verification, which is also illustrated in Fig. 7.2.
7.2.2.1 Component modeling and system integration First, components are modeled in the state space form using an analytical modeling method presented in 7.2.1. Normally, SSA method is adopted to build state space models of components and then a system model is integrated for eigenvalue analysis. For a largescale power system with a high penetration of power electronic converters, there would be a lot of state variables from the detailed models of components. To address the high computation burden and complexity for
Overview of stability analysis methods in power electronics Chapter  7
175
FIGURE 7.2 Procedures of smallsignal analysis with the eigenvalue method and impedancebased method.
deriving the overall system state space model, the CCM or a modulebased approach can be employed for system integration [14,15]. In the CCM, the power system is divided into multiple components with the interconnected relationship. The state space models of components are obtained and linearized as Dx_ ¼ Ai Dx þ Bi Du Dyi ¼ Ci Dx þ Di Du
(7.6)
where Ai, Bi, Ci, and Di are state matrices for the ith subsystem. The interconnection relationship is extracted from the network architecture, given by u ¼ L1 y þ L2 a b ¼ L3 y þ L4 a
(7.7)
where L1, L2, L3, and L4 are component interconnection matrices. Then the smallsignal model of the whole system is integrated as Dx_ ¼ FDx þ GDa Db ¼ HDx þ JDa where F ¼ A þ BL1 ðI DL1 Þ1 C; G ¼ BL1 ðI DL1 Þ1 DL2 þ BL2 ;
(7.8)
176 Control of Power Electronic Converters and Systems
H ¼ L3 ðI DL1 Þ1 C; J ¼ L3 ðI DL1 Þ1 DL2 þ L4 . The CCM and modular approach provide a modular and scalable solution for the largescale power electronicebased power system integration.
7.2.2.2 Eigenvalue stability analysis 7.2.2.2.1 Eigenvalue analysis The linearized state space model in (7.8) is analyzed by examining the eigenvalue of the system matrix F. The eigenvalue analysis shows different oscillation modes and damping characteristics. For a complex eigenvalue li that corresponds to an oscillatory mode of the system in (7.9), the frequency fosci and damping ratio zi of the oscillation are expressed as (7.10) and (7.11). li ¼ si jui ui 2p si zi ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ s2i þ u2i fosci ¼
(7.9) (7.10) (7.11)
The damping ratio determines the rate at which the amplitude of the oscillations decreases. The dominant eigenvalues at different oscillation frequency ranges can be identified by eigenvalue analysis. 7.2.2.2.2 Participation factor analysis Then participation factor analysis is performed [14,15,29], which calculates the contribution of each state to a certain mode, given by jwki jjvik j pki ¼ P jwki jjvik j
(7.12)
where pki is the participation of kth state to ith mode; wki and vik are kth elements in the ith left eigenvector and right eigenvector, respectively. The participation factor matrix can identify the states that have dominant effects on the dominant eigenvalues. 7.2.2.2.3 Sensitivity analysis System stability can be predicted by calculating eigenvalue loci under different operating conditions [14,15,29]. The system becomes unstable if positive eigenvalues appear. Parameters associated with the dominant states will be investigated through sensitivity analysis. The effect of these parameters (e.g., line impedance, filters, controller gain, delay time, PLL bandwidth, etc.) on system stability can be assessed by calculating the eigenvalue loci with the variation of these parameters. Then the ranges of the parameters for system stable operation of
177
Overview of stability analysis methods in power electronics Chapter  7
the system can be obtained. This also provides a guideline to improve the system damping by tuning control parameters.
7.2.2.3 Verification Timedomain simulations and experiments are typically performed to study stable and unstable conditions to verify the analysis and results. 7.2.3 Impedancebased method Impedancebased methods are widely used to analyze power electronicebased systems. The detailed procedure of the smallsignal stability analysis using an impedanceebased approach is presented, including impedance modeling/ measurement, stability analysis and verification, which is also illustrated in Fig. 7.2.
7.2.3.1 Impedance modeling 7.2.3.1.1 Impedance modeling In impedancebased method, a system can be divided into a source subsystem and a load subsystem, as shown in Fig. 7.3 [8]. Zout_S and Zin_L are the output impedance of source and input impedance of load in the sourceeload interface, respectively. The total input to output transfer function is obtained as GSL ¼
Vout L Zin L ¼ GS GL Vin S Zin L þ Zout
S
¼ GS GL
1 1 þ TMLG
(7.13)
where the minor loop gain TMLG is expressed as TMLG ¼
Zout S Zin L
(7.14)
For DC systems, impedance models are derived based on singleinput and singleoutput transfer functions from current to voltage in a smallsignal sense.
FIGURE 7.3 Impedance representation of an interconnected sourceeload system.
178 Control of Power Electronic Converters and Systems
For AC systems, impedance models are built as multipleinput and multipleoutput transfer functions either in dq domain or sequence domain [17]. For example, by applying a dqreference frame transformation to the variables in abc domain, the AC system becomes a DC system in dq domain with the source and load impedances expressed as " " # # ZS dd ðsÞ Zs dq ðsÞ ZL dd ðsÞ ZL dq ðsÞ ZS dq ¼ ; ZL dq ¼ (7.15) Zs qd ðsÞ Zs qq ðsÞ ZL qd ðsÞ ZL qq ðsÞ If the detailed model information is not available, the impedance models can be obtained from experiments or simulations using impedance measurement technique discussed in 7.2.1.4. 7.2.3.1.2 Network partitioning The impedancebased method is actually a local method. The impacts of some components may be neglected due to the lumping effect. Thus, the selection of network partitioning points to divide the system into source and load subsystems may lead to different system stability analysis results. It is also necessary to investigate the stability at different interfacing points.
7.2.3.2 Stability analysis Based on (7.13), as Gs and GL are stable transfer functions, system stability is determined by TMLG. According to Nyquist criterion, the system is stable if and only if the Nyquist contour of TMLG does not encircle (1, 0) point [16]. Many stability criteria are further developed based on this concept [16], such as Middlebrook criterion, Gain Margin and Phase Margin criterion, Opposing
OA GMPM 1/GM
ESAC
PM
Unit circle Middlebrook
1
FIGURE 7.4 Stability boundaries based on different stability criteria.
Overview of stability analysis methods in power electronics Chapter  7
179
Component criterion, Root Exponential Stability Criterion, etc. Fig. 7.4 depicts the stability boundaries based on different stability criteria. For DC system, system stability can be evaluated directly in this way. For threephase AC system, impedances of AC systems can be represented in dq domain or sequence domain. Then system stability can be evaluated and predicted by using the generalized Nyquist stability criterion to the impedance ratio [17]. To study the interaction between a converter system and the grid, a specific interfacing point needs to be selected with the converter as the load and the rest of the system as the source. Then the interaction dynamics can be analyzed in detail. The stability region can be predicted by investigating the impedance ratio with the stability criterion by varying operating profiles and different parameters to be studied. This method also provides a designoriented analysis. If a new converter system is to be integrated into the grid, given the interface and the grid impedance, the converter should be designed to ensure its impedance compatibility to the grid impedance with the stability criterion for stable operation. Moreover, with the plotted impedance characteristics, the interaction effects at different frequency ranges can be identified.
7.2.3.3 Verification of analysis The stability analysis results should be verified by simulation or experimental results, otherwise, the impedance models should be reexamined in frequency domain. 7.2.4 Comparison of methods The eigenvalue method has advantages in identifying oscillation modes and instability roots of system variables and it is preferred for comprehensive systematic analysis. However, it requires to know the full system information to perform stability analysis, which may be kept confidential by different vendors. The impedancebased method analyses the whole system based on the I/O character (impedance) of each subsystem, once the impedance of each subsystem is obtained, system stability can easily be assessed; adding or removing or modifying a subsystem will not affect other subsystems and the new system can easily be reformulated. Thus, it is suitable for analyzing interactions of subsystems and the design of converters. Moreover, the impedancebased method has black box feature, which makes it a powerful technique when the system details are unavailable or for online stability assessment. However, the conservativeness of some impedance stability criteria may lead to
180 Control of Power Electronic Converters and Systems
TABLE 7.1 Comparison of eigenvalue method and impedancebased method. Stability analysis requirements
Eigenvalue method
Impedance method
Identification of oscillation modes
O
e
Participation factor of state variables
O
e
Black box (privacy)
e
O
Converter designeoriented analysis
e
O
Interaction effects of two subsystems
e
O
conservative design and it is unable to identify the oscillation modes and participation factors. A comparison is made as a criterion for selecting the smallsignal stability analysis method based on different requirements, which is shown in Table 7.1.
7.3 Largesignal stability analysis methods Largesignal stability (i.e., transient stability) is the stability subject to largesignal disturbances caused by faults, protection, sudden load connections or disconnections, etc [21,22]. Two kinds of approaches are used to conduct largesignal stability analysis, one is using timedomain simulations with switched models [23], and the other is to use Lyapunovbased analytical methods [21]. In this section, timedomain simulation method is briefly introduced, and three Lyapunovbased analytical methods are illustrated to estimate domain of attraction that can ensure system stability under largesignal disturbances.
7.3.1 Timedomain simulations Timedomain simulation approach is a common way to investigate system largesignal stability and electromechanical transient simulation tools have been developed for analyzing system behavior under different largesignal disturbances. It provides high accuracy and validity. But the problem is the extremely high computation burden as numerous simulations should be conducted to ensure system stability over a wide operating range under various disturbances [22].
7.3.2 Lyapunovbased analytical methods Lyapunovbased analytical methods can estimate the domain of attraction, within which the system is stable under largesignal disturbances. Then
181
Overview of stability analysis methods in power electronics Chapter  7
numerous timedomain simulations are not needed. Three analytical tools for largesignal stability analysis are introduced here.
7.3.2.1 TakagieSugeno multimodel method TS multimodeling method can model nonlinear systems with multiplication, division, exponential, or square root nonlinearities. In this method, a nonlinear system is represented by a set of linear models based on a set of “ifthen” fuzzy rules. At each rule, local behavior of the nonlinear model will be represented by a local linear model. The TS models are equivalent to the nonlinear model with sufficient number of fuzzy rules [24]. For the ith rule of total r rules, the relevant local model is expressed as ( _ ¼ Ai xðtÞ þ Bi uðtÞ xðtÞ (7.16) yðtÞ ¼ Ci xðtÞ where Ai, Bi, and Ci are constant matrices. Then the nonlinear model can be expressed as 8 r X > > > _ ¼ hi ðAi xðtÞ þ Bi uðtÞÞ > xðtÞ < i¼1
(7.17)
r X > > > yðtÞ ¼ hi Ci xðtÞ > : i¼1
where hi is the respective weight of the ith rule that
r P i¼1
hi ¼ 1; hi 0.
The nonlinear system is asymptotically stable if the linear matrix inequality (LMI) condition in (7.18) is satisfied. ( M ¼ MT > 0 (7.18) ATi M þ MAi < 0; ci ˛ f1; 2; .; rg where is M is a matrix and its existence indicates the stability of the system. Once the matrix of LMI problem is known, the Lyapunov function is selected as VðxÞ ¼ xT Mx
(7.19)
The following algorithm is utilized to estimate the domain of attraction: 1.  Set all nonlinearities to the operating point values fjmin ¼ fjmax ¼ fj(0), j ¼ 1,2,.q (xjmin ¼ xjmax ¼ 0), q is the number of nonlinearities of the function f. 2.  Check LMI problem in (7.19), if it is feasible, go to Step 3, else go to Step 4.
Algorithm 7.1.
182 Control of Power Electronic Converters and Systems
3.  Decrease fjmin and increase fjmax by modifying xjmin and xjmax, j ¼ 1,2,., q. Then go to Step 2. 4.  The estimated domain of attraction can be calculated by xjmin and xjmax, j ¼ 1,2,., q. This method can estimate domain of attraction with an obtained Lyapunov function. However, if the number of nonlinearities increases, the order of matrix will grow rapidly and the problem will be very complicated to solve.
7.3.2.2 BraytoneMoser’s mixed potential This method is based on Lyapunovtype mixed potential function using the elements and circuits of the studied system [25]. Based on circuit topology and Kirchhoff’s laws, the dynamics of a nonlinear circuit can be expressed by a set of inductor currents i and capacitor voltages v. Then the mixed potential is developed as Pðv; iÞ ¼ AðiÞ BðvÞ þ Ci; g $ vD
(7.20)
where g is a constant matrix determined by circuit topology; current potential A(i) and voltage potential B(v) are the sum of integration of current set and voltage set, respectively, given by XZ AðiÞ ¼ vr di (7.21) r ˛ Ni
BðiÞ ¼
G
XZ r ˛ Nv
ir dv
(7.22)
G
A candidate Lyapunov function is selected as P ðv; iÞ ¼ l$Pðv; iÞ þ
C
D C
D
1 vPðv; iÞ 1 vPðv; iÞ 1 vPðv; iÞ 1 vPðv; iÞ ; $ ; $ þ 2 vv C vv 2 vi L vi (7.23)
According to Lyapunov theorem and LaSalle invariance principle [25], an estimated domain of attraction can be obtained as largesignal stability boundary of the system. However, this method is mainly for nonlinear LC circuits, other nonlinear systems such as motor drive system, it is often difficult to find a Lyapunov function.
Overview of stability analysis methods in power electronics Chapter  7
183
7.3.2.3 Optimal Lyapunov function generation The problem of finding the largesignal stability boundary can be formulated as an optimization problem which is to find the maximum domain of attraction for a proper Lyapunov function V(x) [26]. The optimization problem is formulated as (7.24) and a genetic algorithm is applied to solve it. min Subject to
VðxÞ _ 0 VðxÞ
(7.24)
Several candidate Lyapunov functions are presented in Ref. [30]. To solve the optimization problem, it is better to have more variables to ensure enough degrees of freedom. Therefore, this method is more helpful to solve complex systems. As can be observed from the above methods, largesignal stability analysis is much more complicated than smallsignal stability analysis. The major challenge for Lyapunovbased analytical method is to find a proper Lyapunov function for a highorder power electronicebased power system.
7.4 Case studies with practical examples 7.4.1 Smallsignal stability analysis To illustrate the smallsignal analysis method of power electronicebased power system using the eigenvalue method and impedancebased method in Section 7.2, a gridconnected threeinverter system in Fig. 7.5 is studied by using eigenvalue method and impedancebased method as a demonstration. The inverters are in currentcontrolled mode with PR controllers in the stationary frame (only current loop is considered) [9], as shown in Fig. 7.6. System parameters and control parameters are given in Table 7.2.
FIGURE 7.5 A gridconnected threeinverter system for case study.
184 Control of Power Electronic Converters and Systems
+ –
Lf
Lg
Cf
io(abc)
PCC vo(abc) PLL
abc
PWM generation 2k s kcp+ 2 cr 2 s + ω0
θ PLL
αβ
io(αβ) +
io(αβ)* αβ
io(dq)* dq
FIGURE 7.6 Control implementation of the gridconnected inverter using PR controllers.
TABLE 7.2 System parameters for system in Fig. 7.3. Inverters Inv. A
Inv. B
Inv. C
10
Switching frequency [kHz] Filter values
Lf[mH] Cf[uF] Lg[mH]
3 3 3
2 5 2
1.5 5 1.5
Parasitic values
rLf[mU] rCf[mU] rLg[mU]
0.0628 0.1061 0.0942
0.0471 0.0637 0.0628
0.0471 0.0637 0.0471
Control gain
Kp kR
20 600
12 600
12 600
Grid impedance
LS[mH] Rs [U]
2ee3 0.4ee3
7.4.1.1 Eigenvalue method By following the procedure in Section 7.2.2, the eigenvalue method is applied for stability analysis of the system in Fig. 7.5. 7.4.1.1.1 System modeling The most widely used SSA method is applied. According to the system topology in Fig. 7.5 and control structure in Fig. 7.6, state space model of Inverter i (i ¼ A, B, and C) and grid can be derived as
Overview of stability analysis methods in power electronics Chapter  7
8 diLfd;i rLf ;i 1 1 > > ¼ iLfd;i þ uiLfq;i þ vind;i vCfd;i > > Lf ;i Lf ;i dt Lf ;i > > > > > > > di r 1 1 > > Lfq;i ¼ Lf ;i iLfq;i þ uiLfd;i þ vinq;i vCfq;i > > Lf ;i Lf ;i dt Lf ;i > > > > > > > dvCfd;i 1 1 > > > ¼ iLfd;i þ uvCfq;i iLgd;i > > C C dt f ;i f ;i > > > > > > > dvCfq;i 1 1 > > ¼ iLfq;i þ uvCfd;i þ iLgq;i > > C C dt > f ;i f ;i > > > > > > > < diLgd;i ¼ rLg;i iLgd;i þ uiLgq;i þ 1 vCfd;i 1 vpccd;i Lg;i Lg;i dt Lg;i > > > > diLgq;i rLg;i 1 1 > > > ¼ iLgq;i þ uiLgd;i þ vCfq;i vpccq;i > > dt L L L > g;i g;i g;i > > > > > > dx1d;i > > ¼ iLgd;i iLgd;i > > dt > > > > > > dx1q;i > > > ¼ iLgq;i iLgq;i > > dt > > > > > > > vind;i ¼ kcp;i iLgd;i iLgd;i þ kcr;i X1d;i > > > > > > > > :v ¼ k i i inq;i cp;i Lgq;i Lgq;i þ kcr;i X1q;i 8 diGd rS 1 1 > > ¼ iGd þ uiGq þ vGd vpccd > > LS LS dt LS > > > > > > diGq rS 1 1 > > ¼ iGq þ uiGd þ vGq vpccq > < LS LS dt LS > dvpccd 1 1 > > ¼ igd þ uvpccq þ iGd > > C C dt > pcc pcc > > > > > dvpccq 1 1 > > ¼ igq uvpccd þ iGq : Cpcc Cpcc dt
185
(7.25)
(7.26)
The smallsignal models of Inverter i (i ¼ A, B, and C) and grid can be obtained by linearizing (7.25) and (7.26) at the operating point as Dx_inv;i ¼ Ainv;i Dxinv;i þ Binv;i CG DxG
(7.27)
186 Control of Power Electronic Converters and Systems
Dx_G ¼ AG DxG þ
X
BG Cinv;i Dxinv;i
(7.28)
i¼11n
where Dxinv;i ¼ DiLfd;i DiLfq;i Dvcfd;i Dvcfq;i DiLgd;i DiLgq;i DX1d;i Di1q;i T T DxG ¼ Dvpccd Dvpccq Digd Digq By using component connection approach or modulebased approach, the overall smallsignal model of the system is expressed as 3 2 Ainv;A 0 0 Binv;A CG 7 6 6 0 Ainv;B 0 Binv;B CG 7 7 6 (7.29) As ¼ 6 7 6 0 0 Ainv;C Binv;C CG 7 5 4 BG Cinv;A BG Cinv;B BG Cinv;C AG Then system stability can be evaluated by examining eigenvalues of the system matrix As. Fig. 7.7 shows the corresponding eigenvalue loci. As can be observed, the modes l1114 are in the right half plane and the system will be unstable. Then the participation factor analysis is conducted to localize the instability roots. Following (7.12), the participation factors for the system are calculated and depicted in Fig. 7.8. The dominant states to the unstable modes l1114 are iLfd,C, iLfq,C, vCfd,C, vCfq,C, iLgd,C, iLgq,C with the participation factors 0.124, 0.124, 0.249, 0.249, 0.125, and 0.125, respectively. Therefore, Inverter C has a dominant impact on the unstable result. Fig. 7.9 shows the eigenvalue loci with the variation of the current control gain from 9 to 14. As can be observed, when the current control gain is larger than 11, some eigenvalues move to the right half plane. Therefore, to ensure system stability, controller gain should be selected as 10 or smaller.
FIGURE 7.7 Eigenvalue loci of the system under study in Fig. 7.3.
Overview of stability analysis methods in power electronics Chapter  7
187
FIGURE 7.8 Participation factors for the system under study in Fig. 7.3.
FIGURE 7.9 Eigenvalue loci with the variation of the current control gain kcp,C in Inverter C from 9 to 14.
7.4.1.2 Impedancebased method 7.4.1.2.1 Impedance modeling Based on [9], the output admittance of Inverter i with an LCL filter Yoi(i ¼ A, B, C) can be derived as Yoi;i ¼
Yo;i ðsÞ 1 þ Gcgi ðsÞGPWM ðsÞYgi;i ðsÞ
with Ygi;i ðsÞ ¼
ZCf ;i ZCf ;i ZLf ;i þ ZLg;i ZLf ;i þ ZCf ;i ZLg;i
(7.30)
188 Control of Power Electronic Converters and Systems
Yo;i ðsÞ ¼
ZCf ;i þ ZLf ;i ZCf ;i ZLf ;i þ ZLg;i ZLf ;i þ ZCf ;i ZLg;i Gcgi ðsÞ ¼ kcp;i
kcr;i s þ u21
s2
GPWM ðsÞ ¼ e1:5Ts s The grid admittance Yg is derived as Yg ¼
1 RS þ sLS
(7.31)
Here the integration of Inverter C to the gridconnected inverter system is studied with Inverters A þ B as an example. Then partitioning point can be selected where the Inverter C is seen as the load side and the rest of the system is the source side, with the minor loop gain calculated as TMLG ¼
YoC Yg þ YoA þ YoB
(7.32)
Fig. 7.10depicts the Nyquist plot for the system with original parameters in red (dark gray in print version) color. It can be observed that Nyquist plot encircles (1, 0j) point and thus, the system is unstable with the integration of Inverter C. The parameter is redesigned to stabilize system, i.e., to make Nyquist plot encircle (1, 0j) point. The system is stable with proportional gain redesigned as 10, as it is shown in Fig. 7.10 with blue (dark gray in printed version) color line.
FIGURE 7.10 Nyquist plot for the system with original control parameter (unstable) and redesigned control parameter (stable).
Overview of stability analysis methods in power electronics Chapter  7
189
7.4.1.3 Verification Simulations are done in Matlab/Simulink based on Figs. 7.5 and 7.6 for the verification of the results of the above two methods. Fig. 7.11 demonstrates the simulation results of output currents of three inverters and the grid voltage with the original parameters in Table 7.2. The results indicate that the system is unstable, which verifies the analytical results of eigenvalue method in Section 7.4.1.1 and impedancebased method in Section 7.4.1.2. Fig. 7.12 presents the simulation results with the redesigned proportional gain in Inverter C, as suggested by the eigenvalue method and impedancebased method in Section 7.4.1. The results of the current dynamics and voltage dynamics show that the system is stable, which validates the analytical results. A laboratory platform of the gridconnected multiinverter system is established for further verification of the results, as shown in Fig. 7.13. Three inverters are integrated into the grid simulator though LCL filters and grid line inductor. The parameters are the same as that for simulations. The control algorithm is implemented in dSPACE 1007 to generate PWM signals for the inverters.
ioA (A)
10 0
ioB (A)
10 0.95 10
ioC (A)
0.97
0.98
0.99
1
0.96
0.97
0.98
0.99
1
0.96
0.97
0.98
0.99
1
0.96
0.97
0.98
0.99
1
0 10 0.95 10
vpcc (V)
0.96
0
10 0.95 400 200 0 200 400 0.95
Time(s) FIGURE 7.11 Simulation results with the original control parameter (kcp,C ¼ 12) for the system under study.
190 Control of Power Electronic Converters and Systems
ioA (A)
10 0
ioB (A)
10 1.5 10
ioC (A)
1.52
1.53
1.54
1.55
1.51
1.52
1.53
1.54
1.55
1.51
1.52
1.53
1.54
1.55
1.51
1.52
1.53
1.54
1.55
0 10 1.5 10
vpcc (V)
1.51
0
10 1.5 400 200 0 200 400 1.5
Time(s) FIGURE 7.12 Simulation results with the redesigned control parameter (kcp,C ¼ 10) for the system under study.
FIGURE 7.13 Experimental platform using threeinverter systems connected to a grid.
The experimental results of the system with original parameters listed in Table 7.2 are shown in Fig. 7.14. The current response of three inverters and the voltage response at PCC show that the system is unstable, which verifies the analytical and simulation results.
Overview of stability analysis methods in power electronics Chapter  7
191
FIGURE 7.14 Experimental results with the original control parameter (kcp,C ¼ 12) for the system under study.
Fig. 7.15 shows the experimental results with the redesigned current loop proportional gain of Inverter C to be 10. The system is stable, which validates the analytical and simulation results are feasible to be used.
FIGURE 7.15 Experimental results with the redesigned control parameter (kcp,C ¼ 10) for the system under study.
192 Control of Power Electronic Converters and Systems
7.4.2 Largesignal stability analysis on a power electronic system Here an example of a large signal stability analysis is shown for a simple DC system with constant power load (CPL) by using TS multimodeling approach [20,24] illustrated in Section 7.3.2.1. Stability issues caused by a high penetration of power electronic converters in DC microgrid attract much attention in recent years. Tightly controlled power electronic converters behave as CPLs with destabilizing effects. The interaction of CPLs with an LC filter is studied in a simple DC distribution system, as shown in Fig. 7.16. The parameters for the system under study are given in Table 7.3. The model of the system is expressed as 3 2 rf 1 2 3 " # 7 x 6 Lf Lf _ x 1 7 6 1 x_ ¼ 4 5 ¼ 6 (7.33) ¼ A1 ðx2 Þx 7 x_2 5 4 1 x ps0 2 f1 ðx2 Þ C CVs0 with f1 ðx2 Þ ¼
1 . x2 þ Vs0
FIGURE 7.16 A simplified DC distribution system.
TABLE 7.3 Parameters for the DC distribution system. DClink voltage (E)
200 V
DClink filter (Lf, rf, Cf)
39.5 mH, 1.1 U, 0.501 mF
Constant power load Ps
300 W
Overview of stability analysis methods in power electronics Chapter  7
193
As can be observed from (7.33), there is only one nonlinearity. Following the TS modeling method in 7.3.1, suppose x2 belongs to the interval [x2min, x2max], which is to be determined by Algorithm 7.1. Let 8 1 > > > < f1min ¼ x 2max þ Vs0 (7.34) > 1 > > : f1max ¼ x2min þ Vs0 Here two local models using two fuzzy rules can be applied: Rule 1: if x2 is x2min, the model is obtained as 3 2 rf 1 6 Lf Lf 7 7 6 x_ ¼ 6 7x ¼ A1 x 5 4 1 ps0 f1min C CVs0 Rule 2: if x2 is x2max, the model is obtained as 3 2 rf 1 6 Lf Lf 7 7 6 x_ ¼ 6 7x ¼ A2 x 5 4 1 ps0 f1max C CVs0
(7.35)
(7.36)
Then the border of the domain can be calculated considering the LMI condition in (7.37) and the algorithm in Section 7.3.1. 8 M ¼ MT > 0 > > < (7.37) AT1 M þ MA1 < 0 > > : T A2 M þ MA2 < 0 Then the domain of attraction is calculated as 103:1x21 þ 2:86x1 x2 þ 1:31x22 ¼ 10292 (7.38) " # 103:1 1:43 with M obtained as M ¼ at x2min ¼ 89:4. 1:43 1:31 The domain of attraction using TS multimodeling method is plotted in Fig. 7.17. The effect of the load power on the domain of attraction can be explored and it is shown that the maximum load of the system is 500 W beyond which the domain of attraction vanishes. Fig. 7.18 shows the timedomain simulation result when load power increases from 200 to 400 to 600 W. It shows that system becomes unstable when load increases from 400 to 600 W, which is coherent with the stability region of the analytical method.
194 Control of Power Electronic Converters and Systems
FIGURE 7.17 Estimated domain of attraction using TS multimodeling method for largesignal stability analysis.
FIGURE 7.18 Timedomain simulation of the DC system with load power variation.
Overview of stability analysis methods in power electronics Chapter  7
195
7.5 Summary In this chapter, we introduce smallsignal stability analysis and largesignal stability methods. First, modeling methods are introduced, and a proper modeling method can be selected based on information availability and accuracy requirement. Then the commonly used smallsignal stability analysis tools, i.e., eigenvalue method and impedancebased method, are demonstrated in details, with case studies and simulation/experimental verifications. It shows that the eigenvalue method is a good solution to find the unstable root of the overall system and design the stable parameters; while impedance method is suitable to investigate the interactions of two subsystems or when the information of subsystems is not available. The largesignal stability analysis tools are introduced which include timedomain simulation method and Lyapunovbased analytical methods with a small case study of a DC system for demonstration. For smallsignal stability analysis, it is necessary to develop a technique that is able to identify the unstable root and does not require detailed model information, i.e., have both advantages of the eigenvalue method and impedancebased method. For largesignal stability analysis, it is challenging to find a proper Lyapunov function for large power electronicebased power systems to estimate domain of attraction, which should be explored in the future.
References [1] [2]
[3]
[4]
[5]
[6]
[7]
[8]
F. Blaabjerg, Z. Chen, S.B. Kjaer, Power electronics as efficient interface in dispersed power generation systems, IEEE Trans. Power Electron. 19 (5) (2004) 1184e1194. M. ElShimy, A. Sharaf, H. Khairy, G. Hashem, Reducedorder modelling of solarPV generators for smallsignal stability assessment of power systems and estimation of maximum penetration levels, IET Gener. Transm. Distrib. 12 (8) (2018) 1838e1847. Y. Mishra, S. Mishra, F. Li, Z.Y. Dong, R.C. Bansal, Smallsignal stability analysis of a DFIG based wind power system under different modes of operation, IEEE Trans. Energy Convers. 24 (4) (2009) 972e982. N. Pogaku, M. Prodanovic, T.C. Green, Modeling, analysis and testing of autonomous operation of an inverterbased microgrid, IEEE Trans. Power Electron. 22 (2) (2007) 613e625. E. Mollerstedt, B. Bernhardsson, Out of control because of harmonicsan analysis of the harmonic response of an inverter locomotive, IEEE Control Syst. Mag. 20 (4) (2000) 70e81. M. Amin, M. Molinas, Understanding the origin of oscillatory phenomena observed between wind farms and HVdc systems, IEEE J. Emerging Sel. Top. Power Electron. 5 (1) (2017) 378e392. F. Blaabjerg, R. Teodorescu, M. Liserre, A. V Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (2006) 1398e1409. J. Sun, Smallsignal methods for AC distributed power systems–a review, IEEE Trans. Power Electron. 24 (11) (2009) 2545e2554.
196 Control of Power Electronic Converters and Systems [9] X. Wang, F. Blaabjerg, W. Wu, Modeling and analysis of harmonic stability in an AC powerelectronics based power system, IEEE Trans. Power Electron. 29 (12) (2014) 6421e6432. [10] X. Wang, F. Blaabjerg, Harmonic stability in power electronicbased power systems: concept, modeling, and analysis, IEEE Trans. Smart Grid 10 (3) (May 2019) 2858e2870. [11] Y. Wang, X. Wang, F. Blaabjerg, Z. Chen, Harmonic instability assessment using statespace modeling and participation analysis in inverterfed power systems, IEEE Trans. Ind. Electron. 64 (1) (2017) 806e816. [12] E. Ebrahimzadeh, F. Blaabjerg, X. Wang, C.L. Bak, Harmonic stability and resonance analysis in large PMSGbased wind power plants, IEEE Trans. Sustain. Energy 9 (1) (2018) 12e23. [13] S. Eftekharnejad, V. Vittal, G.T. Heydt, B. Keel, J. Loehr, Small signal stability assessment of power systems with increased penetration of photovoltaic generation: a case study, IEEE Trans. Sustain. Energy 4 (4) (2013) 960e967. [14] Q. Xu, P. Wang, J. Chen, C. Wen, M.Y. Lee, A modulebased approach for stability analysis of complex moreelectric aircraft power system, IEEE Trans. Transp. Electrif. 3 (4) (2017) 901e919. [15] Y. Wang, X. Wang, Z. Chen, F. Blaabjerg, Smallsignal stability analysis of inverterfed power systems using component connection method, IEEE Trans. Smart Grid 9 (5) (2018) 5301e5310. [16] S.D. Sudhoff, S.F. Glover, Admittance space stability analysis of power electronic systems, IEEE Trans. Aerosp. Electron. Syst. 36 (3 PART 1) (2000) 965e973. [17] B. Wen, R. Burgos, D. Boroyevich, P. Mattavelli, Z. Shen, AC stability analysis and dq frame impedance specifications in powerelectronicsbased distributed power systems, IEEE J. Emerging Sel. Top. Power Electron. 5 (4) (2017) 1455e1465. [18] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, Z. Shen, Smallsignal stability analysis of threephase AC systems in the presence of constant power loads based on measured dq frame impedances, IEEE Trans. Power Electron. 30 (10) (2015). [19] A. Riccobono, M. Mirz, A. Monti, Noninvasive online parametric identification of threephase AC power impedances to assess the stability of gridtied power electronic inverters in LV networks, IEEE J. Emerging Sel. Top. Power Electron. 6 (2) (2017) 629e647. [20] M. Amin, M. Molinas, Understanding the origin of oscillatory phenomena observed between wind farms and HVDC systems, IEEE J. Emerging Sel. Top. Power Electron. 5 (1) (2016) 378e392. [21] D. Marx, P. Magne, B. NahidMobarakeh, S. Pierfederici, B. Davat, Large signal stability analysis tools in DC power systems with constant power loads and variable power loadsA review, IEEE Trans. Power Electron. 27 (4) (2012) 1773e1787. [22] M. Kabalan, P. Singh, D. Niebur, Large signal Lyapunovbased stability studies in microgrids: a review, IEEE Trans. Smart Grid 8 (5) (Sepember 2017) 2287e2295. [23] E. Nasrazadani, et al., Stability analysis of unbalanced distribution systems with synchronous machine and DFIG based distributed generators, IEEE Trans. Smart Grid 5 (5) (2014) 2326e2338. [24] P. Magne, B. NahidMobarakeh, S. Pierfederici, General active global stabilization of multiloads DCpower networks, IEEE Trans. Power Electron. 27 (4) (April 2012) 1788e1798. [25] A. Griffo, J. Wang, Large signal stability analysis of ’more electric’ aircraft power systems with constant power loads, IEEE Trans. Aerosp. Electron. Syst. 48 (1) (January 2012) 477e489.
Overview of stability analysis methods in power electronics Chapter  7 [26] [27] [28]
[29]
[30]
197
B.P. Loop, S.D. Sudhoff, S.H. Zak, E.L. Zivi, Estimating regions of asymptotic stability of power, IEEE Trans. Control Syst. Tech. 18 (5) (2010) 1011e1022. S.R. Sanders, G.C. Verghese, Synthesis of averaged circuit models for switched power converters, IEEE Trans. Circuits Syst. 38 (8) (1991) 905e915. H. Ebrahimi, H. Elkishky, A novel generalized statespace averaging (GSSA) model for advanced aircraft electric power systems, Energy Convers. Manag. 89 (January 2015) 507e524. G.O. Kalcon, G.P. Adam, O. AnayaLara, S. Lo, K. Uhlen, Smallsignal stability analysis of multiterminal VSCbased DC transmission systems, IEEE Trans. Power Syst. 27 (4) (2012) 1818e1830. _ Stanislaw, Methods of Optimal Lyapunov C.J. Sullivan, S.D. Sudhoff, E.L. Zivi, H.Z. Function Generation with Application to Power Electronic Converters and Systems, 2007, pp. 267e274.
This page intentionally left blank
Chapter 8
Cyber security in power electronic systems Subham Sahoo Department of Energy Technology, Aalborg University, Aalborg East, Nordjylland, Denmark
8.1 Introduction The development of carbonneutral electrical power systems is one of the most important global technical targets and a challenge at the same time in this century. It will not only minimize emissions and the impact of global warming but will also reduce the overall public reliance on unreliable fossil fuel supplies. Widescale deployment of renewable energy sources (RES) such as wind and photovoltaic (PV) transmission systems, energy storage systems (ESSs), electric vehicles (EVs), and highvoltage DC (HVDC) transmission systems is seen as critical initiatives for achieving this objective [1]. Under these circumstances, gridtied voltage source converters (VSCs) play a key role as they act as the most common energy conversion interfaces between these technologies and the electrical grid [2]. It should also be noted that VSCs allow the creation of intelligent microgrids (MGs), which are seen as intermediate aggregation entities that can either function in standalone mode or facilitate the integration of distributed energy resources in gridtied mode on a large scale [3,4]. As the number of VSCs in the power grid increases, their impact on these grids is also becoming more pronounced. With the grid modernization being carried out swiftly, several VSCs are integrated into the existing distribution network to provide gridsupportive services. In reality, the assimilation of these facilities has led to a plight, creating a direct tradeoff between reliability and security for larger interconnected VSC network. In fact, such largescale monitoring using supervisory control and data acquisition (SCADA) makes it highly susceptible to malicious intrusions. In addition, the reliability aspect involved in deep integration of communication layers in order to achieve synchronization often leads to new safety issues. These threats range from theft to cyber attack, which may eventually lead to system collapse, cascaded failure, damage to consumer loads, endangered activities in the energy market, etc [5]. Many cyber power outage Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.00007X Copyright © 2021 Elsevier Ltd. All rights reserved.
199
200 Control of Power Electronic Converters and Systems
incidents in Brazil have been published [6], including the SQL Slammer worm attack, the Stuxnet attack, and numerous industrial calamities. In addition, the McAfee Study [7] reported that 80% of the utility companies encountered at least one denial of service (DoS) attack in their communication network, with 85% of unit data penetrated by an adversary. Since the most popular mode of communication in smart grids is wireless, IT security customers control different data protection policies to tackle data transmission device unreliability. However, smartly designed cyber attacks with abundant system information create disparity in securing the electrical grid, as they easily bypass the model verification tests [8]. It emanates additional vulnerabilities from the perspective of control systems in the smart grid, since the newly established methods of verification are mostly based on information technologies. Intelligent attacks often seek to monitor the integrity of the system as a concealed disorder. This mandates the detection of cyber attacks in a timely manner in order to avoid unnecessary system casualties, taking into account a considerable timescale separation of the control phases of the VSC with a value of no more than 0.1 s. Apart from the said liabilities, it also violates confidentiality and the optimal functioning of the system almost immediately on one hand. On the other hand, the attacker can initiate the attack in slightly alarming circumstances to arrange drastic system shutdowns, since the penetration of intelligent attacks continues stealthily without being detected by controltheoretical solutions. Such situations would be considered as alternatives to planning and power backup. Therefore, the emphasis on resolving these critical safety issues needs to be the suitable design for VSC for secure, robust, and intelligent control methodologies. For the purpose of better understanding of security problems in the control of cyber physical power electronic systems, this chapter will discuss the following points: 1. Control and operational challenges faced by the VSCs used in different applications due to cyber attacks. 2. A brief overview of the vulnerabilities in the control and cyber layer of VSCs (in gridconnected and standalone mode) is provided. Further, more aspects on how it disorients their operation from the state of normalcy is detailed. 3. Directions and viewpoints, especially in the design of resilient control formulation for VSCs.
8.2 Cyber physical architecture of power electronic converters A typical architecture of the network of VSCs connected to the AC grid is shown in Fig. 8.1. There are several phases in the total power conversion chain, such as input, inputside power transformer, DC voltage phase, VSC
Cyber security in power electronic systems Chapter  8
A. Inputstage
B. Inputside converters
C. DCstage
DC/D C DC/DC
PV
Control Wind
ESS
MPPT algorithm ChargIng/ discharging algorithm
AC grid
E. ACstage
DC/AC
DClink capacitors
DC microgrid
DC voltage control Power sharing
D. Gridtied VSC Filter
201
HVDC cables
PCC Control
AC grid
Frequency control DC voltage Synchrocontrol nization AC voltage Virtual control impedance or Fault rideadmittance through (outer and AC current inner) control Sampling & Switching
AC microgrid
UPS loads
Cyber layer FIGURE 8.1 Control and physical stages in an individual gridtied voltage source converter system connected in different configurations.
network stage, AC grid stage, and cyber stage, as shown in Fig. 8.1. This kind of power architecture is used most commonly for RES interfaces such as wind and PV [9], ESS [10], and the electric grids EV charging network [11]. In order to improve the strength and reliability of smart grids, it is anticipated that individual VSC systems will be interconnected through a unique allinclusive cyber physical smart grid through communication links in the near future. The control stages which are referred to the above segments are listed in the sections below.
8.2.1 Physical stage The exemplary input power sources/sinks are located on the far left side of Fig. 8.1. Some units in the input stage such as grid or ESS can either inject or absorb the electric power. Inputside converters may also be programmed to automatically shift the operating modes online. These mode shifts are implemented in systems that have multiple units connected to the common DC bus and their purpose is to maintain the power balance in the common DC bus under all operating conditions. For instance, RES may change the operation from MPPT mode to voltage control mode and viceversa. Similarly, ESSs may move from charging mode to DC voltage control mode and back. These mode shifts can be done either through a centralized supervisory control system enabled by communication technologies [12], or in a decentralized way [13,14]. The output of AC gridetied VSC is connected through the interface filter either to an AC electrical power grid, AC MG, or to standalone AC loads, as shown in Fig. 8.1. The most common filter structures are L, LC, and LCL.
202 Control of Power Electronic Converters and Systems
The connection point is commonly referred to as the point of common coupling (PCC). Depending on the type of connection, different standards are applicable. For instance, in the case of grid connection, voltage at the PCC is defined by the legacy AC grid so primary concern of VSC is to regulate the grid current with high quality in steady state and with predefined behavior during transients (e.g., voltage sags, swells, and unbalances) [15]. Recently, an increasing number of gridancillary services related to grid voltage and frequency support are also required [15]. Based on its interconnection with different AC stages, various standards are applicable. For electric grid, the primary concern lies with the regulation of grid current with high power qualitative signatures during transients (voltage sags, swells, and unbalances) [15]. In recent years, increased numbers of gridancillary services are also required in terms of grid stress reduction and frequency support [15]. On the other hand, its output in an inertiafree autonomous system, such as MG systems, is effectively regulated with the ability to share active and reactive control. Such objectives are accomplished by primary control of the quantities described above, which will be discussed later in this chapter.
8.2.2 Cyber stage As smart grid comprising of numerous VSCs, together with conventional synchronous generators, jointly regulate the grid, each of these units is termed as an agent for an exemplary portion of a smart grid with interconnected VSCs. The sensors and controllers that operate jointly in the smart grid are linked by a communication network. Each agent can interact in two different ways: (a) to a central controller and (b) among each other in a distributed manner. A pictorial description of both the cyber structures is provided in Fig. 8.2, where the dotted lines represent the flow of information. Since the control objectives is highly vulnerable to single pointoffailure in a centralized network, the distributed control theory in Fig. 8.2B is used prominently to improve the reliability and scalability in power electronicebased cyber physical systems.
(A)
(B) 1
Central Controller 4
1
2
3
4
2
3
FIGURE 8.2 Communication topologies with four units: (A) Centralized control, (B) Distributed control.
Cyber security in power electronic systems Chapter  8
203
Each agent has a distributed controller that uses data from local and neighboring agents. These data are typically obtained via phasor measurement units (PMUs), consisting of dynamic voltage phasors. Communication between the PMUs and local controllers can be carried out centrally, where measurements are obtained from all the agents for analysis and decision making. The SCADA framework is typically used to alleviate surveillance in smart grid networks [16]. This strategy not only requires substantial means of communication but is also vulnerable to possible cyber attacks, if the number of agents is high. Another method, commonly known as decentralized control, involves a scheme that uses only local steps. Although the communication network here is totally avoided, the capability to monitor is restricted. As already explained above, a distributed control paradigm introduces flexibility since the computational resources are uniformly dispersed across the system to achieve coordination. Hence, low bandwidth communication channels can be employed to achieve the same function. Although it offers an obvious evaluation criterion for intrusion attempts, resistance to cyber attacks cannot automatically be assured for coordinated attacks [17e20]. This can be explained owing to the inadequate information in each node for cyber attack detection. A brief description of the control functions of AC gridetied VSCs according to their timescales is provided in Fig. 8.3. Each control loop (as highlighted in Fig. 8.3) is shown next to each other, indicating that they are concurrently controlled (e.g., active damping and AC current control [21], DClink voltage control and synchronization [22], or fault ride and virtual impedance/admittance control [23]). The VSC’s roles in renewablebased power systems and MGs can be divided into two main categories, i.e., gridfollowing, and gridforming [24]. On the other hand, the internal control loops, as shown in Fig. 8.3, are resilient to cyber attacks, because they operate for each state with a tracking objective. It is worth noting that the internal control loop only becomes vulnerable to cyber attacks when the outer control loop is unattacked. Since the secondary control layer exploits communication to alter the outer control loop references, any baddata injection into the upper control layer (highlighted in Fig. 8.3) disorients stability or causes system shutdown. The shutdown is usually caused by the involuntary activation of overvoltage and overcurrent layers of protection. More discussion on VSC’s vulnerable control layers will be conducted in the next section after a brief theory about cyber security is provided.
8.3 Vulnerability analysis of cyber attacks on control of VSCs 8.3.1 Cyber security Cyber threats are becoming a reality with the emergence of networking technologies. These disruptions can dramatically affect the efficiency of smart
204 Control of Power Electronic Converters and Systems
FIGURE 8.3 Conventional control structure for twolevel VSCdVulnerable control layer against cyber attacks.
grids, as it is evident from multiple realworld examples. Rapidly growing penetration of VSC technologies and their impact on the system is approaching a point, where it is difficult to disregard the risk of cyber attacks. In particular, researchers are focusing more on developing secure control methodologies rather than improving the conventional encryptionbased techniques. Spoofing attacks can usually be triggered on sensors and communication links, where the signals are disrupted, quantized, or coerced. To name a few, false data injection attacks (FDIAs) are triggered by injecting auxiliary signals or altering the output of the sensor measurements [25]. It is generally referred to as the maninthemiddle (MITM) attack when a similar activity is detected in the communication channels [26]. In addition, signal jamming can also be triggered to disrupt signal transmission, which is generally known as service denial (DoS) attack [27]. These are some of the prominent cyber attacks that have precipitated in real time [28].
8.3.2 Vulnerability assessment A conventional singleline representation for the gridforming VSCs is shown in Fig. 8.4. Gridforming VSCs control local voltage and frequency. The general principle is to coordinate primary droop control locally using the available measures to synchronize with other AC sources. This decentralized system is considerably secure from a cyber space perspective, as it is difficult for the attackers to penetrate the physical layer. Additionally, effective physical layer protection alternatives such as beamforming are widely used these days [29]. However, the philosophies of decentralized management suffer from an organizational point of view when it comes to conforming to commercial regulatory requirements. Normally secondary controller conceived this drawback using the details from other VSCs. Referring to the cyber framework in Fig. 8.2, distributed or centralized secondary control systems may be placed on the primary control legislation to account for the offsets. However, this leaves the attackers a large vulnerable space to locate the
Cyber security in power electronic systems Chapter  8
205
∼
Vc ∠ φc
Load
Lg
VSC+LC filter FIGURE 8.4 Simplified representation of gridforming VSC.
attacked data in the sensors, communication links, or the controller. Below are some of the common methods of intrusion approaches to manipulate each component: l
l
l
Sensors: Data from the sensors are normally manipulated within the control platform by penetration of the adversary. Trojan Horse [30] can easily achieve this penetration by using remote systems as host. The sensor output from the acquisition panel is usually within signed 15 V. Acquisition gains using a linear plotting method are used to calibrate it against the actual measurement. The attacker usually attempts to change the acquisition gains which creates a bias in the reported measurements. Communication Links: The communicated data can be manipulated either inside the controller or in a router/encoder/decoder communication point. There are several ways in which the transmitted data can be manipulated, such as infringement of authorization, interruption of signal transmission, illegal opening of information logs, replay of the transmitted information from the past, etc. Controller: As already mentioned, the controller can be illegitimately accessed using Trojan Horse to modify the reference input(s) used for control of VSCs either in the outer control loop or in the secondary controller.
All the vulnerable points for gridforming VSCs have been detailed in Table 8.1. A detailed vulnerability assessment of gridfollowing and gridsupporting VSCs including applications such as DFIG, HVDC, etc., follows the same hypothesis with maximum threat from the controller [31].
8.4 Cyber attack detection and mitigation mechanisms in power electronic systems 8.4.1 Detection When the control layers of a VSC are implemented in realtime processors, intrusion into the control layer only allows access to the reference setpoints (DClink voltage, frequency) during runtime instead of the internal control layers.
206 Control of Power Electronic Converters and Systems
FIGURE 8.5 Basic V f control of gridforming VSCs: Attack elements injected into sensors/ communication link.
TABLE 8.1 Vulnerable points in control of gridforming VSCs in Fig. 8.5. Control layer
Vulnerable?
Current control
e
Voltage control
FDIA on u , V (only without secondary controller)
Secondary control
DoSa/MITMb attack on Vj , uj FDIAd on Vk , uk
c
a
Denial of Service. Maninthemiddle. Communicated. d False data injection attack. b c
Since the internal loops are compiled into the processor’s readonly memory portion, the device operation cannot be dissembled by intrusion into the sensor values. However, when the references are modified to cause instability or activation of the protection layer, the device dynamics can vary. Mathematically, this can be explained using the state space representation of ith VSC for a system with N VSCs using x_i ðtÞ ¼ Axi ðtÞ þ Bui ðtÞ yi ðtÞ ¼ Cxi ðtÞ þ Dui ðtÞ
(8.1)
Cyber security in power electronic systems Chapter  8
207
c i ˛ N, where xi ¼ ½vg ig P Q vdc T and u ¼ ½u vdcref P Q V T with the state parameters xi denoted by grid voltage, grid current, active power, reactive power, DC voltage, respectively; and the input consisting of the reference parameters of frequency, DC voltage, active power, reactive power, and inverter voltage for ith VSC, respectively. Further, x ˛ ℝN , u ˛ ℝM , y ˛ ℝS , A ˛ ℝNN , B ˛ ℝNM , C ˛ ℝPN , and D ˛ ℝPM . Without loss of generality, it can be assumed that each state and output variable can be independently compromised by an attacker. An attack signal xi ðtÞ ˛ ℝPþN depends specifically upon the attack strategy. If S ¼ fx1 ; x2 ; .; xNþP g is a null vector, then the system response is unbiased. To detect the presence of cyber attack elements, a residual signal r: ℝ0 /ℝP test can be followed. It is worth notifying that xi is not a design parameter; as it completely depends on the intent of the attacker. To detect attacks using a centralized attack detection filter based on a modified Luenberger observer, the estimated dynamics of ith VSC with known initial states xð0Þ can be given by x i ðtÞ Gyi ðtÞ xb_ i ðtÞ ¼ ðA þ GCÞb ri ðtÞ ¼ Cb x i ðtÞ yi ðtÞ
(8.2)
where xbi ðtÞ denote the estimated states. Further, xbi ð0Þ ¼ xi ð0Þ and the output injection matrix G ˛ ℝNP is such that (A þ GC) is Hurwitz. Hence, ri ðtÞ r if and only if xi ðtÞ ¼ 0 for t ˛ ℝ0, where r is an infinitesimal value. It is intuitive from Fig. 8.6 that the normal residual test is passed for the violet (light gray in print version) trajectory since the residual value remains within the threshold r. Otherwise, it could be inferred that the ith VSC contains an attack element. As these attacks cause system response alteration due to modified model, the residual item overshoots out of the shaded circle in Fig. 8.6. Hence, any
Normal operation
O
Cyber attack
r
FIGURE 8.6 Attack detection filter lawdTrajectories outside r denote the presence of cyber attack.
208 Control of Power Electronic Converters and Systems
physical disturbances such as load change, faults, and line breakdown will always satisfy the abovementioned detection criteria as the dynamics of the model will always be unchanged using the unbiased measurements during these disturbances. Based on the attacked point(s) in cyber physical power electronic systems, the designed anomaly detection model in (8.2) can be used to detect compromised controllers, sensors, and cyber links. Various cyber attack detection techniques in Table 8.2 have been designed from a control theory perspective to disregard the false data injected by the attacker. To eliminate false data in the control inputs, the vulnerable states are often adjoined using an alias relationship, also termed as the watermarking strategy [32], which theoretically is unknown to the attacker. Further to detect the presence of bad data in incoming/outgoing cyber links via an MITM attack, an improved version of the baddata detector in (8.2) is designed only using local measurements, such that the localization principle can be leveraged [33]. Apart from maligning the communicated signal, they can also be interrupted leading to loss of information, commonly referred as DoS attacks. These attacks can be modeled as a fuzzy disturbance to the plant, leaving behind insufficient information of the attacked agent. To solve this issue, a signal temporal logicebased controller is designed in Ref. [34] to assimilate the
TABLE 8.2 Detectability of different cyber attacks in electrical networks. Vulnerable points Controller
Sensors
Cyber link
Detection techniques
Attack
Remarks
U
Watermarking [32]
Hijacking
False control inputs
U
U
Baddata detectors [35]
FDIA
False sensing
U
Localization [33]
MITM
Observable attacks
U
Signal temporal logic [34]
DoS
Observable attacks
U
U
Hyperproperties [36] PIbased consensus [17e20]
Generalized FDIA
Difficult to detect
U
U
U
Generalized FDIA
Undetectable
e
Cyber security in power electronic systems Chapter  8
209
timestamped behavior of the missing communicated signal. Additionally using an artificial intelligence perspective, datadriven techniques such as artificial neural networkebased controllers can also be designed to predict the missing information by studying the inputeoutput behavior of the system. Extending this theory for interconnected VSCs, the artificial dynamics created by the attack element can be nullified in (1), only when N X
xi ¼ 0
(8.3)
i¼1
holds true. Further, these attacks in the attack set S can be categorized as undetectable from the monitors, if and only if x ˛ ℝN such that sI Aj0 þ Cxj ¼ f, where jSj ¼ f. Such attacks are commonly termed as coordi0 nated attacks, since they easily bypass the attack filters in (8.2). These attacks have also been tabulated in Table 8.2, where generalized FDIAs are detected using a systematic and sophisticated model identification tool by extracting the hyperproperties [36] or analyzing the system outputs using PIbased extended consensus algorithms [17e20], only when the sensors and cyber links are attacked concurrently. On the other hand, if all of the cyber physical entities such as controller, sensors, and cyber links are compromised, it is impossible to detect the attack element using any security conventions since the attack element has completely encapsulated the property of the system and its response to physical disturbances. Using (8.3), it can be concluded that an external entity can manipulate the control inputs either in the controller or on the communication link(s). A brief overview of the attack detection techniques subjected to various vulnerabilities in the control layer has been schematically demonstrated in Fig. 8.7. As the cyber and control layer are closely linked, the vulnerability to cyber attacks worsens for an interconnected VSC network. As attackvulnerable points
FIGURE 8.7 A schematic of detection techniques and vulnerabilities in the control layer for cyber secure measures on control of networked power electronic system.
210 Control of Power Electronic Converters and Systems
increase, the ancillary support provided by interconnected VSCs can be easily misled, resulting in system collapse. These effects inevitably trigger technoeconomic catastrophes by inserting false data attack vectors into the cyber physical layer to malign the electric network.
8.4.2 Mitigation To remove the aforementioned cyber attacks in power electronic systems, the mitigating action needs to be fast, otherwise the network can become unstable or even lead to shutdown. Most cyber security papers in power electronic systems are limited to detection without providing any comprehensive steps of countermeasures for normal system operation to remove the attack element(s). Some papers also tend to remove the compromised information received from the attacked unit(s) is as an elementary approach to prevent the propagation of attack into the system [37]. As a result, the network connectivity is affected degrading the control performance. In Ref. [34], O. Beg et al. have proposed an attack impact quantification technique and suppressed the impact of attack element using a deterministic number in the lowpass filter. However, the scalability of the mitigation approach is not largely discussed. Another welldefined mitigation approach is to employ an observer for each unit to operate with the estimated states using the preattack points upon detection of attack [38]. Even though these approaches are quite efficient, they have modelintensive requirements, where their performance is highly prone to model uncertainties. Moreover, the design of observer can be complex, while its realtime execution may require heavy computational resources. Additionally, an upper boundebased mitigation condition is also proposed in Ref. [39], where the mitigation strategy is selectively determined based on the total number of compromised units, termed as Ftotal, or the local compromised agents in the neighborhood of each unit, termed as Flocal. Although it counteracts against attacks on sensors, actuators, and communication links, it might affect the cyber graph connectivity by unnecessarily abandoning neighborVs information during a load change even when there is no attack. As a result, its operation becomes a point of serious concern for stealth attacks, which entails zero dynamics in distributed networks. Since the abovementioned approaches are based on restrictive assumption on the information exchange in the cyber network, a selfhealing mitigation strategy needs to be developed, which provides maximum resilience for the system to recover without losing the cyber network connectivity. To address these issues, a signal reconstructionebased strategy is devised in Ref. [40], where the presence of a cyber attack in a given agent is classified as an event. As compared to eventtriggered schemes [41], the attack detection criteria can be used as an eventtriggering mechanism for the said countermeasure to operate immediately. As soon as the authentication is checked for the monitor signals, the signal reconstruction process is activated as shown in Fig. 8.8. This mandates the requirement of at least one trustworthy converter,
Cyber security in power electronic systems Chapter  8
FIGURE 8.8 Signal reconstructionebased mitigation strategy against cyber attacks in power electronic systems.
211
212 Control of Power Electronic Converters and Systems
which is used as a reference to reconstruct the digital signal to be used in the compromised agent. This requirement is coordinated based on the control objectives and define an error surface to limit the difference between the trustworthy and constructed signal within jjejj. This leads to triggering instants, wherein the intertriggering space is held to the last triggered value. In this way, the new constructed signal is generated as shown in Fig. 8.8. The resolution of the constructed signal will improve as e is kept smaller. As described in Ref. [20], the value of e can be as small as possible yet it should be sufficiently higher than the measurement noise to avoid unnecessary triggering. Finally using the abovementioned strategies, the most prominent cyber attacks can be removed from the conventional controllers defined for power electronic systems. Although they provide resiliency against false intrusion attempts, there could be many other shortcomings such as stability issues, determination of critical detection, and mitigation time period, which needs further investigation.
8.5 Test cases 8.5.1 Test case I In this scenario, a rampbased cyber attack on frequency reference is conducted on a single gridforming VSC (N ¼ 1) (as shown in Fig. 8.5) operating at a global voltage and frequency reference of 311 V and 50 Hz, respectively. More details on the system and control parameters can be referred here [42]. At t ¼ 0.2 s, a ramp signal ua is injected into the frequency reference for control of gridforming VSC in Fig. 8.9 using ua ¼ u þ 2p1000t ﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄ}
(8.4)
ua
As soon as the attack is conducted in Fig. 8.9, the frequency at which the voltage is formulated ramps up leading to issues such as heating in the LC filter and operability issues of critical loads. As this attack acts like an exogeneous disturbance to the state space equations defined for gridforming VSC in (8.1), a
FIGURE 8.9 A rampbased attack element is injected into the frequency reference of single VSC at t ¼ 0.2 sdFrequency increases leading to heating and operability issues of critical loads.
213
Cyber security in power electronic systems Chapter  8
residue can be dynamically calculated using the Luenberger observer. However, such attacks conducted in multiple gridforming VSCs at the same time can bypass the detection norms easily. This has been studied in the next case study, which mandates a robust detection scheme for coordinated attacks.
8.5.2 Test case II In this scenario, a coordinated attack on frequency is carried out in a distributed AC MG (as shown in Fig. 8.5) operating at a global voltage and frequency reference of 311 V and 50 Hz, respectively, with N ¼ 4 gridforming VSCs. Since each VSC is of equal capacity of 10 kVA, the active power droop coefficients mk are equal and, hence, active power will be shared equally. More details on the system and control parameters and control structure can be referred here [43]. As shown in Fig. 8.5, the kth distributed generators consist of a DC source (e.g., renewable energy or ESSs), VSC, LCL filter, and a controller using local measurements. In the system shown in Fig. 8.5 comprising of N agents, each communication digraph is represented via edges to constitute an adjacency matrix A ¼ ½akj ˛ RNN , where the communication weights are given by akj > 0; if jk ; jj ˛ E, where E is an edge connecting two nodes, with jk and jj being the local and neighboring node, respectively. Otherwise, akj ¼ 0. Nk ¼ j jk ; jj ˛ E denotes the set of all neighbors of kth agent. Further, the indegree matrix ZP in ¼ diag{zin } is a diagonal matrix with its elements given by zin ¼ akj. Further, the Laplacian matrix L is defined as j ˛ Nk
L ¼ Zin A. To improve their performance, neighboring VSCs’ measurements, which are transmitted to the local VSC and viceversa, are used in a cooperative k
secondary controller to regulate their respective bus’ average voltage V g and frequency uk. The control objectives of the cooperative controller can be mathematically represented as lim uk ðtÞ ¼ u ; lim V g ðtÞ ¼ V c k ˛ N k
t/N
t/N
(8.5)
where u and V denote the global reference for frequency and voltage, respectively. Detailed control equations of cooperative secondary controller in AC MG can be referred from Refs. [44]. To achieve proportionate active power sharing alongwith frequency restoration, the primary layer droop control is modified into uk ðtÞ ¼ u mk Pk ðtÞ Pref ðtÞ (8.6) k where mp , Pk , and Pref k denote the active power droop coefficient, measured active power, and secondary control active power reference in kth agent.
214 Control of Power Electronic Converters and Systems
Basically, Pref k compensates for the error introduced by the droop coefficient in (8.6). This is done using X ref P_k ðtÞ ¼ k1 ðu uk ðtÞÞ þ k2 akj ðxj ðtÞ xk ðtÞÞ (8.7) j ˛ Nk
with k1 and k2 being positive variables and x ¼ mP. Further, Nk denotes the set of neighbors of kth agent. Substituting (8.7) in (8.6), we obtain LuðtÞ ¼ 0. However, the objectives in (8.5) can be misconstrued in the presence of cyber attacks on the frequency signal in kth agent using ufk ðtÞ ¼ uk ðtÞ þ kuak
(8.8)
where k ¼ 1 denotes the presence of an attack element uak in kth agent, or 0 otherwise. Further, these attacks can be conducted in a coordinated manner to deceive the system operator using _ ¼ ðLuðtÞ þ ua Þ uðtÞ
(8.9)
where u and ua denote column matrices of the measured frequency and attack signal for N VSCs, respectively. Considering the attack model in (8.9), the attack can be termed as _ 1. coordinated attack, if uðtÞ ¼ 0. Such attacks always lead to a stable and feasible solution, thereby satisfying the objectives in (8.5). _ 2. noncoordinated attack, if uðtÞs 0. They disregard the objectives in (8.5). Based on the definition of coordinated attacks, it can be concluded that PN ua ¼ 0 for cooperative synchronization holds true. Conversely, Pk¼1 N a k¼1 u s 0 for noncoordinated attacks. In Fig. 8.10, when an attack of ua ¼ 2p{0.1, 0, 0.1, 0} rad/s is introduced at t ¼ 0.5 s, frequency and active power converge back to the corresponding references, as defined in the control objectives. As defined above, all the necessary conditions are met, which certifies it as a coordinated attack. However, at t ¼ 1.5 s, the attacker maintains this discretion and increases one attack element in ua ¼ 2p{10, 0, 0.1, 0} rad/s. As a result, it can be seen that u1 immediately goes outside the boundary of operation [49.5, 50.5] Hz (as highlighted in Fig. 8.10) defined for the MG. As the frequencies reach close to the aforementioned threshold, it could unnecessarily lead to the activation of protective relays, which could cause shutdown of the whole MG. Finally, the resilience capability of the signal reconstruction strategy in Fig. 8.8 is studied for test case II using only one trustworthy frequency input. In Fig. 8.11, it can be seen that the detection mechanism for the attack model in (8.9) activates the proposed signal reconstruction strategy at t ¼ 0.5 s, which immediately brings the system to follow the control objectives despite the presence of attacks. Following the coordinated attack, it also bypasses the
(A)
(B)
FIGURE 8.10 Case study for N ¼ 4 gridforming VSCs in Fig. 8.5d(a) Frequency, (b) Average Voltages and (c) Active Power. Attacker conducts a coordinated attack first to deceive the system operator and then conducts a noncoordinated attack resulting in operation outside the allowable range.
Cyber security in power electronic systems Chapter  8
(C)
215
(B)
FIGURE 8.11 Case study for N ¼ 4 gridforming VSCs in Fig. 8. 5d(a) Frequency, (b) Active Power. The eventdriven resilient controller mitigates the coordinated attack immediately based on the detection philosophy proposed in Ref. [40].
216 Control of Power Electronic Converters and Systems
(A)
Cyber security in power electronic systems Chapter  8
217
impact of noncoordinated attack at t ¼ 1.5 s. Hence, this strategy can be effectively used to mitigate cyber attacks to maximum degree of resiliency. As it is evident from the impact and behavior of the system, it is important to establish a generic strategy for detection and mitigation of attacks to provide a resilient networked control norm. In addition, network observability needs to be accommodated to develop a robust control mechanism for cyber attacks to improve security in the modern electric grid.
8.6 Conclusions and future challenges In this chapter, from power electronic system point of view, the challenges and vulnerabilities associated with controlling modern gridtied power converters due to cyber attacks were analyzed. Initially, the basic concepts of local control used by VSCs in different fields and applications were updated. Then, we provided an overview of possible attacks and their effect on interconnected converters. A comprehensive tutorial is given about the vulnerable points in the control and communication layer used for controlling gridforming converters. Two test cases considering gridforming VSCs are conducted using these attack models as proof of concept to illustrate the implications of cyber attacks. Cyber attacks with limited complexity have been shown to result in system failure, cause disruption, and potentially harm to consumer appliances. To resolve these issues, resilient control systems need to be established as a potential scope of work to reduce the effect of cyber attacks on the electrical grid. The design of resilient strategies includes a clear understanding of the layer of control and security. Ideally, eliminating the communication channel to promote localized control strategies will enable the protection of the electronic power converters. However, from an efficiency viewpoint, this notion elevates up as an overstatement. Therefore, it is important to restrict the cyber physical interactions to a minimum synergy by targeting a manageable system output tradeoff. Accommodating these perspectives, designing resilient technology and planning a line of defense against cyber attacks is a new objective to improve the security and efficiency of the power electronic converters in the electric grid.
References [1] [2]
[3]
B.V. Mathiesen, H. Lund, K. Karlsson, 100% renewable energy systems, climate mitigation and economic growth, Appl. Energy 88 (2) (2011) 488e501. F. Blaabjerg, R. Teodorescu, M. Liserre, A.V. Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (October 2006) 1398e1409. J. Rocabert, A. Luna, F. Blaabjerg, P. Rodriguez, Control of power converters in AC microgrids, IEEE Trans. Power Electron. 27 (11) (November 2012) 4734e4749.
218 Control of Power Electronic Converters and Systems [4] T. Dragicevic, X. Lu, J.C. Vasquez, J.M. Guerrero, DC microgridsPart II: a review of power architectures, applications, and standardization issues, IEEE Trans. Power Electron. 31 (5) (May 2016) 3528e3549. [5] Y. Mo, T.H.J. Kim, K. Brancik, D. Dickinson, H. Lee, A. Perrig, B. Sinopoli, “Cybere physical security of a smart grid infrastructure, Proc. IEEE 100 (1) (2011) 195e209. [6] F. Pasqualetti, F. Dorfler, F. Bullo, Controltheoretic methods for cyberphysical security: geometric principles for optimal crosslayer resilient control systems, IEEE Contr. Sys. Mag. 35 (1) (2015) 110e127. [7] S.A. Baker, N. Filipiak, K. Timlin, In the Dark: Crucial Industries Confront Cyber Attacks, McAfee, 2011. Incorporated. [8] F. Pasqualetti, F. Do¨rfler, F. Bullo, Attack detection and identification in cyberphysical systems, IEEE Trans. Automat. Contr. 58 (11) (2013) 2715e2729. [9] J.M. Carrasco, L.G. Franquelo, J.T. Bialasiewicz, E. Galvan, R.C. PortilloGuisado, M.A.M. Prats, J.I. Leon, N. MorenoAlfonso, Powerelectronic systems for the grid integration of renewable energy sources: a survey, IEEE Trans. Ind. Electron. 53 (4) (June 2006) 1002e1016. [10] C.A. Hill, M.C. Such, D. Chen, J. Gonzalez, W.M. Grady, Battery energy storage for enabling integration of distributed solar power generation, IEEE Trans. Smart Grid 3 (2) (June 2012) 850e857. [11] M. Yilmaz, P.T. Krein, Review of battery charger topologies, charging power levels, and infrastructure for plugin electric and hybrid vehicles, IEEE Trans. Power Electron. 28 (5) (May 2013) 2151e2169. [12] T. Dragicevic, J. Guerrero, J. Vasquez, D. Skrlec, Supervisory control of an adaptivedroop regulated dc microgrid with battery management capability, IEEE Trans. Power Electron 29 (2) (2014) 695e706. [13] J. Schonberger, R. Duke, S. Round, DCBus signaling: a distributed control strategy for a hybrid renewable nanogrid, IEEE Trans. Ind. Electron. 53 (5) (Oct. 2006) 1453e1460. [14] D. Boroyevich, I. Cvetkovic, D. Dong, R. Burgos, F. Wang, F. Lee, Future electronic power distribution systems a contemplative view, in: 2010 12th International Conference on Optimization of Electrical and Electronic Equipment, IEEE, May 2010, pp. 1369e1380. [15] “Ieee Standard for Interconnection and Interoperability of Distributed Energy Resources with Associated Electric Power Systems Interfaces,” IEEE Std 15472018 (Revision of IEEE Std 15472003), April 2018, pp. 1e138. [16] S.A. Boyer, SCADA: Supervisory Control and Data Acquisition, International Society of Automation, 2009. [17] S. Sahoo, S. Mishra, J.C. Peng, T. Dragicevic, A stealth cyber attack detection strategy for dc microgrids, IEEE Trans. Power Electron. 34 (8) (2019) 8162e8174, https://doi.org/ 10.1109/TPEL.2018.2879886. [18] S. Sahoo, J.C. Peng, A. Devakumar, S. Mishra, T. Dragicevic, “On detection of false data in cooperative dc microgridsV”a discordant element approach, IEEE Trans. Ind. Electron. 67 (8) (2020) 6562e6571. [19] S. Sahoo, J.C. Peng, S. Mishra, T. Dragicevic, Distributed screening of hijacking attacks in dc microgrids, IEEE Trans. Power Electron. 35 (7) (2020) 7574e7582. [20] S. Sahoo, T. Dragicevic, F. Blaabjerg, Resilient operation of heterogeneous sources in cooperative dc microgrids, IEEE Trans. Power Electron. 35 (12) (2020) 12601e12605, https://doi.org/10.1109/TPEL.2020.2991055.
Cyber security in power electronic systems Chapter  8 [21]
[22]
[23] [24]
[25] [26] [27] [28] [29] [30] [31]
[32] [33] [34] [35] [36] [37] [38]
[39]
219
W. Wu, Y. Liu, Y. He, H.S.H. Chung, M. Liserre, F. Blaabjerg, Damping methods for resonances caused by lclfilterbased currentcontrolled gridtied power inverters: an overview, IEEE Trans. Ind. Electron. 64 (9) (Sept 2017) 7402e7413. L. Harnefors, X. Wang, A.G. Yepes, F. Blaabjerg, Passivitybased stability assessment of gridconnected vscs: an overview, IEEE J. Em. Sel. Topics Power Electron. 4 (1) (March 2016) 116e125. D.M. Vilathgamuwa, P.C. Loh, Y. Li, Protection of microgrids during utility voltage sags, IEEE Trans. Ind. Electron. 53 (5) (Oct 2006) 1427e1436. K.D. Brabandere, B. Bolsens, J.V. den Keybus, A. Woyte, J. Driesen, R. Belmans, A voltage and frequency droop control method for parallel inverters, IEEE Trans. Power Electron. 22 (4) (July 2007) 1107e1115. Y. Liu, P. Ning, M.K. Reiter, False data injection attacks against state estimation in electric power grids, ACM Trans. Inf. Syst. Secur. 14 (1) (2011) 13. F. Callegati, W. Cerroni, M. Ramilli, Maninthemiddle attack to the https protocol, IEEE Security and Privacy 7 (1) (2009) 78e81. A.D. Wood, J.A. Stankovic, Denial of service in sensor networks, Computer 35 (10) (2002) 54e62. G. Liang, J. Zhao, F. Luo, S.R. Weller, Z.Y. Dong, A review of false data injection attacks against modern power systems, IEEE Trans. Smart Grid 8 (4) (2016) 1630e1638. Y.S. Shiu, S.Y. Chang, H.C. Wu, S.C.H. Huang, H.H. Chen, Physical layer security in wireless networks: a tutorial, IEEE Wireless Comm. 18 (2) (2011) 66e74. A. Stefanov, C.C. Liu, Cyberpower system security in a smart grid environment, in: 2012 IEEE PES Innovative Smart Grid Technologies (ISGT), IEEE, 2012, pp. 1e3. S. Sahoo, T. Dragicevic, F. Blaabjerg, “Cyber security in control of gridtied power electronic convertersV“challenges and vulnerabilities, IEEE J. Emerg. Select. Topics Power Electron. (2019), pp. 1e1. B. Satchidanandan, P.R. Kumar, “Dynamic watermarking: active defense of networked cyberV“physical systems, Proc. IEEE 105 (2) (2017) 219e240. K. Pan, A. Teixeira, M. Cvetkovic, P. Palensky, Cyber risk analysis of combined data attacks against power system state estimation, IEEE Trans. Smart Grid 10 (3) (2018) 3044e3056. O.A. Beg, L.V. Nguyen, T.T. Johnson, A. Davoudi, Signal temporal logicbased attack detection in dc microgrids, IEEE Trans. Smart Grid 10 (4) (2018) 3585e3595. E. Handschin, F.C. Schweppe, J. Kohlas, A. Fiechter, Bad data analysis for power system state estimation, IEEE Trans. Power Apparatus Syst. 94 (2) (1975) 329e337. O.A. Beg, T.T. Johnson, A. Davoudi, Detection of falsedata injection attacks in cyberphysical dc microgrids, IEEE Trans. Ind. Informat. 13 (5) (2017) 2693e2703. S. Sundaram, C.N. Hadjicostis, Distributed function calculation via linear iterative strategies in the presence of malicious agents, IEEE Trans. Automat. Contr. 56 (7) (2010) 1495e1508. K. Manandhar, X. Cao, F. Hu, Y. Liu, Detection of faults and attacks including false data injection attack in smart grid using kalman filter, IEEE Trans. Control Network Syst. 1 (4) (2014) 370e379. S. Sundaram, B. Gharesifard, Consensusbased distributed optimization with malicious nodes, in: 2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton), IEEE, 2015, pp. 244e249.
220 Control of Power Electronic Converters and Systems [40] S. Sahoo, Y. Yang, F. Blaabjerg, Resilient synchronization strategy for ac microgrids during cyber attacks, IEEE Trans. Power Electron. 36 (1) (2021) 73e77, https://doi.org/10.1109/ TPEL.2020.3005208. [41] S. Sahoo, S. Mishra, An adaptive eventtriggered communicationbased distributed secondary control for dc microgrids, IEEE Trans. Power Electron. 9 (6) (2017) 6674e6683. [42] T. Dragicevic, Model predictive control of power converters for robust and fast operation of ac microgrids, IEEE Trans. Power Electron. 33 (7) (2018) 6304e6317. [43] Q. Shafiee, J.M. Guerrero, J.C. Vasquez, “Distributed secondary control for islanded microgridsea novel approach, IEEE Trans. Power Electron. 29 (2) (2013) 1018e1031. [44] R. Rana, S. Sahoo, S. Mishra, J.C. Peng, Performance validation of cooperative controllers in autonomous ac microgrids under communication delay, in: 2019 IEEE Power Energy Society General Meeting (PESGM), 2019, pp. 1e5.
Chapter 9
Advanced modeling and control of voltage source converters with LCL filters Bochen Liu1, Dao Zhou1, Frede Blaabjerg1 1
Department of Energy Technology, Aalborg University, Aalborg, Denmark
9.1 Introduction In this chapter, various switched models of the LCLfiltered voltage source converters (VSCs) in abc, ab, and dqreference frames are firstly introduced in Section 9.2. Especially, the original sinusoidal AC voltage and currents are transformed to the constant DC components in dq frame, together with the rotating grid phasor. This can facilitate the simplification of the steadystate grid current control and help to build the continuous timeindependent averaged model in dq frame. However, this model can only predict the converter openloop behavior with a given switching function, and it cannot model the crosscoupling of different harmonic components. By addressing this harmonic coupling issue and considering unbalanced grid situations, the dynamic phasor model is introduced, where the averaged values of different harmonic components are included and thus is also called generalized averaged model (GAM). Next, in order to further describe the transient response of each harmonic component in the VSC, which can be regarded as a linear timeperiodic (LTP) system, the harmonic state space (HSS) model can be used. Besides, this model is developed by imitating the linear timeinvariant (LTI) system and therefore the already developed control and analysis methods for LTI systems can also be applied. Afterward, in terms of the converter control, a fundamental dqframee based DClink voltage and alternating current control are firstly introduced in Section 9.3. Therein, each variable is expressed by the complex form and is ready to further analyze the system performance (e.g., stability and harmonic effect) using the tools of complex space factors and complex transfer function. In order to tackle the resonance phenomenon caused by LCL filter, different damping methods including passive damping (PD) and active damping (AD) Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000159 Copyright © 2021 Elsevier Ltd. All rights reserved.
221
222 Control of Power Electronic Converters and Systems
are introduced to suppress the peak resonance. Moreover, an alternating current control procedure is presented for unbalanced grid situations, where the gridphase extraction and positive/negativesequence decoupling submodules are explained in detail. Besides, considering the widely varying grid impedance in increasing distribution generation systems, the converter stability analysis in weak grids is conducted in Section 9.4. By integrating the capacitor current feedback into the system control so as to damp the LCL filter resonance, the system control procedure is firstly illustrated in ab and dq frame. On this basis, the equivalent VSC output impedance can be derived and the converter stability analysis can be implemented. Simulation results are presented to validate the analysis.
9.2 Modeling of the VSCs with LCL filters Firstly, which is developed in DCeDC power converters, the state space averaging approach [1] is often used to filter out the switching ripples and thus an averaged nonlinear but timeinvariant model can be obtained. However, this approach cannot directly be applied to VSCs due to the timeperiodic operating trajectory of the AC system [2]. Hence, a reference frame transformation technique (i.e., Clarke transformation and Park transformation) is introduced to transfer the static threephase system to a rotating twophase system, where the AC operating trajectory is transformed as the DC operating point. On the other hand, in an unbalanced power grid, more frequency terms need to be considered due to the positivesequence and negativesequence components of the AC system. By addressing this, a generalized averaging model (GAM [3,4], also known as the dynamic phasor model [5e7]) and the HSS model [8e10] can be developed to transform the discrete switching events into a continuous model, which is much more easier to analyze. Due to the page limitation, only the large signal model using these three modeling methods is illustrated. But the similar modeling procedure can be applied to derive the smallsignal model for linearizing at an equilibrium operating point.
9.2.1 Modeling methods in balanced threephase systems A typical VSC with LCL filter is shown in Fig. 9.1 and the threephase system equations can be obtained from the circuit 8 dif udc > > þ Rf if ¼ S ucf un I > Lf > dt 2 > > > > > > ducf > > > < Cf dt ¼ if ig (9.1) > > dig 0 > > Lg þ Rg ig ¼ ucf ug þ un un I > > dt > > > > > > > : C dudc ¼ idc ST if dt
Advanced modeling and control Chapter  9
223
Rdc idc +
Q1
Q3
Q5 ifa ifb ifc
udc C Q2

Q4
Q6
Lf +Rf
Lg+Rg
ucfa ucfb ucfc
iga uga igb ugb igc ugc
un'
Cf
un
FIGURE 9.1 Typical structure of threephase gridtied voltage source converter with LCL filter.
where ig ¼ [iga igb igc]T is the gridside current vector, if ¼ [ifa ifb ifc]T the inverterside current vector, icf ¼ [icfa icfb icfc]T the filtering capacitor current vector, ucf ¼ [ucfa ucfb ucfc]T the capacitor voltage vector, ug ¼ [uga ugb ugc]T the grid voltage vector, S ¼ [Sa Sb Sc]T the control signal vector (Sa Sb Sc ¼ 1, 1), un and u0n are the voltages at neutral points, and I is a unit column vector defined as [1 1 1]T. Due to the Kirchhoff’s laws, there are 8 > < iga þ igb þ igc ¼ 0 icfa þ icfb þ icfc ¼ 0 (9.2) > : ifa þ ifb þ ifc ¼ 0 Combining Eqs. (9.1) and (9.2), the neutral point voltages un and u0n can be derived as 8 udc 1 > > < un ¼ 6 ðSa þ Sb þ Sc Þ 3 ðucfa þ ucfb þ ucfc Þ (9.3) > > : u0 ¼ udc ðS þ S þ S Þ 1 ðu þ u þ u Þ a b c ga gb gc n 3 6 In balanced threephase power grids, uga þ ugb þ ugc ¼ 0 and uca þ ucb þ ucc ¼ 0, and thus Eq. (9.3) is simplified to be udc ðSa þ Sb þ Sc Þ un ¼ u0n ¼ (9.4) 6 According to Eq. (9.4), the new expressions for balanced threephase systems can be written as 8 > dif udc > > > Lf þ Rf if ¼ S ucf un I > > dt 2 > > > > > du > > Cf cf ¼ if ig < dt (9.5) > di > g > Lg þ Rg ig ¼ ucf ug > > > dt > > > > du > dc T > > : C dt ¼ idc S if
224 Control of Power Electronic Converters and Systems
Eq. (9.5) is the switched model of the VSC with LCL filters in balanced threephase power systems. Without loss of generality, applying a moving average technique to an nlength signal vector x ¼ [x1, x2, x3 . xn], the result is Z t Z t Z t Z t 1 x1 ðsÞds; x2 ðsÞds; x3 ðsÞds . xn ðsÞds CxDavg ¼ Tsw tTsw tTsw tTsw tTsw (9.6) Therein, Tsw is the width of the averaged time window, and it is usually equal to one switching period in power electronic converters. Applying moving average technique to the switched model in Eq. (9.5), the averaged model of the VSC with LCL filter can be derived as 8 > > > dCif Davg Cudc Davg > > Lf þ Rf Cif Davg ¼ CSDavg Cucf Davg Cun DavgI > > dt 2 > > > > > > dCucf Davg > > ¼ Cif Davg Cig Davg < Cf dt (9.7) > dCig Davg > > > Lg þ Rg Cig Davg ¼ Cucf Davg Cug Davg > > dt > > > > > dCudc Davg > > > ¼ Cidc Davg CST Davg Cif Davg > :C dt It should be noted that by assuming that the DC output voltage udc and the converterside current if hardly change in one switching period, the items Cudc SDavg and CST if Davg in Eq. (9.5) are approximately equal to Cudc DavgCSDavg and CST Davg Cif Davg in Eq. (9.7), respectively. In this way, the variables in Eq. (9.5) are transformed to be continuous, and thus Eq. (9.7) is named as continuoustime averaged model of VSC converter. However, the model in Eq. (9.7) is still nonlinear and time varying due to the AC currents (i.e., if, ig) and AC voltages (i.e., ucf, ug). In order to linearize the averaged model in Eq. (9.7), a coordinate transformation technique is usually adopted. In balanced power systems, the AC voltages or currents contain no zerosequence components. Hence, by using Clarke transformation, the threephase abc frame can be converted into the twophase static abreference frame, i.e., 2 3 2 3 1 1 xa 1 6 7 xa 2 2 26 7 6 7 (9.8) ¼ 6 pﬃﬃﬃ pﬃﬃﬃ 7 4 xb 5 34 xb 3 35 xc 0 2 2 ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} Tab
where xa, xb, xc represent the threephase voltages or currents and xa, xb are the converted variables in abreference frame. The constant coefficient 2/3 is used
Advanced modeling and control Chapter  9
225
pﬃﬃﬃﬃﬃﬃﬃﬃ pﬃﬃﬃ to keep the same peak value in both coordinate frames ( 2 3 or 2=3 for rootmeansquare and powerinvariant scaling, respectively). Although the threephase quantities are transformed into twophase quantities, xa and xb are still timedependent variables. Thus, a Park transformation is needed to further transform the static twophase ab frame to the synchronous rotating dq frame, which is cos ut sin ut xd xa ¼ (9.9) sin ut cos ut xq xb ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} Tdq
Therein, xd, xq are the transformed components in dqreference frame, and u is the angular speed of the rotating frame, which can be obtained by grid synchronization technique. Taking the voltage vector ug in Eq. (9.1) as an example, if Vg is defined as the peak value of the grid phase voltage, the threephase voltages are ug ¼ ½uga ugb ugc T ¼ Vg ½cosðutÞ cosðut 2p=3Þ cosðut þ 2p=3ÞT (9.10) Then the dq transformation can be conducted as ug;dq ¼ ½ugd ugq T ¼ Tdq $Tab $ug ¼ ½Vg 0T
(9.11)
Consequently, the three static variables in abcreference frame can be represented by two DC variables in the dqreference frame. By applying the Clarke transformation in Eq. (9.8) to the threephase static switched model Eq. (9.5), the expressions of the VSC with LCL filter in ab coordinate can be achieved as 8 dif;ab udc > > þ Rf if;ab ¼ Sab ucf;ab > Lf > dt 2 > > > > > > ducf;ab > > > < Cf dt ¼ if;ab ig;ab (9.12) > > di g;ab >L > þ Rg ig;ab ¼ ucf;ab ug;ab g > > dt > > > > > > > : C dudc ¼ idc 3ST if;ab 2 ab dt where if,ab ¼ [ifa ifb]T, ig,ab ¼ [iga igb]T, ucf,ab ¼ [ucfa ucfb]T, ug,ab ¼ [uga ugb]T and Sab ¼ [Sa Sb]T, all of which are converted into twophase static ab components. As the control signals are symmetrical in a balanced power system, the term unI in Eq. (9.5) is omitted by ab transformation. Similarly, based on Eq. (9.12) and the Park transformation Eq. (9.9), the expressions in the rotating dqreference frame can be derived as
226 Control of Power Electronic Converters and Systems
8 > dif;dq udc > > Lf þ Rf if;dq ¼ uLf Jif;dq þ Sdq ucf;dq > > dt 2 > > > > > > ducf;dq > > > < Cf dt ¼ uCf Jucf;dq þ if;dq ig;dq > > dig;dq > > þ Rg ig;dq ¼ uLg Jig;dq þ ucf;dq ug;dq Lg > > dt > > > > > > dudc 3 > > ¼ idc STdq if;dq :C 2 dt
(9.13)
where if,dq ¼ [ifd ifq]T, ig,dq ¼ [igd igq]T, ucf,dq ¼ [ucfd ucfq]T, ug,dq ¼ [ugd vgq]T and Sdq ¼ [Sd Sq]T, all of which are converted into twophase synchronous dq components. J is an assistant matrix equaling to 0 1 J¼ (9.14) 1 0 Applying the moving average technique to Eq. (9.13), the averaged dqframe model of the VSC for a balanced threephase system can thus be obtained, which is 8 > > dCif;dq Davg Cudc Davg > > Lf þ Rf Cif;dq Davg ¼ uLf JCif;dq Davg þ CSdq Davg Cucf;dq Davg > > > dt 2 > > > > > > dCucf;dq Davg > > ¼ uCf JCucf;dq Davg þ Cif;dq Davg Cig;dq Davg > < Cf dt > > dCig;dq Davg > > þ Rg Cig;dq Davg ¼ uLg JCig;dq Davg þ Cucf;dq Davg Cug;dq Davg > > Lg > dt > > > > > > dCudc Davg 3 > > ¼ Cidc Davg CSTdq Davg Cif Davg > :C 2 dt (9.15) Eq. (9.15) is the averaged model of the VSC with LCL filter in dqreference frame. Compared to the original averaged timevariant model Eq. (9.7) in abcreference frame, the AC variables (e.g., if, ig) are transferred into DC operating variables (e.g., if,dq, ig,dq), and thus the dqframe model becomes nonlinear and time invariant. It should be noted that the crosscoupling between the daxis and qaxis components also appears (i.e. uLfJif,dq, uLgJig,dq) due to the Park transformation.
9.2.2 Modeling methods in unbalanced threephase systems As discussed above, the balanced threephase power system can be modeled by averaging the variables within one switching period. However, in unbalanced threephase system, more frequency coupling terms need to be
Advanced modeling and control Chapter  9
227
considered. Besides, the state averaged model can only predict the converter dynamics below half the switching frequency, and the model is restricted to the assumption of small switching ripples. In order to extend these limitations, a multifrequency model is introduced to represent the behavior of converters containing AC stages, especially to handle the highorder harmonics. Two modeling methods of the VSC with LCL filters will be introduced for unbalanced threephase systems, i.e., GAM (also called dynamic phasor model) and HSS model.
9.2.2.1 Generalized averaged model The GAM was firstly proposed to capture the switchingfrequency component of DCeDC converters. The GAM is based on the fact that a periodic waveform x(t) can be represented by infinite complex Fourier series, which is xðtÞ ¼
þN X
CxDk ðtÞ$ejkut
(9.16)
k¼N
The coefficient of the kth harmonic is defined by Z 1 T CxDk ðtÞ ¼ xðsÞ$ejkus ds T tT
(9.17)
where T ¼ 2p/u and u is the constant fundamental frequency pulsation. It can be found that k ¼ 0 in Eq. (9.17) corresponds to the moving average operator in Eq. (9.6) with T equaling the switching period Tsw. There are two fundamental properties in GAM, as can be expressed by d d CxDk ðtÞ ¼ C xDk ðtÞ jkuCxDk ðtÞ dt dt X Cx$yDk ðtÞ ¼ CxDki ðtÞ$CyDi ðtÞ
(9.18) (9.19)
i
Eqs. (9.18) and (9.19) describe the properties of the derivative of the moving average and the variables coupling in GAM, respectively. For the convenience of analysis, the Fourier coefficients are simplified by CxDk ðtÞ / CxDk ;
Cx$yDk ðtÞ/Cx$yDk
(9.20)
Furthermore, the following relationships between the real (denoted by “R”) and imaginary (denoted by “I”) parts of the kth coefficient and the (k)th coefficient can be readily obtained and they are satisfied for arbitrary harmonic component 8 Z 1 t > > > CxD ¼ xðsÞcosðkusÞds ¼ CxDkR > < kR T tT (9.21) Z t > > 1 > > xðsÞsinðkusÞds ¼ CxDkI : CxDkI ¼ T tT
228 Control of Power Electronic Converters and Systems
Therefore, if only firstorder harmonic (i.e., fundamental component) is studied, the actual signal can be modeled as xðtÞ z CxD1 ejut þ CxD1 ejut ¼ 2½CxD1R cosðutÞ CxD1I sinðutÞ
(9.22)
On the basis of Eqs. (9.17)e(9.21) and combining with the original switched model (cf. Eq. 9.1) of VSC with LCL filter, the kthorder harmonic component contained in the AC currents/voltages can be obtained with 8 dCif Dk 1 > > > Lf þ Rf Cif Dk ¼ jkuLf Cif Dk þ Cudc SDk Cucf Dk Cun IDk > > 2 dt > > > > > > dCucf Dk > > ¼ jkuCf Cucf Dk þ Cif Dk Cig Dk > Cf < dt (9.23) > > dCig Dk > 0 > > > Lg dt þ Rg Cig Dk ¼ jkuLg Cig Dk þ Cucf Dk Cug Dk þ C un un IDk > > > > > > > > : C dCudc Dk ¼ Cidc Dk CST if Dk dt where the symbol meanings are the same as Eq. (9.1). Next, by setting k to different values (e.g., 1, 5, 7, 11, 13 . for AC variables and 0, 2, 6, 12 . for DC variables), the corresponding kthorder harmonic can be solved using Eq. (9.23), and then the GAM of the VSC with LCL filter can be achieved by adding up these harmonic components. As a result, if up to hthorder harmonics are considered, the final results are 8 h > P > > > d Cif Dk h h h h h > X X X X > 1X > k¼1 > Cif Dk ¼ ð jkuLf Cif DkÞ þ Cudc SDk Cucf Dk Cun IDk þ Rf L > f > > 2 dt > k¼1 k¼1 k¼1 k¼1 k¼1 > > > > > > h P > > > > h h h X X > d k¼1 Cucf Dk X > > > Cf ð jkuCf Cucf DkÞ þ Cif Dk Cig Dk ¼ > > dt < k¼1
k¼1
k¼1
> h > P > > d Cig Dk > h h h h h > X X X X X > k¼1 > > jkuL þ þ R L Ci D ¼ Ci D Cu D Cu D þ C un u0n IDk > g g g g g cf g k k k k > dt > > k¼1 k¼1 k¼1 k¼1 k¼1 > > > > > > h > > d P Cu D > dc k > h h X X > > > C k¼0 > Cidc Dk CST if Dk ¼ > : dt k¼0 k¼0 (9.24)
Clearly, if more harmonics are taken into account, the GAM will be more accurate, but in the meantime, the model will become more complex. In sight of this, a balance between the modeling accuracy and complexity needs to be
Advanced modeling and control Chapter  9
229
considered in practical applications, which usually depends on the harmonic components of interest. If the firstorder harmonic components (i.e., fundamental components) of the AC variables (e.g., ig ; if ) and the zerothorder components (i.e., average values) of the DC variables (e.g., udc ; iload ) are considered in Eq. (9.24), the modeling procedure of GAM is explained as the following. Firstly, substituting h ¼ 1 and h ¼ 0 into the first three expressions and the last expression in Eq. (9.24), respectively, the result is 8 dCif D1 1 > > > Lf þ Rf Cif D1 ¼ juLf Cif D1 þ Cudc SD1 Cucf D1 Cun ID1 > > 2 dt > > > > > > dCucf D1 > > ¼ juCf Cucf D1 þ Cif D1 Cig D1 > Cf < dt (9.25) > > dCig D1 > 0 > þ Rg Cig D1 ¼ juLg Cig D1 þ Cucf D1 Cug D1 þ C un un ID1 Lg > > > dt > > > > > > > : C dCudc D0 ¼ Cidc D0 CST if D0 dt
T T with Cif D1 ¼ ½Cifa D1; Cifb D1; Cifc D1 , Cig D1 ¼ Ciga D1; Cigb D1; Cigc D1 , T
Cucf D1 ¼ ½Cucfa D1; Cucfb D1; Cucfc D1 . Based on [11], the Fourier coefficients of the switching functions S ¼ [Sa Sb Sc]T can be obtained with 2 3 m 7 3 2 3 3 6 j 2 2 2 6 7 0 CSa D0 CSa D1 6 m 2p 7 7 7 6 7 7 6 6 6 (9.26) CSD0 ¼ 4 CSb D0 5 ¼ 4 0 5 CSD1 ¼ 4 CSb D1 5 ¼ 6 j ej 3 7 6 7 2 6 7 0 CSc D0 CSc D1 4 m 4p 5 j ej 3 2 where m is the modulation index. On this basis, the term Cudc SD1 can be expressed by 3 2 3 2 Cudc D0CSa D1 Cudc Sa D1 7 6 7 6 Cudc SD1 ¼ 4 Cudc Sb D1 5 ¼ 4 Cudc D0CSb D1 5 (9.27) Cudc Sc D1
Cudc D0CSc D1
and the term CSa ifa D0 in CST if D0 ¼ CSa ifa D0 þ CSb ifb D0 þ CSc ifc D0 can be derived as CSa ifa D0 ¼ CSa D1Cifa D1 þ CSa D1Cifa D1 ¼ 2ðCSa D1RCifa D1R þ CSa D1I Cifa D1I Þ
(9.28)
and similar results can be obtained for CSb ifb D0 and CSc ifc D0. Combining with Eqs. (9.21)e(9.28) and separating the real part and imaginary part for each expression, the final form of Eq. (9.25) can be derived as
230 Control of Power Electronic Converters and Systems 8 > > > > dCif D1R 1 1 X > > þ Rf Cif D1R ¼ uLf Cif D1I þ Cudc D0CSD1R Cucf D1R þ Cucfi D1R Lf > > 2 3 i¼a;b;c dt > > > > > > > > dCi D 1 1 X > > Lf f 1I þ Rf Cif D1I ¼ uLf Cif D1R þ Cudc D0CSD1I Cucf D1I þ Cucfi D1I > > 2 3 i¼a;b;c dt > > > > > > > > dCucf D1R > > ¼ uCf Cucf D1I þ Cif D1R Cig D1R Cf > > dt > > > > < dCucf D1I ¼ uCf Cucf D1R þ Cif D1I Cig D1I Cf dt > > > > > dCig D1R 1 X 1 X > > > Cugi D1R þ Cucfi D1R > Lg dt þ Rg Cig D1R ¼ uLg Cig D1I þ Cucf D1R Cug D1R þ 3 > 3 i¼a;b;c > i¼a;b;c > > > > > > 1 X 1 X > dCig D1I > > þ Rg Cig D1I ¼ uLg Cig D1R þ Cucf D1I Cug D1I þ Lg Cugi D1I þ Cucfi D1I > > dt 3 3 i¼a;b;c > i¼a;b;c > > > > > X > dCudc D0 > > > ¼ Cidc D0 2 C ðCSi D1RCifi D1R þ CSi D1I Cifi D1I Þ > > dt > i¼a;b;c > : (9.29)
Then Eq. (9.29) can be directly used to model the VSC with an LCL filter. Threephase voltages can become unbalanced and distorted because of the effect of the nonlinear loads and transient grid faults. In such case, one general way is to rewrite the AC voltages as
vabc ¼ ½va vb
8 3 2 > cos kut þ fþk > > 7 6 > > 7 6 > > > 6 7 > > 7 6 2 þN þN < X þk 7 X 6 k 0k þk 6 cos kut p þ f 7 vþk ¼ vc T ¼ þ v þ v V 3 abc abc abc 7 6 > 7 6 k¼1 k¼1 > > > 7 6 > > 7 6 > > 4 5 4 > þk > cos kut p þ f : 3 9 3 > cos kut þ fk > > 6 2 0k 3> >
7 > 7 6 cos kut þ f > 4 = 7 6 k p þ f cos kut 7 k 6 0k 6 0k 7 þV 6 7 þ V 4 cos kut þ f 5 3 6 > > 7
7 6 > cos kut þ f0k > > 2 4 k 5 > > cos kut p þ f ; 3 2
(9.30)
where superscripts þk, k, and 0k represent, respectively, the positive, negative, and zerosequence components of the kthorder harmonic of the
Advanced modeling and control Chapter  9
231
voltage vector vabc. Likewise, the unbalanced AC currents can be generally expressed as iabc ¼ ½ia ib ic T ¼
þN X k þ i iþk abc abc k¼1
2 3 39 > sin kut þ 4þk sin kut þ 4k > 6 6 7 7> > > 6 6 7 7> > > 6 6 7 7 2 4 þk 7 k 7= 6 6 p þ 4 sin kut þk 6 sin kut p þ 4 k 7þI 6 7 I 6 ¼ 3 3 6 7 7> > 6 6 7 7> k¼1 > >
> > 6 6 7 7> > 4 > > 4 2 4 5 5> > > þk k > > p þ 4 p þ 4 sin kut sin kut ; : 3 3 8 > > > > > > > > þ N X
f D1 > > L þ Rf Ci Cu D1 Cu f > f D1 ¼ juLf Cif D1 þ c D1 > dt 2 > > > > < dCu c D1 (9.33) ¼ juCf Cu Cf c D1 þ Cif D1 Cig D1 > dt > > > > > > > dCi D > : Lg g 1 þ Rg Ci D1 ¼ juLg Ci D1 þ Cu D1 Cu D1 g g c g dt
Eq. (9.33) is the combination of the positive and negativesequence GAMs, denoted by the superscripts þ and , respectively.
232 Control of Power Electronic Converters and Systems
9.2.2.2 Harmonic state space model A typical LTP system can be defined by _ ¼ AðtÞxðtÞ þ BðtÞuðtÞ xðtÞ yðtÞ ¼ CðtÞxðtÞ þ DðtÞuðtÞ
(9.34)
where x(t), u(t), y(t) are the, respectively, state, input, and output variables and A(t), B(t), C(t), D(t) are timeperiodic matrices, i.e., A(t) ¼ A(t þ T) and similarly for others. In order to apply the frequency separation property to LTP systems, instead of the expression Eq. (9.16) in GAM, another form of exponentially modulated periodic (EMP) signal expressed by the following is used: þN X
xðtÞ ¼ est
xk ejkut
(9.35)
k¼N
where s is a complex number in order to represent the transient evolution of the harmonic components. Then, by posing the variables x(t), u(t), y(t) in Eq. (9.34) as EMP signals and the timeperiodic matrices A(t), B(t), C(t), D(t) in Eq. (9.34) as complex Fourier series, i.e., 8 þN þN þN X X X > st > > xk e jkut ; yðtÞ ¼ est Yk e jkut ; uðtÞ ¼ est Uk e jkut ; > < xðtÞ ¼ e k¼N
k¼N
k¼N
þN þN þN þN > X X X X > > > Ak e jkut ; BðtÞ ¼ Bk e jkut CðtÞ ¼ Ck e jkut ; DðtÞ ¼ Dk e jkut : AðtÞ ¼ k¼N
k¼N
k¼N
k¼N
(9.36) the LTP system description can be transferred to
8 X þN þN þN X X > ðjkuþsÞt ðjkuþsÞt > ðjku þ sÞX e ¼ A X e þ Bkm Um eðjkuþsÞt > k km m > < k¼N k;m¼N k;m¼N þN þN þN > X X X > > > Yk eðjkuþsÞt ¼ Ckm Xm eðjkuþsÞt þ Dkm Um eðjkuþsÞt : k¼N
k;m¼N
k;m¼N
(9.37)
The m in Eq. (9.37) indicates the frequency crosscoupling among difference harmonic components. Considering the harmonic balance principle, which refers to the linear independency of every harmonic component, the summation symbol and the term eðjkuþsÞt can be canceled on the both sides of the expressions in Eq. (9.37), leading to
Advanced modeling and control Chapter  9
233
8 þN þN X X > > > sX ¼ A X þ Bkm Xm jkuXk k km m > < m¼N
m¼N
(9.38)
þN þN > X X > > > Ckm Xm þ Dkm Um : Yk ¼ m¼N
m¼N
Then, the HSS form can be obtained by including all harmonic component coefficients of Eq. (9.38) in a matrix, resulting in sX [ ðANÞXDBU (9.39) Y [ CXDDU Therein, A, B, C, and D are Toeplitz matrices constructed by respective Fourier coefficients of A(t), B(t), C(t), and D(t). For example, instead of the infinite Fourier series of Eq. (9.36), if up to hthorder components are considered in practice, A can be written as 2 3 A0 A1 / Ah 0 0 0 6A 1 0 0 7 6 1 1 1 1 7 6 7 6 « 1 A0 A1 1 Ah 0 7 6 7 6 7 (9.40) A ¼ 6 Ah 1 A1 A0 A1 1 Ah 7 6 7 6 0 7 A 1 A A 1 « h 1 0 6 7 6 7 0 1 1 1 1 A1 5 4 0 0 0 0 Ah / A1 A0 and similarly for B, C, and D. On the other hand, X, U, and Y in Eq. (4.39) denote respective Fourier coefficients of their complex Fourier series expansion, and N is the diagonal matrix of the sequence fjkugþh k¼h , resulting in 2
6 6 6 6 6 6 X¼6 6 6 6 6 4
Xh « X1 X0 X1 « Xh
3
2
6 7 6 7 6 7 6 7 6 7 6 7 7U¼6 6 7 6 7 6 7 6 7 4 5
Uh « U1 U0 U1 « Uh
3
2
6 7 6 7 6 7 6 7 6 7 6 7 7Y¼6 6 7 6 7 6 7 6 7 4 5
Yh « Y1 Y0 Y1 « Yh
3
2
6 7 6 7 6 7 6 7 6 7 6 7 7N¼6 6 7 6 7 6 7 6 7 4 5
3
jhu
7 7 7 7 7 7 7 7 7 7 7 5
1 ju 0 ju 1 jhu
(9.41)
In order to obtain the HSS model of the VSC in Fig. 9.1, the original switched model Eq. (9.1) is firstly transferred to the LTP form
234 Control of Power Electronic Converters and Systems
8 Rf udc 1 un > > if . ¼ if þ S ucf I > > Lf Lf 2Lf Lf > > > > > > > 1 1 > . > > < ucf ¼ Cf if Cf ig
(9.42)
> > Rg 1 1 1 > > un u0n I ig . ¼ ig þ ucf ug þ > > > Lg Lg Lg Lg > > > > > > > : : udc ¼ 1 idc 1 ST if C C
If the average and firstorder harmonic components are considered for each state variable, the HSS model of Eq. (9.42) can be derived based on Eqs. (9.39)e(9.41). For example, the differential expression of a phase converterside current is 3 2 Sa : udc 1 1 6 Rf 7 ifa ¼ $ifa þ (9.43) $4 ucfa 5 Lf Lf 2Lf Lf un and the HSS model of Eq. (9.43) can be derived as sIfa ¼ ðAifa Nifa ÞIfa DBifa Uifa
Ifa ¼ ½ Ifa1 Ifa0 Ifa1 T , Un1 Un0 Un1 T ,
Therein, the matrices Uifa ¼ ½ Sa1 Sa0 Sa1 Ucfa1 Ucfa0 Ucfa1 Nifa ¼ diag½ ju0 0 ju0 , and Aifa, Bifa are 2
Rf 6 Lf 6 6 6 Aifa ¼ 6 6 0 6 6 4 0
0
Rf Lf
0
3 2 Udc0 0 7 6 2Lf 7 6 7 6 7 6 Udc1 6 0 7 B ¼ ifa 7 6 2Lf 7 6 7 6 4 Rf 5 0 Lf
Udc1 2Lf
0
Udc0 2Lf
Udc1 2Lf
Udc1 2Lf
Udc0 2Lf
(9.44)
1 Lf
1 Lf
0
0
0
1 Lf
0
0
1 Lf
0
0
1 Lf
0
0
0
3 0 7 7 7 7 0 7 7 7 7 15 Lf (9.45)
Similar to Eq. (9.44), the other expressions in Eq. (9.42) can also be transformed into HSS models. Firstly, Eq. (9.42) can be rewritten as the LTP state space form (similar to Eq. 9.34), which is
Advanced modeling and control Chapter  9
2
Rf 6 6 Lf 2 3 6 6 1 . 6 i 6 f . 7 6 6 ucf 7 6 Cf 6 . 7¼6 6 ig 7 6 4 5 6 6 0 : udc 6 6 6 4 1 T S C
1 Lf
3 2 1 un S7 2Lf 7 Lf 6 72 7 i 3 6 6 7 f 6 0 0 76 6 76 ucf 7 6 7 76 þ 6 7 74 i 5 6 1 u u0 7 g n 6 Lg n 0 7 6 7 udc 6 7 4 7 5 0 0
0
0
1 Cf
1 Lg
0
Rg Lg
0
235
3 0
0
0
1 Lg
0
7 7 72 3 07 7 I 76 7 74 ug 5 07 7 idc 7 7 15 C (9.46)
Then, the HSS model of Eq. (9.46) can be obtained by following Eqs. (9.39)e(9.41). For a clear view, the HSS model of each phase in Eq. (9.46) is shown separately. By setting i ¼ a, b, c, there are sucfi ¼ Nucfi ucfi þ Bucfi Uucfi Z
2
ucfi1
6 6 ucfi ¼ 6 ucfi0 4 ucfi1
3
2
7 6 7 6 7 Ncfi ¼ 6 5 4
2
3
ju0
7 7 7 Uucfi 5
0 ju0
ifi1
3
2 1 6 7 6 ifi0 7 6 Cf 6 7 6 6 7 6 6 7 6 6 ifi1 7 6 6 7 ¼6 7 Bucfi ¼ 6 6 6i 7 6 6 gi1 7 6 6 7 6 6 7 4 6 igi0 7 4 5
1 Cf
1 Cf
igi1
3
1 Cf 1 Cf
7 7 7 7 7 7 7 7 7 7 1 5 Cf (9.47)
sigi ¼ ðAigi Nigi Þigi þ Bigi Uigi Z 2
2
igi1
6 6 igi ¼ 6 6 igi0 4 igi1
3 7 7 7 Aigi 7 5
Rg 6 6 Lg 6 6 6 6 ¼6 6 6 6 6 6 4
3
Rg Lg
7 7 2 7 ju0 7 7 6 7 6 7Nigi ¼ 6 7 6 7 4 7 7 Rg 7 5 Lg
2
Bigi
6 1 6 6 Lg 6 6 6 6 6 ¼6 6 6 6 6 6 6 4
Uigi ¼ ucfi1
3 7 7 7 7 5
0 ju0
(9.48)
3
1 Lg
1 Lg
1 Lg
1 Lg
1 Lg ucfi0
1 Lg
ucfi1
ugi1
1 Lg
ugi0
ugi1
7 7 7 7 7 7 7 7 7 7 7 7 7 1 7 7 Lg 5
un1 u0n1
un0 u0n0
un1 u0n1
T
236 Control of Power Electronic Converters and Systems sudc ¼ Nudc udc þ Budc Uudc 2
udc1
6 6 udc ¼ 6 udc0 4 udc1 2 1 6 6C 6 6 6 Budc ¼ 6 6 6 6 6 4
3
2
7 6 7 6 7Nudc ¼ 6 5 4
Z
3
ju0
7 7 7Uudc ¼ ½ idc1 idc0 idc1 ifa1 ifa0 ifa1 ifb1 ifb0 ifb1 ifc1 ifc0 ifc1 5
0 ju0
1 C 1 C
1 Sa0 C
1 Sa1 C
0
1 Sb0 C
1 Sb1 C
0
1 Sc0 C
1 Sc1 C
1 Sa1 C
1 Sa0 C
1 Sa1 C
1 Sb1 C
1 Sb0 C
1 Sb1 C
0
1 Sc0 C
0
1 Sa1 C
1 Sa0 C
0
1 Sb1 C
1 Sb0 C
0
1 Sc1 C
3 7 7 7 7 7 1 Sc1 7 7 C 7 7 7 5 1 Sc0 C 0
(9.49)
Similar procedure can be applied to derive the HSS model of the controller, which is not shown here due to page limitation.
9.2.3 Simulation examples Taking the GAM as an example, a simulation model using PLECS is built to validate the effectiveness of the modeling method, and the system parameters are shown in the following Table 9.1. Applying the conventional dqaxis current control (converter current feedback), which will be introduced in the next section, the steadystate working waveforms including the grid voltages, the gridside inductor current, and the converterside current are shown in Fig. 9.2.
TABLE 9.1 Parameters of 15 kW gridtied voltage source converters with LCL filter. Operating power Po (kW)
7.5
Grid peak phase voltage Ug (V)
311
DClink voltage Vdc (V)
650
Converterside inductor Lf (mH)
3
Capacitor branch Cf (uF)
25
Gridside inductor Lg (mH)
1.8
Grid voltage frequency fg (Hz)
50
Switching frequency fsw (kHz)
4
Converterside inductor equivalent resistance Rf (U)
0.15
Gridside inductor equivalent resistance Rg (U)
0.15
Advanced modeling and control Chapter  9
(A)
(B) ugb
ugc
uga
igb
igc
iga
(C)
ifb
237
ifa
ifc
(D) igq igd
FIGURE 9.2 Steadystate working waveforms of the voltage source converter with LCL filter with Po ¼ 7.5 kW. (A) Grid voltages, (B) converterside currents, (C) gridside currents, and (D) gridside dqaxis currents.
For the purpose of validating the GAM model, the simulated phase a current on the converter side and grid side is compared with the modeled a phase current from the firstorder GAM of the VSC with LCL filter, as shown in Fig. 9.3A and B, respectively. In the modeling procedure, Eq. (9.22) is used to generate the actual AC current from the calculated real and imaginary parts of the firstorder Fourier series coefficient in GAM. In the figure, Iga1(t), Ifa1(t) are the modeled currents and ifa, iga are the simulated currents. It can be seen that the firstorder generalized average model can track the fundamental component of the converter current and grid current properly.
(A)
Rf
Rg
0.15 : I fa1 t
(B) Rf
Rg
0.15 : I ga1 t
i fa
iga
FIGURE 9.3 Comparison of the simulated a phase current and the modeling a phase current from firstorder GAM under Rf ¼ Rg ¼ 0.15 U. (A) Converterside current comparison, (B) gridside current comparison.
238 Control of Power Electronic Converters and Systems
(A) I ga1 t
Rf
Rg
0.15 :
Rf
Rg
(B) I ga1 t
0
FIGURE 9.4 Modeling startup waveforms of a phase current under (A) Rf ¼ Rg ¼ 0.15 U and (B) Rf ¼ Rg ¼ 0 in the GAM of voltage source converter with LCL filter.
Besides, by setting the resistance parts in the converter inductor and grid inductor as zero, the startup waveform of the modeled a phase current is shown in Fig. 9.4B. As comparison, the startup waveform with nonzero resistance is shown in Fig. 9.4A. It can be seen that the resonance in the beginning is gradually debilitated in Fig. 9.4A due to the existence of the resistance. On the contrary, the resonance caused by LCL filter will retain in the current waveforms in Fig. 9.4B if the resistive parts are zero. From this point, the GAM can also be used to study the damping effect of different damping methods, which can be achieved by proper LCL filter design in hardware or AD control from software perspective.
9.3 Alternative current control of the VSCs with LCL filters There are mainly two types of L and LCL filters for the grid connection of VSCs. In terms of Ltype filter, one drawback is the need of higher value of inductance to decrease the harmonics of the line current, and another one is the necessity of higher switching frequency to achieve desired dynamic performance. If LCLtype filter is used, the size of passive elements and the switching frequency can be lower [12]. However, using LCL filter might cause instability problems at zero impedance occurred by its resonance frequency. In order to avoid this instability problem, the inherent, active, or PD methods can be employed [12e15].
239
Advanced modeling and control Chapter  9
9.3.1 Control in synchronous reference (dq) frame The power circuit of VSCs with the LCL filter is as shown in Fig. 9.1. Due to the fact that the impedance of the capacitor branch in LCL filter is negligibly small at grid frequency, the LCL filter converges to L filter at low frequencies. The merit of LCL filter becomes evident at higher frequencies, because the impedance of the capacitor branch decreases considerably and the highfrequency ripple attenuation extent rises from 20 to 60 dB/dec. Thus, the LC part of the LCL filter is in charge of the attenuation of the highfrequency current ripple and current controllers do not have to deal with the highfrequency ripple conduction. Besides, proportionaleintegral (PI)based current controllers have limited control bandwidth that primarily depends upon the sampling frequency of the system, and this does not enable the controllers to regulate the highfrequency oscillations. Therefore, PIbased current control can be done in accordance with the L filter approximation by neglecting the influence of the capacitor branch. On this basis, the system dqframe control of VSC is shown in Fig. 9.5. Consequently, the averaged dqframe model in Eq. (9.13) can be simplified as ! d i dq Vdc ! ! ! ðLf þ Lg Þ þ ðRf þ Rg Þ i dq ¼ d dq ! u g;dq jug ðLf þ Lg Þ i dq dt 2 (9.50) Note that in Eq. (9.50), complex vectors are introduced for the convenience ! ! u g;dq ¼ ugd þ jugq . of the analysis, i.e., i dq ¼ id þ jiq , d dq ¼ dd þ jdq , !
Rdc idc +
Q1
Q3
Q5
udc C Q2
Q4
udc
Q6
Lg+R g ifa Lf +R f ifb ifc uca ucb ucc Cf
iga uga igb ugb igc ugc ugabc
un'
un abc/dq
PWM
θ
PLL
abc/dq dq/αβ
igdq Gci(s)
igdq*
udc*
Igd*
udc
FIGURE 9.5 System dqframe control of voltage source converter in stiff grid.
240 Control of Power Electronic Converters and Systems
ug is the grid angular frequency and dd, dq are the dqframe transmission of averaged switching functions. In order to model a worstcase undamped scenario, the resistive part of the LCL filter components is often neglected in controller design phase, i.e., Rf and Rg in Eq. (9.50) equal zero, leading to ! d i dq Vdc ! ! ¼ d dq ! ðLf þ Lg Þ u g;dq jug ðLf þ Lg Þ i dq (9.51) dt 2 ! Then, the relationship of the sdomain currents I dq ðsÞ ¼ Id ðsÞ þ jIq ðsÞ ! and input variables D dq ðsÞ ¼ Dd ðsÞ þ jDq ðsÞ can be deduced from (9.51), which are ! I dq ðsÞ Vdc (9.52) PðsÞ ¼ ! ¼ 2ðs þ ju g ÞðLf þ Lg Þ D dq ðsÞ The model of Eq. (9.52) can be equivalently expressed in the transfer function matrix form as Id P1 ðsÞ P2 ðsÞ Vdc Dd Vdc Dd ¼ PðsÞ$ ¼ (9.53) $ 2 Dq 2 Dq Iq P2 ðsÞ P1 ðsÞ with s P1 ðsÞ ¼ 2 2 s þ ug ðLf þ Lg Þ
ug P2 ðsÞ ¼ 2 s þ u2g ðLf þ Lg Þ
(9.54)
In Eq. (9.53), a nonzero P2(s) indicates that there exists axis crosscoupling in the converter plant, which means that the input variable Dd or Dq will not only control the corresponding current (Id or Iq) but also disturb the other current (Iq or Id). Thus, a decoupling control is needed. Depending on the current sensor positions, there are two types of current controllers, i.e., converter current feedback (CCF) and grid current feedback (GCF). Accordingly, there are two types of decoupling schemes corresponding ! ! ! ! to i dq ¼ i f ;dq ¼ ifd þ jifq or i dq ¼ i g;dq ¼ igd þ jigq in Eq. (9.50). If the converter complex output voltages are introduced as Vdc ! ðDd þ jDq Þ u o ¼ uod þ juoq ¼ 2
(9.55)
the controller block diagrams of these two decoupling methods are as shown in Fig. 9.6. The superscript notation denotes the reference value. For the purpose of achieving zero displacement between the grid current and the grid voltage (i.e., unity power factor), the given qaxis current should be zero if GCF is used (cf. igq ¼ 0 in Fig. 9.6B). Otherwise, if the CCF is employed, the influence of the capacitor branch should be taken into account since the grid
241
Advanced modeling and control Chapter  9
(A) i*fd
(B)
u gd * uod
PI
u gd
* igd
* uod
PI
i fd
Z g L f Lg
igd
Z g L f Lg
i fq
Z g L f Lg
igq
Z g L f Lg
i*fq
* uoq
PI Zg C f u gd
* igq
* uoq
PI 0
u gq
u gq
FIGURE 9.6 Decoupling block diagrams of daxis and qaxis currents. (A) Converterside current feedback, (B) gridside current feedback.
current is not directly controlled, and hence a nonzero ifq is required. According to the second expression in Eq. (9.13), the qaxis components can be expanded to be Cf
ducq ¼ ug Cf ucd þ ifq igq dt
(9.56)
In steady state, it can be derived from Eq. (9.56) that ifq ¼ ug Cf ucd ¼ ug Cf ugd
(9.57)
which is as shown in Fig. 9.6A. Then, the overall block diagram of the current loop control can be obtained, as shown in Fig. 9.7. The variables in the figure are expressed in the form of complex vectors. The current controller part in Fig. 9.7 is equivalent to the decoupling control as in Fig. 9.6A or Fig. 9.6B, depending on the current sensor positions. Therein, GPI(s) denotes the PI controller and can be expressed by GPI ðsÞ ¼ Kp þ
u g , dq I dq*
Ki s
(9.58)
u g , dq GL s
Gd s
GPI s
jZ g L f Lg
Current controller
idq
jZ g L f Lg
Plant of VSC using Lfilter
FIGURE 9.7 Overall block diagram of the current loop control.
242 Control of Power Electronic Converters and Systems
Besides, Gd (s) means the transfer function of the total time delay caused by sampling, computation, updating of the compare registers, and the zeroorderhold effect of pulse width modulation [16], which can be estimated as Gd ðsÞ ¼ esTd zes$1:5Tsamp
(9.59)
with Td as the time delay and Tsamp as the sampling cycle of the controller. The last part in Fig. 9.7 is the simplified Ltype filter plant (i.e., Eq. 9.51) where GL ðsÞ ¼
1 sðLf þ Lg Þ
(9.60)
9.3.2 Resonance damping technique Despite that LCL filter has the advantages of smaller size and better harmonic attenuation performance over the L filter, the utilization of LCL filters makes the control design more complicated. One main drawback is the amplification of undesired harmonic components around the resonant frequency. Thus, the harmonics generated by VSC at this frequency are amplified through the LCL filter and injected into the grid, leading to the inevitable closedloop control instability. In the literature, there are various methods to deal with this phenomenon, such as PD provided with resistors connected in several ways to the LCL filter [17], AD supplied with the modification of the current control structure using filter capacitor current [14,15], or capacitor voltage [18], inherent damping (ID) [14] achieved by only using the converterside current feedback and sensorless ADbased estimation of the state variables by using complex state observers [19]. Some of these methods are selected and will be introduced in the following. The resistive part of the LCL filter components is neglected to nullify all internal damping, and the grid voltages are assumed to contain only positivesequence fundamental components. Thus, the simplified equivalent circuit per phase is shown as in Fig. 9.8, and the plant model transfer function Gp(s) (which is also the LCL filter sdomain admittance) is YLCL ðsÞ ¼ Gp ðsÞ ¼
If (s)
Ig ðsÞ 1 ¼ 3 Uo ðsÞ Lf Cf Lg s þ ðLf þ Lg Þs
Lf
Ig (s)
(9.61)
Lg
Icf (s) Uo (s)
Cf
FIGURE 9.8 Simplified equivalent circuit per phase at nonfundamental frequencies of Fig. 9.5.
Advanced modeling and control Chapter  9
Uo s
1 sL f
I f s
I cf s
1 U cf s sC f
1 sLg
243
Ig s
FIGURE 9.9 Block diagram of LCL network shown in Fig. 9.8.
Bode Diagram
Magnitude (dB)
100 50 0 50 100 90
Phase (deg)
Undamped 135 180 225 270 102
Damped 103
104
105
Frequency (rad/s)
FIGURE 9.10 Bode plot of the openloop voltage source converter with LCL filter under nondamped and damped conditions.
where the capitalized notations represent the variables in the sdomain and Uo(s) is sfunction of the converter output voltages. The expression of Gp(s) can also be obtained using the block diagram of the LCL filter plant shown in Fig. 9.9. Based on Eq. (9.61), the magnitude and phase response of the undamped LCL filter is plotted in Fig. 9.10. An obvious resonance can be observed at the resonance frequency, which should be avoided since the rapid phase transition may cause instability issues. With the utilization of damping methods, the resonance can be suppressed as shown in Fig. 9.10 with the “damped” label, where the peaky magnitude response and the rapid phase transition are softened.
9.3.2.1 Passive damping technique There are several ways to implement PD [12], among which the minimization of the damping power losses is the vital point to determine the most suitable PD configuration. In light of comprehensive analysis on distinct resistor
244 Control of Power Electronic Converters and Systems
If (s)
Lf
Ig (s)
Lg
Icf (s) Cf
Uo (s)
Rd FIGURE 9.11 Equivalent per phase circuit of LCL filter with passive damping.
Uo s
1 sL f
I f s
I cf s
1 U s Rd cf sC f
1 sLg
Ig s
FIGURE 9.12 Block diagram of LCL network with passive damping.
configurations, inserting simple resistors into the capacitor branch in LCL filters is proved to deliver the least PD losses [20,21], the structure of which is as shown in Fig. 9.11. The equivalent block diagram of the LCL filter with PD is shown in Fig. 9.12, and the transfer function between the grid current and the converter output voltage can be derived as GPD ðsÞ ¼
Ig ðsÞ Rd Cf s þ 1 ¼ 3 Uo ðsÞ Lf Cf Lg s þ Cf Rd ðLf þ Lg Þs2 þ ðLf þ Lg Þs
(9.62)
The magnitude and phase response are plotted in Fig. 9.13, where the peak resonance response is clearly suppressed. In the lowfrequency range, the impact of PD is not considerable due to the much lower impedance of the damping resistor Rd than that of the filter capacitor Cf . In higher frequency range, the impedance of damping resistor is increased, and the highfrequency current components (especially the switching harmonics) begin to flow into the grid. Therefore, there is a tradeoff between the system stability and the low switching ripple injected to the grid.
9.3.2.2 Active damping technique using filter capacitor current feedback Due to the fact that the resonance is caused by the filter capacitor, it is reasonable to process the current/voltage information of the capacitor part. The block diagram shown in Fig. 9.14 employs the capacitor current feedback to realize AD control. The AD block K(s) can be modeled as a simple proportional controller with a constant gain Kd. Seen from Fig. 9.14, it is not convenient to analyze the resonance damping because the feedback current is tapped from the middle of the LCL plant. By moving the connection point backward to the grid current side, an equivalent block diagram can be obtained
Advanced modeling and control Chapter  9
245
Bode Diagram
Magnitude (dB)
100 50 0
Gp s GPD s
50
Phase (deg)
100 90 135 180 225 270 2 10
10
3
10
4
10
5
Frequency (rad/s)
FIGURE 9.13 LCL filter.
Comparison of the attenuation capability between undamped and passivedamped
K (s) Uo s
1 sL f
I f s
I cf s
1 U cf s sC f
1 sLg
Ig s
LCL filter plant model Gp(s) FIGURE 9.14 damping.
Block diagram of the LCL network with capacitor current feedback for active
s 2 K d Lg C f Uo s
U oc s
Gp (s)
Ig s
FIGURE 9.15 Equivalent block diagram of the LCL network using active damping.
as shown in Fig. 9.15. The LCL filter plant is denoted by Gp(s) (cf. Eq. 9.61) in the figure. It can be found that the converter output voltage to the LCL plant is modified by the AD block, and the transfer function from the original Uo(s) to the new Uo0 ðsÞ is
246 Control of Power Electronic Converters and Systems Bode Diagram
Magnitude (dB)
100
0
Gp s GAD s YLCL , eq s
100
Phase (deg)
200 90 0 90 180 270 102
103
104
105
Frequency (rad/s)
FIGURE 9.16 Impact of the capacitor current feedback on resonance damping.
GAD ðsÞ ¼
Uo0 ðsÞ Lf Cf Lg s3 þ ðLf þ Lg Þs ¼ Uo ðsÞ Lf Cf Lg s3 þ Kd Cf Lg s2 þ ðLf þ Lg Þs
(9.63)
Then the equivalent admittance transfer function between the grid current and the converter output voltage can be derived as YLCL;eq ðsÞ ¼
Ig ðsÞ U 0 ðsÞ 1 ¼ Gp ðsÞ$ o ¼ 3 Uo ðsÞ Uo ðsÞ Lf Cf Lg s þ Kd Cf Lg s2 þ ðLf þ Lg Þs (9.64)
Comparing Eq. (9.64) and undamped Eq. (9.61), the s2term appears in the denominator resulting from the capacitor current feedback, which can effectively avoid a phase jump and the resonance peak. In order to verify this, the magnitude and phase responses of Gp(s), GAD(s), and YLCL,eq(s) are plotted in Fig. 9.16. It can be seen that the AD block transfer function GAD(s) behaves like a notch filter (NF), providing a negative peak response as opposed to the undamped LCL filter resonance. If a proper Kd value is selected, these two peaks might cancel out each other at the resonant frequency and an improved damped system can be obtained, as highlighted by the solid red (gray in print version) line in the figure. 9.3.2.3 Active damping technique under grid current feedback The closedloop system consisting of the LCL filter plant Gp(s), the PI current controller GPI(s), and the gridside current feedback ig is as shown in Fig. 9.17.
Advanced modeling and control Chapter  9
247
K (s)
i*
ie
§ 1 · K p ¨1 ¸ T is ¹ ©
uo
GPI (s)
1 sL f
if
icf
1 sC f
ucf
1 sLg
ig
LCL filter plant Gp(s)
FIGURE 9.17 Block diagram of the gridside current feedback control without/with active damping.
It should be noted that the time delay, as shown in the overall system control block diagram Fig. 9.7, is treated as pure amplifier gain here and thus is integrated into the frontend current controller [14]. Detailed information of the impact of time delay on the controller design and system stability can be found in Refs. [22e24], where the method of equivalent impedance model in discrete domain is often employed. The openloop transfer function without AD in Fig. 9.17 can be derived as GPI ðsÞGp ðsÞ ¼
Kp ðs þ 1=Ti Þ Lf Cf Lg s4 þ ðLf þ Lg Þs2
(9.65)
from which the thirdorder term is absent, indicating the difficulty to reach stability for closedloop control. In spite of the PD resulting from the resistive part of the LCL components, the damping extent is generally inadequate. Thus, the aforementioned AD block is added, as shown by the dashed feedback in Fig. 9.17. The openloop function changes to Ig ðsÞ Kp ðs þ 1=Ti Þ ¼ Ie ðsÞ Lf Cf Lg s4 þ Kd Cf Lg s3 þ ðLf þ Lg Þs2
(9.66)
The current controller parameters do not appear in the resulting s3 term in Eq. (9.66), revealing that GPI(s) has no impact on the resonant damping under GCF. Besides, the damping extent can also be evaluated by the damping factor z, which can be regulated by sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ Kd Lf þ Lg (9.67) ¼ 2zures ¼ 2z Lf Lf Lg Cf In Eq. (9.67), the term Kd/Lf is the proportion of the constant coefficients of the second highest order variable (s3 in Eq. 9.66) to the highest order variable (s4 in Eq. 9.66) and ures is the resonant angular frequency. Generally, an optimum damping factor z ¼ 0.707 is recommended in literature [14,15]. In order to validate the effectiveness of the AD in GCF control, the simulated working waveforms of the converterside current and the gridside
248 Control of Power Electronic Converters and Systems
(A)
Undamped
Active Damping
ifb ifc ifa
(B)
Undamped
Active Damping
ifd ifd
FIGURE 9.18 Simulated waveforms of the converterside current under grid current feedback control with Po ¼ 7.5 kW. (A) Threephase converterside AC currents, (B) transformed dq frame converterside currents.
current are shown in Figs. 9.18 and 9.19, respectively. The converter parameters are as listed in Table 9.1, except that the equivalent series resistance of the converter inductor and the grid inductor is set at zero in undamped case. In the following Figs. 9.18 and 9.19, the capacitor current feedbackebased AD is activated at t ¼ 0.04 s, and it can be seen that the resonant threephase AC currents are immediately damped and reach steady state after one fundamental operating period.
9.3.2.4 Active damping technique under converter current feedback A block diagram of the converterside current feedback control is shown in Fig. 9.20, which can be equivalently modified to the control in Fig. 9.21. Compared to the GCF control in Fig. 9.17, an additional GCF appears in the loop, which can be seen as the ID characteristic of CCF. The openloop transfer function can be derived as Ig ðsÞ Kp ðs þ 1=Ti Þ ¼ Ie0 ðsÞ Lf Cf Lg s4 þ Kp Cf Lg s3 þ ðKp Cf Lg =Ti þ Lf þ Lg Þs2
(9.68)
Advanced modeling and control Chapter  9
(A)
Undamped
249
Active Damping
igb igc iga
(B)
Undamped
Active Damping
igd igq
FIGURE 9.19 Simulated waveforms of the gridside currents under grid current feedback control with Po ¼ 7.5 kW. (A) Threephase gridside AC currents, (B) transformed dq frame gridside currents.
i*
§ 1 · K p ¨1 ¸ © Ti s ¹
ie
uo
if
1 sL f
GPI(s)
icf
1 sC f
ucf
1 sLg
ig
LCL filter plant Gp(s)
FIGURE 9.20 Block diagram of converterside current feedback control.
K (s)
Lg C f s 2
i*
ie'
ie
§ 1 · K p ¨1 ¸ © Ti s ¹
GPI(s)
FIGURE 9.21
uo
1 sL f
if
icf
1 sC f
ucf
1 sLg
ig
LCL filter plant Gp(s)
Equivalent block diagram of converterside current feedback with active damping.
250 Control of Power Electronic Converters and Systems
Comparing Eq. (9.68) with Eq. (9.65), it can be found that due to the existence of the extra GCF in CCF, the s3 term also appears in the denominator of the openloop transfer function, indicating an intrinsic damping control of CCF. Similarly, the damping can be obtained with Kp ¼ 2zures Lf
(9.69)
where either Kp or Lf can be used to adjust the damping factor z. If the required damping factor cannot be achieved by only adjusting Kp or Lf, then additional damping terms K(s) could be introduced into the loop. Consequently, the openloop transfer function becomes Ig ðsÞ Kp ðs þ 1=Ti Þ ¼ Ie0 ðsÞ Lf Cf Lg s4 þ ðKp þ Kd ÞCf Lg s3 þ ðKp Cf Lg =Ti þ Lf þ Lg Þs2
(9.70)
Likewise, the damping can be adjusted by Kp þ Kd ¼ 2zures Lf
(9.71)
In order to compare the effectiveness of the damping method under GCF and CCF control, the closedloop poles corresponding to Eqs. (9.65), (9.66), (9.68), and (9.70) are shown in Fig. 9.22. The original GCF (denoted by blue
FIGURE 9.22 Pole locations under grid current feedback (GCF) control and converter current feedback (CCF) control using active damping.
Advanced modeling and control Chapter  9
(A)
7.5 kW
15 kW
(B)
7.5 kW
15 kW
igc igb iga
ifc ifb ifa
(C)
7.5 kW
251
15 kW
(D)
7.5 kW
ifd
15 kW
igd
FIGURE 9.23 Simulated waveforms under converter current feedback with Po changing from 7.5 to 15 kW at t ¼ 0.04 s. (A) Threephase converterside AC currents, (B) threephase gridside AC currents, (C) dq frame converterside currents, (D) dq frame gridside currents.
(gray in print version) points) control is not stable since there are poles in the right s plane, while the other three cases can achieve stability if proper LCL filter components and controller/damping parameters are selected. By operating the converter with CCF control method, the simulated waveforms including the converterside current and the gridside current are shown in Fig. 9.23. In order to keep the grid currents in phase with grid voltages (i.e., unity power factor), the qaxis reference current is given as ug Cf ugd , as shown in Fig. 9.6. The operating power is changed from 7.5 to 15 kW at t ¼ 0.04 s in Fig. 9.23. It can be seen that there is no clear resonance in the converter current and the grid current due to the ID characterization in CCF control, which is in agreement with the discussions before.
9.3.3 Control under unbalanced grid voltages As discussed previously, the AC voltages/currents can be divided into positiveand negativesequence components when the grid voltages are unbalanced or distorted. Thus, a common way to control the grid current under unbalanced power system is to regulate the positive and negativesequence currents separately based on synchronous dqreference frame [25]. Of course, grid currents can also be controlled in a stationary reference frame (e.g., ab frame) and resonant controllers (e.g., proportional resonant controller) instead of PI control are often utilized in this situation. In order to control the grid currents under unbalanced grid voltages, the dq components of the positive and negativesequence voltages/currents should be
252 Control of Power Electronic Converters and Systems
dq
u g, dq
T
u g , abc
LPF
u g, dq
LPF
u g, dq
Tdq2
αβ abc Tdq
u
g , dq
Tdq2
DDSRF FIGURE 9.24 Block diagram of decoupled synchronous reference frame (DDSRF) to be applied in unbalanced grid.
extracted from the original signals at first. Focusing on the fundamental components (i.e., firstorder harmonic) of the AC variables, if Ugþ and Ug are supposed being as the amplitudes of the positive and negativesequence voltages, the respective dq components can be obtained through Park transformation (cf. Eq. 9.9) and the reverse Park transformation, i.e., " þ# Ugd Uga 1 cosð2utÞ cos ut sin ut þ Ug (9.72) ¼ Ugþ ¼ þ Ugb 0 sinð2utÞ sinut cos ut Ugq ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} þ Tdq
"
Ugd Uga cos ut sin ut þ cosð2utÞ 1 ¼ ¼ Ug þ Ug (9.73) Ugb sinut cos ut sinð2utÞ 0 Ugq ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} Tdq
It can be seen that the DC values of the transformed results are equal to the amplitudes of positive/negativesequence voltages, and the twice angular frequency (i.e., 2u) terms indicate the component coupling between daxis and qaxis. From this perspective, a decoupling network [26] similar to Fig. 9.6 can be used to cancel out the effect of the 2u oscillations, which is as shown in Fig. 9.24. þ2 2 are and Tdq In Fig. 9.24, the transformation matrices Tdq h iT cosð2utÞ sinð2utÞ þ2 2 Tdq ¼ Tdq ¼ (9.74) sinð2utÞ cosð2utÞ The LPF block in Fig. 9.24 is a firstorder lowpass filter, and the final outputs Cuþ g;dq D and Cug;dq D are the average decoupled dq voltages of the positiveand negativesequence components, respectively.
Advanced modeling and control Chapter  9
u g , abc
αβ abc
Tdq
NF
u g, dq
Tdq
NF
u g, dq
253
FIGURE 9.25 Block diagram of average dqcomponent calculation using notch filters to be applied in unbalanced grid.
On the other hand, from a filtering point of view, the positive/negativesequence components can also be obtained by using a NF and the transfer function of which can be expressed by GNF ðsÞ ¼
s2
s2 þ u2n þ 2un s þ u2n
(9.75)
where un is the notch frequency and it should be equal to the 2u value in Eq. (9.74). The schematic block is shown in Fig. 9.25. Other than the dqcomponent calculation as discussed above, the grid synchronization technique to extract the grid voltageephase angle is another key part in the unbalanced current control. The commonly used dual secondorder generalized integratorephaselocked loop (DSOGIPLL) is introduced into the control system [27], and the schematic diagram is shown in Fig. 9.26. Therein, the secondorder generalized integrator (SOGI) submodule is the secondorder generalized integrator, which is as shown in Fig. 9.27. The input
(A)
u gcD
u gD
SOGI
2u gD
qu gc D
12
u gd
Tdq u gc E
ug E
2u gE
SOGI qu gc E
(B)
u gc D
u gD
SOGI
2u gD
qu gcD
12
u gc E
SOGI qu gc E
2u
T 1
GPIPLL(s) Z1 1/s Kp+Ki /s
T 1
u gd
T ug E
GPIPLL(s) Z1 Kp+Ki/s 1/s
12
dq
gE
u gq
u gq
12
FIGURE 9.26 Schematic diagram of the DSOGIPLL. (A) Positivesequence grid voltagee phase angular extraction, (B) negativesequence grid voltageephase angular extraction.
254 Control of Power Electronic Converters and Systems
u gD u g E
+
SOGI
_ 6
k
1 s
+ _6
qu gc D qu gc E
1 s
ω1
u gc D u gc E
FIGURE 9.27 Schematic diagram of the secondorder generalized integrator (SOGI) block in Fig. 9.26.
(A)
(B)
Grid voltage positive αβ direct and quadrature components uEc
uDc
quDc
quEc
Grid voltage negative αβ direct and quadrature components uEc
uDc
quEc
quDc
FIGURE 9.28 Phasor diagram of (A) grid voltage positive ab direct/quadrature components and (B) negative ab direct/quadrature components.
to SOGI can be either the aaxis or the baxis component, and the outputs qu0ga qu0gb denote the orthogonal signals of u0ga and u0gb with a 90 degrees lag, respectively. The relationships among these output signals of SOGI are shown in Fig. 9.28. Therefore, the following equations can be derived for calculating the positive and negativesequence components in abreference frame. u0a qu0b ¼ u0aþ þ u0a qu0bþ þ qu0b ¼ u0aþ qu0bþ ¼ 2u0aþ (9.76) qu0a þ u0b ¼ qu0aþ þ qu0a þ u0bþ þ u0b ¼ qu0aþ þ u0bþ ¼ 2u0bþ
(9.77)
u0a þ qu0b ¼ u0aþ þ u0a þ qu0bþ þ qu0b ¼ u0a þ qu0b ¼ 2u0a
(9.78)
qu0a þ u0b ¼ qu0aþ þ qu0a þ u0bþ þ u0b ¼ qu0aþ þ u0b ¼ 2u0bþ (9.79) On the basis of average dqcomponent extraction (cf. Fig. 9.24 or Fig. 9.25) and the grid synchronous method DSOGIPLL (cf. Fig. 9.26), the overall current control procedure under unbalanced grid voltages can be achieved, as shown in Fig. 9.29. Therein, the decoupling positive and negativesequence current control blocks are shown in Fig. 9.30.
Advanced modeling and control Chapter  9
255
Rdc idc Q1
+
Q3
Q5
udc C Q2
Q4
Q6
Lg+Rg ifa Lf +Rf ifb ifc uca ucb ucc Cf

DDSRF DSOGIPLL
i
θ+
Decoupling negativesequence current control
u gd
id*
id
PI Zg L f Lg
Tdq
Zg L f L g
Zg L f L g
Decoupling positivesequence current control
Tdq
Zg L f L g iq
u gq
u gd
PI
uo*, abc
PI iq*
θ
Block diagram of the overall current control under unbalanced grid voltages.
id*
iq
i dq
+ dq
Decoupling positivesequence current control
id
un'
un PWM
FIGURE 9.29
iga uga igb ugb igc ugc ugabc
uo*, abc
PI iq*
u gq
Decoupling negativesequence current control
FIGURE 9.30 Block diagram of the decoupling positive and negativesequence current control.
In Fig. 9.30, the reference values of the positive/negativesequence dqaxis currents are usually obtained from an outer power control loop in unbalanced grid. There are several power control methods in the literature [28], such as instantaneous activereactive control, positive and negativesequence control, average activereactive control, and balanced positivesequence control (BPSC). In this chapter, the BPSC method is considered, and the goal is to inject a set of balanced sinusoidal current with only positivesequence components into the grid. In order to validate the current control under unbalanced grid voltages, a PLECS simulation model with the parameters shown in Table 9.2 is built. Considering the BPSC power control method, the reference negativesequence dqaxis currents are zero. Besides, in order to avoid the current resonance in
256 Control of Power Electronic Converters and Systems
TABLE 9.2 Parameters of 15 kW gridtied voltage source converters with LCL Filter under unbalanced grid voltages. Operating power Po (kW)
7.5/15
Balanced grid peak phase voltage Ug (V)
311
DClink voltage Vdc (V)
650
Converterside inductor Lf (mH)
3
Capacitor branch Cf (uF)
25
Gridside inductor Lg (mH)
1.8
Grid voltage frequency fg (Hz)
50
Switching frequency fsw (kHz)
4
The proportional gain in SOGI block k (Fig. 9.27)
10
The proportional gain in DSOGIPLL block Kp (Fig. 9.25)
10
The integral coefficient in DSOGIPLL block Ki (Fig. 9.25)
1000
Unbalanced a phase grid voltage uga (V)
311 cos(ugt)
Unbalanced b phase grid voltage ugb (V)
280 cos(ugt 2p/3)
Unbalanced c phase grid voltage ugc (V)
342 cos(ugt þ2p/3)
GCF as discussed before, the CCFbased control is applied to the converter. Thus, the reference positivesequence qaxis current is set as ug Cf uþ gd to maintain unity power factor, where ug is the grid voltage frequency, Cf is the capacitor in LCL filter, and uþ gd is the positivesequence daxis grid voltage. The simulated working waveforms of the converterside currents and gridside currents are shown in Figs. 9.31e9.33. During the simulation procedure, the reference positive daxis converterside current is changed from 16 to 32 A at time t ¼ 0.06 s. In Fig. 9.31, the unbalanced threephase AC grid voltages (uga, ugb, ugc) and the balanced gridside currents (iga, igb, igc) are in phase with each other, indicating a unity power factor. By using the DDSRF technique, the positive and negativesequence dqaxis converterside currents are extracted and shown in Fig. 9.32. Similarly, the transient response of the gridside currents is shown in Fig. 9.33. The converter can effectively follow the reference value, which implies the feasibility of the decoupled double synchronous reference frame current control method.
9.4 Impedancebased stability analysis under weak grid conditions In weak grid conditions, one big difference from the strong power system is that the gridside impedance might vary widely, which is mainly dependent on
Advanced modeling and control Chapter  9
(A)
257
ugc ugb uga
(B)
ifd+ =32 A
ifd+ =16 A
ifc
(C)
ifb
ifa
ifd+ =32 A
ifd+ =16 A
igc
igb
iga
FIGURE 9.31 Simulated waveforms under unbalanced grid voltages with the given positivesequence daxis converter current changing from 16 to 32 A at t ¼ 0.06 s. (A) Grid voltages, (B) converterside currents, (C) gridside currents.
(A)
(B)
ifd + ifq +
ifdifq
FIGURE 9.32 Simulated transient response of dqaxis converterside currents with the given positivesequence daxis converter current changing from 16 to 32 A at t ¼ 0.06 s. (A) Positivesequence dq currents, (B) negative dq currents.
258 Control of Power Electronic Converters and Systems
(A)
(B)
igd+
igd
igq+
igq
FIGURE 9.33 Simulated transient response of dqaxis gridside currents with the given positivesequence daxis converter current changing from 16 to 32 A at t ¼ 0.06 s. (A) Positivesequence dq currents, (B) negative dq currents.
the linear transformers and distribution feeders. Consequently, the grid impedance is possibly close to or larger than the converterside impedance, especially when multiple gridconnected converters are interconnected at the point of common coupling (PCC). In that case, the grid impedance cannot be neglected as in the strong power system. In order to reveal the effect of the grid impedance on the converter stability, the impedancebased analysis method is introduced in the following.
9.4.1 System control of the LCLfiltered VSCs in ab frame and dq frame Based on the derived abframe model (9.12), the system control scheme in ab frame is as shown in Fig. 9.34. As discussed previously, in order to damp the resonance caused by the LCL filter, the filter capacitor currentebased AD method is applied. Compared to the stronggrid configuration, the grid impedance between the PCC and the grid voltage is considered and is denoted idc +
Q1
Q3
Q5 ifa ifb ifc
udc C Q2
Q4
Lf +Rf
Lg+Rg
Lt+Rt uga ugb ugc
iga igb igc vpcc
Q6

ucfa ucfb ucfc un
Hcf
Cf Hg
abc/αβ
abc/αβ
PLL θ
PWM abc/αβ
icfαβ – +
Gc(s)
igαβ – i * + gαβ
Ig*
FIGURE 9.34 Block diagram of the system control in abreference frame.
un'
Advanced modeling and control Chapter  9
259
idc +
Q1
Q3
Q5 ifa ifb ifc
udc C Q2
Q4
Lf +Rf
Lg+Rg
Lt+Rt uga ugb ugc
iga igb igc
un'
vpcc
Q6
ucfa ucfb ucfc
PWM
Cf
un
Hcf
dq/αβ
Hg
abc/dq
θ
PLL
abc/dq abc/dq
icfdq – +
Gc(s)
igdq – i * + gdq
Ig*
FIGURE 9.35 Block diagram of the system control in dqreference frame.
by Lt þ Rt in Fig. 9.34. Therein, igab ¼ iga
igb
T
is the reference grid
T
current, igab ¼ ½ iga igb is the actual grid current, and icf ab ¼ ½ icf a icf b T is the capacitor branch current in the LCL filter, all of which are represented by the abframe form. Hcf and Hg are the sensor gains of the grid current igabc and the capacitor branch current icfabc, respectively. Gc(s) is the current controller. Note that the DClink voltage is assumed constant and the current control is the focus. Besides, the grid voltage phase q is detected using the sampled threephase voltages at PCC. It can be easily found that there is a phase difference between the PCC voltages and the actual grid voltages, which will cause a small error between the detected q and the actual q. The impact of this error and the methods to omit it are described in detail in Refs. [29,30]. In terms of the reference current amplitude Ig , it is directly given in Fig. 9.34, instead of being generated by the outer voltage loop as in Fig. 9.5. Similarly, based on the dqframe model (9.13), the grid current control diagram in dq frame is as shown in Fig. 9.35, where
T igdq ¼ igd igq is the reference grid current, igab ¼ ½ igd igq T is the actual grid current, and icfdq ¼ ½ icfd icfq T is the capacitor branch current in the LCL filter, all of which are represented by the dqframe form. According to the control procedures in Figs. 9.34 and 9.35, the block diagrams of the grid current sdomain control in ab frame and dq frame are shown in Figs. 9.36 and 9.37, respectively. The VSC is represented by a proportional gain KPWM, and considering the practical time delay in digital control, the delay block Gd(s) is introduced and can be expressed by Gd ðsÞ ¼ e1:5sTs
(9.80)
Hcf igα* (s)
Gc(s)
Gd(s)
vpccα(s)
ucfα(s) KPWM
uoα(s)
1 ZLf (s) i (s) fα
icfα (s)
Zcf (s)
Hg Hcf igβ * (s)
Gc(s)
Current Controller FIGURE 9.36
Gd(s)
uoβ(s)
igα(s)
vpccβ(s)
ucf β(s) KPWM
1 ZLg(s)
1 ZLf (s) i (s) fβ
icf β (s)
Zcf (s)
1 ZLg(s)
igβ(s)
Plant of VSC and LCLfilter
Block diagram of the sdomain current controller for the LCLfiltered voltage source converter (VSC) under stationary ab frame.
260 Control of Power Electronic Converters and Systems
Hg
Hg Hcf
ucfd(s)
*
igd (s)
Gc(s)
Gd(s)
Gd(s)
KPWM
KPWM
uod(s)
uoq(s)
ifd(s) 1 ZLf (s)
icfd(s)
1 ZLg(s)
Zcf (s)
ωgLf
ωgCf
ωgLg
ωgLf
ωgCf
ωgLg
Zcf (s)
1 ZLg(s)
1 ZLf (s) i (s) fq
icfq(s)
igd(s)
igq(s)
vpccq(s)
Hcf Hg Current Controller
Plant of VSC and LCLfilter
FIGURE 9.37 Block diagram of the sdomain current controller for the LCLfiltered voltage source converter (VSC) under rotating dq frame.
Advanced modeling and control Chapter  9
igq * (s)
Gc(s)
vpccd(s)
261
262 Control of Power Electronic Converters and Systems
where Ts is the sampling period. Alternatively, Gd(s) can also be expressed by Pade approximation and the accuracy can be regulated by selecting different orders [31]. With respect to the current controller Gc(s), it can be denoted by a quasiproportionalresonant (PR) controller in ab frame or a PI controller in dq frame, by considering the sinusoidal form and the constant value of the grid current in ab frame and dq frame, respectively. Thus, the Gc(s) in Fig. 9.36 can be expressed by Gc ðsÞ ¼ Kp þ
s2
2Kr uc s þ 2uc s þ u20
(9.81)
where Kp and Kr are the proportional and resonance coefficients, respectively, and uc ¼ 5 rad/s (assuming grid fundamental frequency is within 50 Hz 0.8 Hz), u0 is the fundamental angular frequency.
9.4.2 Impedancebased stability analysis Taking the abframe control as an example, the gridconnected converter model in weak grids can be represented by a current source irab ¼ ½ ira irb T in parallel with an output impedance Zo(s), as shown in Fig. 9.38. Therein, Zt(s) denotes the grid impedance, ugab ¼ ½ uga ugb T is the grid voltage, and vpccab ¼ ½ vpcca vpccb T is the PCC voltage. Therefore, the grid current igab ¼ ½ iga igb T can be calculated by Zo ðsÞ 1 irab ðsÞ ugab ðsÞ Zo ðsÞ þ Zt ðsÞ Zo ðsÞ þ Zt ðsÞ 1 ugab ðsÞ ¼ irab ðsÞ 1 þ Zt ðsÞ=Zo ðsÞ Zo ðsÞ
igab ðsÞ ¼
(9.82)
On the other hand, the achannel control block in Fig. 9.36 can be simplified, as shown in Fig. 9.39. Due to that the achannel and bchannel control blocks are similar in Fig. 9.36, the achannel is selected as an example here. The aggregated transfer functions Ga1(s) and Ga2(s) can be derived as KPWM Gc ðsÞGd ðsÞZcf ðsÞ ZLf ðsÞ þ Zcf ðsÞ þ KPWM Hcf Gd ðsÞ
igαβ (s) irαβ (s)
Zo(s)
PCC
vpccαβ (s)
Ga1 ðsÞ ¼
Z t(s) ugαβ (s)
FIGURE 9.38 Equivalent circuit of gridconnected converter system.
(9.83)
Advanced modeling and control Chapter  9
263
vpccα(s) igα* (s)
Ga1(s)
Ga2(s)
igα(s)
Hg (s) FIGURE 9.39 Equivalent transformation of the achannel current control block in Fig. 9.36.
Ga2 ðsÞ ¼
ZLf ðsÞ þ Zcf ðsÞ þ KPWM Hcf Gd ðsÞ ZLf ðsÞZLg ðsÞ þ ½ZLf ðsÞ þ ZLg ðsÞZcf ðsÞ þ KPWM Hcf Gd ðsÞZLg ðsÞ (9.84)
Then the closedloop control of the grid current iga(s) in Fig. 9.39 can be obtained with iga ðsÞ ¼
Ga1 ðsÞGa2 ðsÞ Ga2 ðsÞ i ðsÞ vpcca ðsÞ 1 þ Hg Ga1 ðsÞGa2 ðsÞ ga 1 þ Hg Ga1 ðsÞGa2 ðsÞ ¼ Gig ðsÞiga ðsÞ
1 vpcca ðsÞ Zo ðsÞ
(9.85)
Comparing Eqs. (9.82) and (9.85), the current loop gain Gig(s) and the converter output capacitance Zo(s) can be expressed by Gig ðsÞ ¼ ¼
KPWM Gc ðsÞGd ðsÞ s Lf Cf Lg þ s KPWM Lg Cf Hcf Gd ðsÞ þ sðLf þ Lg Þ þ KPWM Hg Gc ðsÞGd ðsÞ (9.86) 3
2
Zo ðsÞ ¼ ¼
Ga1 ðsÞGa2 ðsÞ 1 þ Hg Ga1 ðsÞGa2 ðsÞ
1 þ Hg Ga1 ðsÞGa2 ðsÞ Ga2 ðsÞ
s3 Lf Cf Lg þ s2 Lg Cf KPWM Hcf Gd ðsÞ þ sðLf þ Lg Þ þ KPWM Hg Gc ðsÞGd ðsÞ s2 Lf Cf þ sCf KPWM Hcf Gd ðsÞ þ 1 (9.87)
According to Eq. (9.82), if the converter is stable in strong grid (i.e., Zt(s) ¼ 0), the stability of the LCLfiltered VSC is determined by Zt(s)/Zo(s). Using the parameters listed in Table 9.3, the Bode plots of Zt(s) and Zo(s) are shown in Fig. 9.40. It can be seen that phase difference at the crossing frequency of Zt(s) and Zo(s) is larger than 180 degrees, indicating an unstable grid current control according to the Nyquist stability criterion [32]. It should be
264 Control of Power Electronic Converters and Systems
TABLE 9.3 Parameters of 15 kW gridtied voltage source converter with LCL Filter under weak grids. Operating power Po (kW)
7.5
Balanced grid peak phase voltage Ug (V)
311
DClink voltage Vdc (V)
650
Converterside inductor Lf (mH)
3
Capacitor branch Cf (uF)
10
Gridside inductor Lg (mH)
1.8
Grid voltage frequency fg (Hz)
50
Switching frequency fsw (kHz)
4
Switching frequency fs (kHz)
8
Proportional gain of the quasiPR controller Kp
0.5
Integral coefficient of the quasiPR controller Kr
50
Sensor gain of the capacitor current Hcf
0.1
Sensor gain of the grid current Hg
0.1
Converter gain KPWM
90
Grid impedance Zt(s)
0.5 þ 0.002 s
Bode Diagram
Magnitude (dB)
60
40
20
Zt(s) Zo(s)
0 90
> 180
Phase (deg)
0 90 180 270 360 102
FIGURE 9.40
103
Frequency (rad/s)
104
105
Bode plots of the converter output impedance Zo(s) and the grid impedance Zt(s).
Advanced modeling and control Chapter  9 Bode Diagram
60
Magnitude (dB)
265
40
20
Zt(s) Zo(s)
0 135
Phase (deg)
90 45
< 180
0 45 90 102
103
Frequency (rad/s)
104
105
FIGURE 9.41 Bode plots of the converter output impedance Zo(s) and the grid impedance Zt(s) with reduced time delay.
noted that there are other factors that may lead to an unstable control. For example, the reference current I is directly given in Fig. 9.36. But in practical situation, this reference value is usually generated by the outer DClink voltage control loop, and this will worsen the converter stability [33]. There are many ways to improve the system stability, such as reducing the time delay [34], proper design of the LCL filter (mainly the capacitor [35]), or AD control [36]. Taking the timedelay reduction as an example, if the time delay in Gd(s) is reduced from 1.5Ts to Ts, the obtained Bode plot of Zo(s) is shown in Fig. 9.41. It can be seen that phase difference becomes lower than 180 degrees, indicating a stable grid current control. In order to validate the stability analysis above, the same parameters in Table 9.3 are utilized to simulate the LCLfiltered VSC under the abframe control in Fig. 9.35. The simulation results are shown in Fig. 9.42, including the threephase voltages vpcc,abc at PCC as shown in Fig. 9.42A and the threephase grid currents ig,abc in Fig. 9.42B. At t ¼ 0.05 s, the VSC is connected to the PCC, and it can be seen that the PCC voltages and the gridinjected current are unstable with the time delay td ¼ 1.5Ts. Then the time delay is reduced to td ¼ Ts at the time instant t ¼ 0.13 s in Fig. 9.42. It can be seen the working waveforms become stable after a short transient period, which validates the effectiveness of the timedelay reduction for a stability improvement.
266 Control of Power Electronic Converters and Systems
(A)
(B)
td = 1.5Ts VSC offline
VSC connected
VSC offline
VSC connected td = 1.5Ts
td = Ts
vpccb vpcca vpccc
td = Ts
igb iga igc
FIGURE 9.42 Simulation results of the (A) point of common coupling (PCC) voltage and (B) grid current when different time delays are applied to the LCLfiltered voltage source converter (VSC), using the abframe control shown in Fig. 9.36.
References [1] G.W. Wester, R.D. Middlebrook, Lowfrequency characterization of switched DCDC converters, in: Proc. IEEE Power Processing and Electronics Specialists Conf, 1972, pp. 9e20. [2] A.R. Wood, D.J. Hume, C.M. Osauskas, Linear analysis of waveform distortion for HVDC and facts devices, in: Proc. (Cat. No.00EX441) Ninth Int Conf. Harmonics and Quality of Power, vol. 3, 2000, pp. 967e972. [3] S.R. Sanders, J.M. Noworolski, X.Z. Liu, G.C. Verghese, Generalized averaging method for power conversion circuits, IEEE Trans. Power Electron. 6 (2) (April 1991) 251e259. [4] V.A. Caliskan, G.C. Verghese, A.M. Stankovic, Multifrequency averaging of DC/DC converters, in: Proc. 5th IEEE Workshop Computers in Power Electronics, 1996, pp. 113e119. [5] A.M. Stankovic, S.R. Sanders, T. Aydin, Dynamic phasors in modeling and analysis of unbalanced polyphase AC machines, IEEE Trans. Energy Convers. 17 (1) (2002) 107e113. [6] P.C. Stefanov, A.M. Stankovic, Modeling of UPFC operation under unbalanced conditions with dynamic phasors, IEEE Trans. Power Syst. 17 (2) (2002) 395e403. [7] A.M. Stankovic, T. Aydin, Analysis of asymmetrical faults in power systems using dynamic phasors, IEEE Trans. Power Syst. 15 (3) (2000) 1062e1068. [8] N.M. Wereley, Analysis and Control of Linear Periodically Time Varying Systems, Ph.D. thesis, Massachusetts Institute of Technology, 1990.
Advanced modeling and control Chapter  9 [9] [10] [11]
[12] [13]
[14]
[15] [16] [17]
[18]
[19] [20]
[21]
[22]
[23]
[24]
[25] [26]
267
J.R.C. Orillaza, A.R. Wood, Harmonic statespace model of a controlled TCR, IEEE Trans. Power Deliv. 28 (1) (2013) 197e205. J.J. Rico, M. Madrigal, E. Acha, Dynamic harmonic evolution using the extended harmonic domain, IEEE Trans. Power Deliv. 18 (2) (2003) 587e594. Y. Peng, Z. Shuai, Y. Li, J.M. Guerrero, Z.J. Shen, Dynamicphasor modeling and transient analysis of inverterbased microgrid under unbalanced and harmonic condition, in: 2018 IEEE Energy Conversion Congress and Exposition (ECCE), IEEE, 2018, pp. 3760e3764. M. Liserre, F. Blaabjerg, S. Hansen, Design and control of an LCLfilterbased threephase active rectifier, IEEE Trans. Ind. Appl. 41 (5) (2005) 1281e1291. J. Dannehl, F.W. Fuchs, S. Hansen, P.B. Thøgersen, Investigation of active damping approaches for pibased current control of gridconnected pulse width modulation converters with LCL filters, IEEE Trans. Ind. Appl. 46 (4) (2010) 1509e1517. Y. Tang, P.C. Loh, P. Wang, F.H. Choo, F. Gao, Exploring inherent damping characteristic of LCLfilters for threephase gridconnected voltage source inverters, IEEE Trans. Power Electron. 27 (3) (2011) 1433e1443. S.G. Parker, B.P. McGrath, D.G. Holmes, Regions of active damping control for LCL filters, IEEE Trans. Ind. Appl. 50 (1) (2013) 424e432. J. Wang, J.D. Yan, L. Jiang, J. Zou, Delaydependent stability of singleloop controlled gridconnected inverters with LCL filters, IEEE Trans. Power Electron. 31 (1) (2015) 743e757. B.G. Cho, S.K. Sul, LCL filter design for gridconnected voltagesource converters in high power systems, in: 2012 IEEE Energy Conversion Congress and Exposition (ECCE), IEEE, 2012, pp. 1548e1555. J. Dannehl, C. Wessels, F.W. Fuchs, Limitations of voltageoriented pi current control of gridconnected PWM rectifiers with LCL filters, IEEE Trans. Ind. Electron. 56 (2) (2008) 380e388. V. Blasko, V. Kaura, A novel control to actively damp resonance in input LC filter of a threephase voltage source converter, IEEE Trans. Ind. Appl. 33 (2) (1997) 542e550. R. PenaAlzola, M. Liserre, F. Blaabjerg, R. Sebastian, J. Dannehl, F.W. Fuchs, Analysis of the passive damping losses in LCLfilterbased grid converters, IEEE Trans. Power Electron. 28 (6) (2012) 2642e2646. M. Liserre, A. Dell’Aquila, F. Blaabjerg, Stability improvements of an LCLfilter based threephase active rectifier, in: 2002 IEEE 33rd Annual IEEE Power Electronics Specialists Conference. Proceedings (Cat. No. 02CH37289), IEEE, vol. 3, 2002, pp. 1195e1201. J. Xu, S. Xie, T. Tang, Active dampingbased control for gridconnected LCLfiltered inverter with injected grid current feedback only, IEEE Trans. Ind. Electron. 61 (9) (2013) 4746e4758. W. Xia, J. Kang, Stability of LCLfiltered gridconnected inverters with capacitor current feedback active damping considering controller time delays, J Mod. Power Syst. Clean Energy 5 (4) (2017) 584e598. X. Wang, F. Blaabjerg, P.C. Loh, Gridcurrentfeedback active damping for LCL resonance in gridconnected voltagesource converters, IEEE Trans. Power Electron. 31 (1) (2015) 213e223. A. DoriaCerezo, M. Bodson, Design of controllers for electrical power systems using a complex root locus method, IEEE Trans. Ind. Electron. 63 (6) (2016) 3706e3716. P. Rodrı´guez, J. Pou, J. Bergas, J.I. Candela, R.P. Burgos, D. Boroyevich, Decoupled double synchronous reference frame PLL for power converters control, IEEE Trans. Power Electron. 22 (2) (2007) 584e592.
268 Control of Power Electronic Converters and Systems [27] Y. Yang, L. Hadjidemetriou, F. Blaabjerg, E. Kyriakides, Benchmarking of phase locked loop based synchronization techniques for gridconnected inverter systems, in: 2015 9th International Conference on Power Electronics and ECCE Asia (ICPEECCE Asia), IEEE, 2015, pp. 2167e2174. [28] R. Teodorescu, M. Liserre, P. Rodriguez, Control of Grid Converters under Grid Faults, 2007. [29] D. Yang, X. Wang, F. Liu, K. Xin, Y. Liu, F. Blaabjerg, Symmetrical PLL for SISO impedance modeling and enhanced stability in weak grids, IEEE Trans. Power Electron. 35 (2) (2019) 1473e1483. [30] X. Wang, L. Harnefors, F. Blaabjerg, Unified impedance model of gridconnected voltagesource converters, IEEE Trans. Power Electron. 33 (2) (2017) 1775e1787. [31] X. Wang, F. Blaabjerg, P.C. Loh, An impedancebased stability analysis method for paralleled voltage source converters, in: 2014 International Power Electronics Conference (IPECHiroshima 2014ECCE ASIA), IEEE, 2014, pp. 1529e1535. [32] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, Z. Shen, Inverse nyquist stability criterion for gridtied inverters, IEEE Trans. Power Electron. 32 (2) (2016) 1548e1556. [33] D. Lu, X. Wang, F. Blaabjerg, Impedancebased analysis of dclink voltage dynamics in voltagesource converters, IEEE Trans. Power Electron. 34 (4) (2018) 3973e3985. [34] D. Yang, X. Ruan, H. Wu, A realtime computation method with dual sampling mode to improve the current control performance of the LCLtype gridconnected inverter, IEEE Trans. Ind. Electron. 62 (7) (2014) 4563e4572. [35] X. Wang, F. Blaabjerg, P.C. Loh, Passivitybased stability analysis and damping injection for multiparalleled VSCS with LCL filters, IEEE Trans. Power Electron. 32 (11) (2017) 8922e8935. [36] L. Harnefors, A.G. Yepes, A. Vidal, J. DovalGandoy, Passivitybased controller design of gridconnected VSCs for prevention of electrical resonance instability, IEEE Trans. Ind. Electron. 62 (2) (2014) 702e710.
Chapter 10
Phaselocked loops and their design Wenzhao Liu, Frede Blaabjerg Department of Energy Technology, Aalborg University, Aalborg, Denmark
10.1 Introduction The development of the phaselocked loops (PLLs) dates back to 1930s when it was designed to the synchronous reception of radio signals [1]. After that, the PLL technologies have been widely developed in different industry areas such as communication systems, motor drive systems, contactless power supplies, and grid synchronization of the power electronicebased renewable energy systems, etc. Furthermore, the detailed applications can be divided into the estimation of fundamental parameters of electrical signals [2], electrical synchronization of power quality instruments [3], measurement of harmonics and interharmonics [4], control of AC and DC electrical machines [5], implementing adaptive filters and robust controllers, islanding detection of microgrids [6], grid faults and voltage sags detection [7,8], etc. In recent years, the PLLs are probably the most widely used in gridconnected synchronization applications. For the renewablebased power generation systems, power converters playing as the interface devices are usually employed connecting to the main grid and local loads. Generally, the synchronization technology is defined as a procedure of coordinating power generation units and main AC grid so that they are able to effectively operate in parallel. This procedure may often require PLLs to extract the grid voltage information such as its amplitude, phase, and frequency and provide necessary reference signals to system controllers [9]. However, the main challenge of all synchronization techniques for power electronicebased electrical system is facing a number of disturbances (i.e., the presence of unbalances, harmonics, asymmetrical amplitude sags/swells and DC offset, etc.) in the grid voltage. These nonideal but widely existing disturbances are depending on many factors, which are mainly attributable to the high penetration of distributed
Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000135 Copyright © 2021 Elsevier Ltd. All rights reserved.
269
270 Control of Power Electronic Converters and Systems
generation systems especially equipped with nonlinear power electronic loads or power electronicebased devices, which may affect the accuracy and/or response speed of synchronization technology. However, it should be noted that a highly important unit in the control of the power converters is the synchronization part, which performs a series of action to ensure the power converter and the main grid to be able to work safely and effectively. The information provided by the synchronization unit may be used for different monitoring and further protection purposes. The synchronization techniques in power converters may be categorized to openloop and closedloop methods [10]. The implementations of closedloop synchronization always require feeding back one or more signals, and there are two main categories (PLLs and frequencylocked loop (FLLs)) [11]. The openloop synchronization is free of any feedback signals in their structures. This chapter focuses on the discussion of popular closedloop PLL and their design in practiced application. On the other hand, much research efforts focused on solving power quality issues have been proposed to improve the performance of PLLs under different working conditions [11e13]. However, these efforts often result in very complicated or highly nonlinear control structures. That lead to some shortcomings such as high computational burden, implementation complexity, difficult to model and do stability analysis for the system, and inefficient under large frequency drifts etc. Therefore, this chapter aims at presenting a comprehensive survey on various PLL synchronization techniques to facilitate quick and proper selection for further development and applications for researchers and engineers.
10.2 PLL’s control and design A gridconnected converter with a closedloop PLL synchronization control to a timevarying grid voltage signal can be simplified as shown in Fig. 10.1. The mentioned PLL is usually devided into three parts: phase detection (PD), loop filter (LF), and voltagecontrolled oscillator (VCO). And the difference between phase angle of the grid voltage input signal and that of the output signal is detected by PD and then sent to the LF. The LF output signal drives the VCO to generate the output signal, which could track the grid voltage phase information. More specifically, the PD is mainly responsible to generate a signal containing the phase error information, and the LF also known as the loop controller, which drives the phase error signal to be zero. Finally, the VCO produces a synchronized unit vector with a phase b q in its outputs. Furthermore, more detailed PLLs can be classified into threephase and singlephase applications. A good PLL usually requires lower computational burden, higher robust grid disturbances (i.e., voltage imbalance, harmonics,
Phaselocked loops and their design Chapter  10
FIGURE 10.1 converter.
271
Classical phaselocked loop (PLL) control structure in threephase gridconnected
frequency variations), and fast dynamic response ability and enhanced stability. The following will present some advanced threephase and singlephase PLLs and their design.
10.3 Threephase PLLs 10.3.1 Conventional synchronous reference frame PLL The synchronous reference frame PLL (SRFPLL) may be the most famous PLL in threephase gridconnected converter applications [2,9,14]. The typical b g , and b SRFPLL structure is shown in Fig. 10.2, where Vb, u q are the estimated amplitude, frequency, and phase angle, respectively. The un is the nominal frequency of the detected grid voltage signal. The operation principle of the SRFPLL can be summarized as follows: firstly, transform the threephase grid voltage Vabc to the Vdq in the synchronous frame, then the PI controller is adopted to suppress the Vq to zero, and consequently align the grid voltage to daxis. The output of the PI controller together with the integrator finally produces the grid voltage phase angle, which is also delivered back to the Park transformation applied in the mentioned power circuit in Fig. 10.1.
272 Control of Power Electronic Converters and Systems
FIGURE 10.2 Schematic diagram of the conventional synchronous reference frameebased phaselocked loop.
In order to further understand the working principle the SRFPLL, it is assumed that the threephase grid voltage (input of the SRFPLL) can be expressed as 3 2 þN X 7 6 V1 cosðq1 Þ þ Vh cosðqh Þ 7 6 N;hs1 7 6 2 3 6 7 Va ðtÞ 7 6 þN X 6 2p 7 6 7 6 V cos q 2p þ 7 V cos q (10.1) V ðtÞ ¼ 4 b 5 6 1 1 h h 3 3 7 7 6 N;hs1 Vc ðtÞ 7 6 6 7 þN X 6 2p 7 5 4 V1 cos q1 þ 2p þ Vh cos qh þ 3 3 N;hs1 where V1 and q1 represent the fundamental frequency amplitude and phase angle of the threephase signals, respectively. Vh and qh represent the horder harmonic amplitude and phase angle of the threephase signals, respectively. 2 3 1 1 1 7 2 2 7 26 6 (10.2) Tclarkðabc/abÞ ¼ 6 pﬃﬃﬃ pﬃﬃﬃ 7 34 3 35 0 2 2 cosðb q 1 Þ sinðb q1Þ Tparkðab/dqÞ ¼ (10.3) q1Þ sinðb q 1 Þ cosðb
Phaselocked loops and their design Chapter  10
273
Applying the Clark and Park transformation, which is expressed as (10.2) and (10.3), into (10.1), the daxis and qaxis component can be obtained as follows: 3 2 þN X b Vh cosðqh q h Þ 7 6 7 6 N;hs1 Vd ðtÞ q1Þ V1 cosðq1 b 7 (10.4) þ6 ¼ þN 7 6 X b Vq ðtÞ V1 sinðq1 q 1 Þ 5 4 b Vh sinðqh q h Þ N;hs1
2
Dqh qh zﬄﬄ}ﬄﬄ{ zﬄﬄﬄﬄ}ﬄﬄﬄﬄ{ 6 ð ð 6 Ð Ð " # 6 ug dt ¼ ðun þ Dug Þdt ¼ un dt þ Dug dt 6 qh 6 ¼6 b 6 qh qh Db qh 6 zﬄﬄ}ﬄﬄ{ zﬄﬄﬄﬄ}ﬄﬄﬄﬄ{ 6ð ð ð ð 4 b g dt ¼ un dt þ D u b g dt b g dt ¼ un þ D u u
3 7 7 7 7 7 7 7 7 5
(10.5)
Substitute (10.5) into (10.4), then the following equations can be obtained: 3 2 þN X V cosðqh b qhÞ 7 " # 6 # " 7 6 N;hs1 h Vd ðtÞ V1 cosðq1 b q1Þ 7 6 þ 6 þN ¼ 7 7 X 6 Vq ðtÞ q1Þ V1 sinðq1 b 4 b Vh sinðqh q h Þ 5 (10.6) " z
V1 þ Dd ðtÞ q 1 Þ þ Dq ðtÞ V1 ðq1 b
#
N;hs1
As it can be observed in (10.6), the signal Vd(t) is a measure of the amplitude of the threephase signals, and Vq(t) contains the phase error information. Based on these characteristics and the diagram in Fig. 10.2, the linearized models of the SRFPLL can be derived as shown in Fig. 10.3. According to this model, the openloop and closedloop transfer functions of the SRFPLL can be determined as follows:^ Db q 1 ðsÞ ¼
D^q1 ðsÞ ¼
Vðkp s þ ki Þ qe ðsÞ s2
Vðkp s þ ki Þ kp s þ ki Dq1 ðsÞ þ 2 Dq ðsÞ s2 þ Vkp s þ Vn ki s þ Vkp s þ Vn ki
(10.7) (10.8)
274 Control of Power Electronic Converters and Systems
(A)
Amplitude model
(B)
Phase angle model FIGURE 10.3 Linearized models of the typical synchronous reference frameebased phaselocked loop. (A) Amplitude model, (B) phase angle model.
b g ðsÞ ¼ Du
s2
Vðkp s þ ki Þ sðkp s þ ki Þ Dug ðsÞ þ 2 Dq ðsÞ s þ Vkp s þ Vki þ Vkp s þ Vn ki D Vb1 ðsÞ ¼
up up DV1 ðsÞ þ Dd ðsÞ s þ up s þ up
(10.9) (10.10)
In fact, these models provide quite useful information for the better understanding of the characteristics of the SRFPLL as follows: l
l
l
In Fig. 10.3, the amplitude V is a gain in the forward path of the SRFPLL model, which means that the variations of the threephase input amplitude may change the loop gain and affect the stability margin and dynamic performance. Consequently, the input amplitude variations will change the loop gain, dynamics, and stability. According to (10.7), where the impact of Dq(s) has been neglected, it has two openloop poles, so the SRFPLL can be treated as a typeII control system. Therefore, the SRFPLL can track phase angle jumps and frequency steps with zero steadystate phase error, but it is difficult to achieve frequency ramps, which may happen in the power system. Furthermore, the phase error during the frequency ramps can be reduced by increasing the PLL bandwidth but degrading the noise immunity of SRFPLL. According to (10.8)e(10.10), the phase angle, frequency, and amplitude estimated by the SRFPLL suffer from the disturbance of Dd(s) and Dq(s). If there is DC offset and unbalances in the input signal, there will be fundamental frequency and doublefrequency components in Dd(s) and
Phaselocked loops and their design Chapter  10
275
Dq(s), and horder harmonic will appear as h 1 order harmonic, which will decrease the tracking accuracy of SRFPLL. More specifically, Figs. 10.4e10.6 show simulation results of SRFPLL under different grid voltage cases. It can be included that the conventional SRFPLL has a very limited capability to mitigate the grid voltage disturbances (i.e., grid voltage imbalance or harmonics), since the fundamental frequency and doublefrequency disturbances will appear or more harmonics will be presented in the control loop of SRFPLL. Nevertheless, narrowing the loop bandwidth of SRFPLL will significantly degrade the dynamic performance and still cannot mitigate the above disturbances. These drawbacks of the SRFPLL are the main motivation to design the PLLs to be more efficient. On the other hand, with the increased penetration of renewable energy sources to the power grid and the proliferation of nonlinear loads have caused serious power quality issues and require better synchronization performance. In order to deal with the above problems, many advanced PLLs are proposed and designed with enhanced disturbance rejection capabilities [9,14].
(A)
(B)
Time (s)
(C)
Time (s)
(D)
Time (s)
Time (s)
FIGURE 10.4 Simulation results of synchronous reference frame phaselocked loop in the threephase normal grid voltage. (A) Threephase grid voltage, (B) grid voltage components at dq frame, (C) estimated phase angle, (D) angular speed.
276 Control of Power Electronic Converters and Systems
(A)
(B)
Time (s)
(C)
Time (s)
(D)
Time (s)
Time (s)
FIGURE 10.5 Simulation results of synchronous reference frame phaselocked loop in the threephase unbalanced grid voltage. (A) Threephase grid voltage (unbalance is 3.5%), (B) grid voltage components at dq frame, (C) estimated phase angle, (D) angular speed.
(A)
(B)
Time (s)
(C)
Time (s)
(D)
Time (s)
Time (s)
FIGURE 10.6 Simulation results of synchronous reference frame phaselocked loop in the threephase grid voltage with harmonics. (A) Grid voltage (4% of fifth harmonics, 3% of seventh harmonic), (B) grid voltage components at dq frame, (C) estimated phase angle, (D) angular speed.
Phaselocked loops and their design Chapter  10
277
10.3.2 Moving average filterebased PLLs Fig. 10.7 shows a conventional SRFPLL with moving average filter (MAF), which is referred as MAFPLL [15]. The MAF is a linearphase filter which can be described as GMAF ðsÞ ¼
1 eTus Tus
(10.11)
Notice that including the MAF inside the SRFPLL control loop can significantly improve its filtering capability, but slow down its dynamic response considerably [15]. The reason is that the MAF inloop will cause a phase delay. It is on the condition that the window length of MAF is equal to the nominal period of the input signals. The selection for the window length is recommended to be equal to the fundamental period of the grid voltage (Tu ¼ T ), when the grid harmonic is unclear and DC offset may be presented in the PLL input [15,16]. Tu is the window length of MAF, more choices for the window length of the MAF such as Tu ¼ T/2 and Tu ¼ T/6 are suitable for applications when there are possible oddorder harmonics in the input of PLLs [17]. The MAF will pass the DC component and completely block frequency components of integer multiples of 1/Tu. Furthermore, in order to improve the dynamic of the MAFPLL while maintaining better harmonicsfilter performance, several methods are proposed in the literature. In Ref. [18], a proportional integral derivative (PID) controller is used instead of the conventional PI controller as the LF of the MAFPLL, which can provide an additional degree of freedom. Therefore, it enables the designer to effectively compensate for the phase delay caused by the MAF by arranging a polezero cancellation in the design [9]. In addition, a special lead compensator is added before the PI controller in the MAFPLL [19], where the transfer function of the compensator is inverse of the MAF’s transfer function; therefore, it will be able to reduce the phase delay in the MAFPLL control loop. It should be emphasized here that, the MAFPLL with a window length equal to the input fundamental period can remove all of the harmonics up to the aliasing frequency in addition to the fundamentalfrequency disturbance components.
FIGURE 10.7 Schematic diagram of the moving average filterebased phaselocked loop.
278 Control of Power Electronic Converters and Systems
10.3.3 Notch filterebased PLLs A notch filter (NF) is a bandrejection filter that significantly attenuates specific frequency signals but passes all other frequency components with negligible attenuation. This feature makes the NF attractive in order to cancel the selected desired harmonic components presented in the input signal [9,14,20]. In fact, the NFs can be divided into adaptive or nonadaptive filters. The former one is very preferred by designers because it is easier to select a narrow bandwidth for NFs to minimize the phase delay in the control loop of the PLL. However, this advantage increases computational burden of the PLL. The structure of NFPLL is similar to the standard MAFPLL, except that the MAF is replaced with NFs as shown in Fig. 10.8. In industry, more than one NF in the PLL control loop can be extended with cascaded topology [21] and parallelconnecting topology [22]. The main difference between these topologies is their frequency estimation method, the latter topology uses the same frequency estimator for all of the NFs. However, in the cascaded topology, every NF is equipped with its own frequency estimator and there is a tradeoff between the filtering capability and computational burden in both topologies. To achieve a satisfactory compromise, three NFs with notch frequencies at 2ug, 6ug, and 12ug are usually suggested by designers to obtain a robust PLL [9].
10.3.4 Sinusoidal signal integratorebased PLLs Fig. 10.9 shows the schematic diagram of sinusoidal signal integrator PLL (SSIPLL) which tracks the grid voltage by extracting the fundamental positivesequence component with the structure of SRFPLL. Therefore, it can operate well under unbalanced and harmonic voltage conditions. The parameter K in Fig. 10.9 is designed to control the loop bandwidth and the response speed of the SSIPLL. In addition, a similar structure with single SSI can be found in Ref. [23], where the positivesequence component is extracted by using a single SSI as a filter for the detected grid voltage, and calculated from grid voltage by
FIGURE 10.8 Schematic diagram of the notch filterebased phaselocked loop.
Phaselocked loops and their design Chapter  10
279
FIGURE 10.9 Schematic diagram of the sinusoidal signal integratorebased phaselocked loop.
delaying the signal with 90 degrees. The main advantages of SSIPLL are immunity to the voltage distortion and being able to operate in unbalanced grid conditions. Moreover, it can be more simplified and applied to singlephase power system with a few modifications [24].
10.3.5 Secondorder generalized integratorebased PLLs The secondorder generalized integratorebased PLL (SOGIPLL) can be treated as SSI and acts as a quadrature signal generator and bandpass filter (BPF) by feeding back its output signal. The SOGIPLL is a useful tool for the extraction and separation of the fundamental positive and negativesequence components of threephase grid signals [9,24]. It implies that its structure is somehow mathematically equivalent to the DSRFPLL [25], DCCFPLL [26], and dual SOGIbased PLL (DSOGIPLL) [11]. In Fig. 10.10, two SOGIs are applied to extract the filtered direct and quadrature components of Va and Vb, and its components are calculated based on the instantaneous symmetrical component method. In addition, in order to improve the harmonic filtering capability of the DSOGIPLL, some additional
FIGURE 10.10 Schematic diagram of the dual secondorder generalized integratorebased phaselocked loop.
280 Control of Power Electronic Converters and Systems
SOGIs tuned at harmonic frequencies can be added to the standard structure [11,27]. Another interesting solution is to apply the thirdorder generalized integratorebased BPF and can be found in Ref. [27], which can further improve the dynamic response ability. It is worth mentioning that an accurate fundamental positive and negativesequence component extraction in DSOGIPLL requires 90 degree phase shift of the grid voltage. This method is not frequencyadaptive and will give rise to errors in the positivesequence estimation. Therefore, a combination of a low pass filter (LPF) and a BPF is developed, where the BPF (Eq. 10.12) only provides harmonic filtering functions and the LPF (Eq. 10.13) offers both the harmonic filtering and 90 degree phase shift at the same time. v0 kus ¼ v s2 þ kus þ u2
(10.12)
qv0 ku2 ¼ 2 v s þ kus þ u2
(10.13)
GBPF ðsÞ ¼ GLPF ðsÞ ¼
Besides, the parameters of the filter can be updated to achieve the frequencyadaptive purpose for the PLL. In addition, it could be applied to singlephase power system, i.e., SSIPLL [28].
10.3.6 Complex coefficient filterebased PLLs The complex coefficient filterebased PLLs (CCFPLLs) can be characterized by introducing an asymmetrical frequency response around zero frequency, which implies the CCFs can make a distinction between the positive and negativesequence components at different frequencies [26]. Therefore, the CCFPLLs are interesting for the selective extraction and/or cancellation of different harmonic components from the grid voltage. Fig. 10.11 shows the schematic diagram of a decoupled CCFPLL (DCCFPLL), which uses two complexcoefficient BPFs as the SRFPLL prefiltering
FIGURE 10.11 Schematic diagram of decoupled complex coefficient filterebased phaselocked loop.
Phaselocked loops and their design Chapter  10
281
stage [26]. The N(s)/D(s) can be extended to different transfer functions to affect the steady and dynamic performance of the CCFPLLs. In addition, the CCFs in the input of the SRFPLL are working in a collaborative way, each CCF is responsible for extracting a particular component of the input signal. The fundamental frequency positive and negativesequence grid voltage components can be expressed as þ Vab ðsÞ ¼
NðsÞðs þ ju0 Þ Vab ðsÞ DðsÞ s2 þ u20 þ 2NðsÞs
(10.14)
Vab ðsÞ ¼
NðsÞðs ju0 Þ Vab ðsÞ DðsÞ s2 þ u20 þ 2NðsÞs
(10.15)
when N(s) ¼ uc, D(s) ¼ 1, the Bode diagram of DCCFPLL can be illustrated as shown in Fig. 10.12. It can be seen that the positivesequence curve has a magnitude of 1.0 p.u at 50 Hz and a magnitude of 0 at 50 Hz. On the contrary, the amplitude of the negativesequence curve is 0 at 50 Hz and 1.0 p.u at 50 Hz. Therefore, the DCCF can achieve accurate decoupling of positive and negativesequence components at the fundamental grid voltage frequency. In addition, the DCCF equipped with LPFs can be used to reduce the potential DC offset components from threephase grid voltage and the structure can be found in Fig. 10.13 [26], where u1 is the cutoff frequency of the LPF inside of the DCCFPLL to extract the potential DC offset components in grid voltage. The u0 represents the detected grid voltage frequency and V0 is the DC offset components measured from the grid voltage.
FIGURE 10.12 Bode diagram of the decoupled complex coefficient filterebased phaselocked loop.
282 Control of Power Electronic Converters and Systems
FIGURE 10.13 Physical implementation diagram of the decoupled complex coefficient filtere based phaselocked loop.
However, it should be noted that the LPFs inside of DCCMPLL may lead to a slow dynamic response. As one possible solution, the dynamic response can be improved by using a PID controller as the LF of the SRFPLL in order to provide a polezero cancellation. On the other hand, the CCFPLLs can be extended to the dominant harmonic components by using extra complex BPFs centered at the desired harmonic frequencies. The CCFPLLs are not limited to the study cases as described in Figs. 10.12 and 10.13. The topology of CCFs can be used as an inloop filter of different PLLs, but the potential efforts and design optimization still need further research investigations under different case studies and tests.
10.3.7 Delayed signal cancellationebased PLLs The delayed signal cancellationebased PLLs (DSCPLLs) are generally designed to improve the filtering capacity of the conventional PLLs due to their easier tailored characteristic under different grid voltage cases [29]. It usually severs as an inloop filter in the SRFPLL or as a preprocessing filter before the SRFPLL input. However, the DSC may increase the phase delay
Phaselocked loops and their design Chapter  10
283
and slow down the dynamic response, and even affect the stability of the PLLs. Therefore, the DSC operators in most cases are usually playing as a preprocessing filter tool to improve the filtering capability of the SRFPLL [9], and the number of DSCs in the PLL control loop depends on the anticipated harmonic components from threephase grid voltages. On the other hand, the frequency estimated by the SRFPLL is often fed back to adapt frequency variations in the system. More DSC operating in the control loop of SRFPLLs will increase the implementation complexity and the computational burden. The frequency feedback loop makes the PLLs system highly nonlinear and hard to analyze in terms of stability. Some alternative approaches are introduced in Refs. [9,30], but still require more computational efforts. In addition, the method of correcting the phase and amplitude errors in PLLs is suggested to reduce computational burden, as the length of the delays of the DSC remains fixed, and the stability analysis can be carried out easily [31]. However, this method does not improve the disturbance rejection ability of the nonadaptive DSCs when the grid frequency deviates from its nominal value, so that there are some limitations for further application under large frequency drifts and severe also during asymmetrical grid voltage faults.
10.3.8 Multiple SRF filterebased PLLs Multiple SRF filterebased PLLs (MSRFPLLs) are another popular extended version of SRFPLLs. As an example, in Fig. 10.14, the DSRFPLL is equipped with two SRFs rotating at the same angular speed but with opposite directions, and a crossfeedback structure is applied to extract and separate the fundamental frequency positive/negativesequence components of the grid voltage [25].
FIGURE 10.14 Schematic diagram of the dual synchronous reference frameebased phaselocked loop [25].
284 Control of Power Electronic Converters and Systems
In addition, it should be noted that the grid voltage imbalance has no steadystate negative effect on the DSRFPLL performance. However, the presence of grid voltage harmonics in the DSRFPLL input may cause oscillatory errors in the estimated quantities. This problem can be solved by adding several SRFs rotating at the targeted harmonic frequencies to the standard structure [25,32]. Therefore, this kind of PLLs is usually called the MSRFPLL. However, more SRFs will cause a considerable increase in the PLL computational burden. A systematic approach for tuning the control parameters of the MSRFPLL can be found in Ref. [33]. It is also worth mentioning that the DSRFPLL is mathematically equivalent with the decoupled double SRFPLL (DDSRFPLL) if the input signal of the PI þ controller is adjusted to Vq;1 .
10.3.9 Other threephase PLLs There are also some other interesting threephase PLLs. The selective cancellation of harmonic components by using a repetitive regulator/controller inside the SRFPLL control loop (RRPLL) is suggested in Ref. [34]. The controller of RRPLL is based on the discrete transformation, and its computational burden is independent of the number of grid voltage harmonics that are intended to be blocked, which means removing a single harmonic or more grid harmonics using this regulator requires the same computational efforts. This is one of the obvious advantages for RRPLL. However, the regulator of RRPLL highly depends on the sampling frequency, and increasing the sampling frequency drives up the potential computational cost. Therefore, the RRPLL may not be suitable for applications where the sampling frequency is very high. On the other hand, in order to remove the fundamental frequency negativesequence component, reforming the imbalanced signals to the balanced ones, a zerocrossing detectionbased PLL (ZCDPLL) is proposed in Ref. [35]. The ZCDPLL is simple to implement and can operate effectively even in the presence of multiple zero crossings in the PLL input. However, the ZCDPLL only considers the amplitude imbalance in the grid voltages, but cannot remove the components of phase imbalance in the grid voltages. Furthermore, the harmonic filtering capability of ZCDPLL is very limited. In Ref. [36], the Space Vector Fourier Transforms (SVFT) are employed as the SRFPLL prefiltering stage. The SVFT can effectively reject almost all of harmonic components with a very low computational effort. However, the recursive implementation of the SVFTbased filter may involve potential stability problems in the power system. This stability issues can be avoided by implementing the SVFT in the nonrecursive form, but at the expense of increasing computational cost.
Phaselocked loops and their design Chapter  10
285
In Ref. [37], a scheme of SecondOrder Lead Compensators (SOLC) included into SRFPLL is proposed. These compensators have pairs of purely imaginary zeros and poles, which means they can provide a selective harmonic cancellation like NFs without causing a big phase delay in the PLLs. Consequently, these compensators can improve the filtering capability but at the cost of a low noise immunity. Furthermore, the threephase PLLs should also have a DC offset rejection ability as required in Ref. [9]. Noted that the presence of the DC offset components may be caused by the incorrected installation of the electrical devices, halfwave rectification, geomagnetic phenomena, and the shortterm grid faults. To deal with these problems, a simple and effective method by adding an integratorbased DC offset estimation loop to the standard PLL structure is proposed in Ref. [38]. In addition, a method by subtracting the abaxis voltage components from their delayed versions and with a frequencyadaptive correction unit improves the robustness perfomance of the PLL [9]. This technique ensures a complete and fast rejection of the DC offset in PLLs. A performance comparison of five DC offset rejection strategies for threephase PLLs can be found in Ref. [39].
10.3.10 Performance comparison and recommendation A summary of the typical threephase PLLs can be found in Fig. 10.15 and their performance comparison between the mentioned PLLs is summarized in Table 10.1. Notice that the performance results reported in Table 10.1 are corresponding to the typical structure of each threephase PLL. In case of the ideal grid synchronization, the SRFPLL is still the most common PLL due to its design simplicity and lower computational burden as well as fast dynamic response ability. However, the SRFPLL is very sensitive to the grid voltage harmonics and imbalances. Therefore, several PLLs such as SSIPLL, SOGIPLL, and DCCF are recommended to enhance the robustness to the system disturbances and unbalanced grid voltage faults. On the other
FIGURE 10.15 Summary of the typical threephase phaselocked loops (PLLs).
Features PLLs type
Design simplicity
Frequency adaptive
Harmonic robustness
Imbalance robustness
Computational burden
Dynamic response
SRFPLL [9]
Good
Average
No
Poor
Average
Fast
MAFPLL [15]
Average
Average
No
Poor
Low
Average
NFPLL [20]
Good
Average
Yes
Average
High
Fast
SSIPLL [23]
Average
Average
Yes
Good
High
Fast
SOGIPLL [24]
Average
Average
Yes
Good
High
Fast
MSRFPLL [25]
Average
Average
Yes
Poor
High
Fast
DCCFPLL [26]
Average
Average
Yes
Good
High
Fast
DSCPLL [29]
Average
Average
No
Good
Low
Average
RRPLL [34]
High
Average
No
Average
High
Average
ZCDPLL [35]
Good
Poor
No
Poor
Low
Fast
SVFTPLL [36]
Average
Average
No
Average
High
Fast
SOLCPLL [37]
Average
Average
No
Average
Average
Fast
286 Control of Power Electronic Converters and Systems
TABLE 10.1 Performance comparison between threephase phaselocked loops (PLLs) shown in Fig. 10.15.
Phaselocked loops and their design Chapter  10
287
hand, the dynamic response and stability performance is the main concern of the PLLs. Almost all of the mentioned PLL in Table 10.1 has a good dynamic response performance, but the MAFPLL, DSCPLL, and ZCDPLL usually need the PIDtype prefilter or extra phase compensator to improve the dynamic performance, which may increase the computational burden. In addition, the ZCDPLL is not suggested to be applied directly in the gridconnected power system, since it is frequency nonadaptive and its harmonic filtering capability is quite limited. It shows that almost all of the PLLs in Table 10.1 benefit from different disturbance rejection capacity and there will be a direct relation between the their filtering capability and computational burden, which means the harmonic filtering capability of PLLs can be improved by adding more extended modules but at the cost of a higher computational burden. In a short conclusion, the operating principle of threephase PLLs was explained and their advantages and disadvantages were briefly discussed in this section. The provided information can be very useful for researchers to select a proper synchronization technique for their particular application.
10.4 Singlephase PLLs In recent years, singlephase PLLs have been widely applied for monitoring and diagnostic purposes in the power electric system and energy areas. With the development of a large number of singlephase PLLs with different topologies and properties, it is necessary to provide an insight into the characteristics of different singlephase PLLs. The details of each typical singlephase PLLs are described in the following section.
10.4.1 Standard PPLLs In fact, the singlephase PLLs are mainly categorized into the powerbased PLLs (PPLLs) and quadrature signal generationebased PLLs (QSGPLLs) [28]. Fig. 10.16 shows the schematic diagram of a standard PPLL, which is a very basic PLL in singlephase applications [28].
FIGURE 10.16 Schematic diagram of a singlephase standard powerebased phaselocked loop.
288 Control of Power Electronic Converters and Systems
b g and b In this standard PPLL, V is the singlephase signal input, u q are the estimated amplitude of frequency and phase, un is the nominal value of the frequency, and the PI controller serves as the LF of PLLs. The PPLL uses a mixer or producttype PD to generate the phase error information. Such PD has a doublefrequency term, which may cause a doublefrequency oscillatory error and DC offset error in the PPLL. The shortcoming of a standard PPLL is that it does not provide any information about the input voltage amplitude. It implies a decoupling between PPLL dynamics and the input voltage amplitude variations.
10.4.2 Low pass filterebased PPLLs In order to mitigate the doublefrequency disturbance terms in the standard PPLL, a PPLL equipped with a highorder infinite impulse response (IIR) or LPF is proposed in Ref. [40]. The schematic diagram of LPFPPLL is shown in Fig. 10.17. Notice that the LPF requires a low cutoff frequency to remove the doublefrequency disturbance item, which results in a considerably high phase delay in the PLLs and causes a very slow dynamic performance. However, the LPFPPLL benefits from a high harmonic filtering capability and noise immunity. Tuning the parameters of the LPFPPLL is carried out by the trialanderror procedure suggested in Ref. [40].
10.4.3 Moving average filterebased PPLLs In a similar way, Fig. 10.18 shows an alternative approach to eliminate the doublefrequency disturbance term by using an inloop MAF in the PPLL [15,41]. The MAF is a linear phaseebased LPF, which is also known as a rectangular window filter, which may cause a slow dynamic response. To mitigate the effect of the dynamics of MAFPLL, the PI controller could be instead a PID controller and arranging a polezero cancellation inside the control loop [15]. However, the polezero cancellation in MAFPLL reduces the ability of rejecting the doublefrequency grid disturbance. Therefore, the
FIGURE 10.17 Schematic diagram of a low pass filter powerebased phaselocked loop PLL.
Phaselocked loops and their design Chapter  10
289
frequencyadaptiveebased MAF is suggested in Ref. [42], and the frequency adaptability of MAFPLL is designed by using a variable sampling frequency and altering the MAF window length according to the frequency variations of the singlephase input signal. In addition, the MAFs can be achieved with both recursive and nonrecursive forms [28]. The recursive realization is usually preferred with lower computational burden, and the amplitude estimation function can be obtained by adding more MAFs into the selected PLLs.
10.4.4 Notch filterebased PPLLs A nonadaptive IIRNFPPLL is proposed to cancel the doublefrequency items in singlephase applications [43]. The structure is similar to the LPFPPLL except that the LPF is replaced by an IIRNF. However, the notch bandwidth of the IIRNF highly depends on the expected range of the grid frequency variations. Therefore, a wide bandwidth for the NF should be carefully designed under the condition of large frequency drifts. Furthermore, it should be noted that the selection effectively removes the doublefrequency term at the expense of causing a large phase delay, and hence, slowing down the transient response of PPLL. On the other hand, the harmonic filtering capability of IIRNFPLLs can be enhanced by additional NFs in a serial manner or in a parallel configuration, and the tuning of the control parameters of IIRNFPPLL can be obtained by using the symmetrical optimum methods [28]. Moreover, the finite impulse response notch filterebased PLL (FIRNFPPLL) is proposed to reject the doublefrequency term in Ref. [44]. The schematic diagram of FIRNFPPLL can be seen in Fig. 10.19. The FIRNF is implemented by using T/4 delay units (T is the grid fundamental period), rather effectively blocks the doublefrequency term even the grid frequency variations are large. In addition, the FIRNFPPLL rejects some specific
FIGURE 10.18 Schematic diagram of the moving average filter powerebased phaselocked loop.
290 Control of Power Electronic Converters and Systems
FIGURE 10.19 Schematic diagram of the finite impulse response notch filterebased phaselocked loop.
harmonic components inside the control loop. However, these advantages are at the expense of a slower transient response of the PPLL due to large phase delay caused by FIRNF unit.
10.4.5 Doublefrequency and amplitude compensationebased PPLLs The schematic diagram of DFACPPLL is shown in Fig. 10.20 [28,45], where an amplitude estimation loop is added to the standard PPLL, and thereby the doublefrequency terms are canceled by compensating the opposite doublefrequency components. On contrary to other singlephase PPLLs, the
FIGURE 10.20 Schematic diagram of the doublefrequency and amplitude compensationebased phaselocked loop [28,45].
Phaselocked loops and their design Chapter  10
291
dynamics of the DFACPPLL is independent of the grid amplitude variations. And the firstorder LPFs used in the DFAC structure contribute to higher harmonic filtering capabilities. It is worth mentioning that the extended version of the DFACPPLL with an excellent harmonic filtering capability is at the cost of slower dynamic response in practice.
10.4.6 Modified mixer PDBased PPLLs A modified mixer PDBased PPLL (MMPDPPLL) to tackle the doublefrequency problem is proposed in Ref. [46]. In this approach, the doublefrequency item is canceled before the PI controller by adding an equal but opposite component, which is constructed by using the information of the estimated phase angle. This approach works effectively in the condition that the grid voltage is always assumed as 1.0 p.u. However, it is difficult to create this condition since an accurate amplitude estimation in real time is required.
10.4.7 Transfer delayedebased PLLs Another important type of singlephase PLL is the QSGPLLs. However, most of QSGPLLs are the variants from the conventional SRFPLL (see Fig. 10.3) with additional filter/circuit/algorithms to generate the required quadrature signals. In Fig. 10.21, the TDPLLs based on the transfer delay create a quadrature signal as one typical approach. The quadrature signal is obtained by delaying the original singlephase signal by T/4, where T is the grid fundamental frequency period. On the other hand, it should be noted that the delayed signal is not perfectly orthogonal to the original signal under grid frequency variations/drifts cases and it is easy to cause the doublefrequency and offset errors in the estimated quantities by the standard structure of the TDPLL. Therefore, some extended TDPLLs have been developed such as NTDPLL [47], ETDPLL [48], ATDPLL [49], VTDPLL [50], etc.
FIGURE 10.21 Schematic diagram of a standard transfer delayedebased phaselocked loop.
292 Control of Power Electronic Converters and Systems
TABLE 10.2 Performance comparison between some singlephase PLLs in terms of calculation operation. Calculation operations Storage samples
Trigonometric functions
Basic calculation (þO)
NTDPLL [47]
80
0
0
ETDPLL [48]
140
0
22
ATDPLL [49]
40
2
4
VTDPLL [50]
80
0
124
As one of the most concerns, their computational burden should be carefully considered. Therefore, the number of operations required for the quadrature signal generation in different TDPLLs is summarized in Table 10.2. Based on these results, it seems that the ATDPLL and ETDPLL are the best choices. However, the VTDPLL with a linear interpolation can be an interesting option when the sampling frequency is high, as it requires lower calculations in practice. The NTDPLL and the VTDPLL are not recommended, as the former suffers from doublefrequency oscillations/ripples under frequency drifts and the latter requires a high computational effort. It should be noted that the ETDPLL is the only available choice among TDPLLs when a high harmonic filtering capability is required.
10.4.8 Inverse park transformationebased PLLs The inverse park transformationebased PLL (IPTPLL) is one of another popular singlephase PLLs and it is shown in Fig. 10.22. Its virtual orthogonal signal is generated by applying the IPT to the filtered dqaxis voltage components. Note that an inloop phase delay may be caused by LPF inside the IPTPLL, but the filtering capability and noise immunity are increased. More control design and detailed analysis of the IPTPLL can be found in Refs. [28,40]. Furthermore, in addition to the fundamental component, extracting the DC offset and some harmonic components may be required for the faulty grid conditions, an extended IPTPLL structure is shown in Fig. 10.23 [51]. However, selecting the number of filtering modules involves a tradeoff between the detection accuracy and computational burden, which should be carefully considered.
Phaselocked loops and their design Chapter  10
293
FIGURE 10.22 Schematic diagram of a standard inverse park transformationebased phaselocked loop.
FIGURE 10.23 Extended structure of the inverse park transformationebased phaselocked loop [51].
10.4.9 Generalized integratorebased PLLs Generalized integratorebased PLLs (GIPLLs) are also very attractive since they can be effectively customized according to different grid scenarios. In Fig. 10.24, the singlephase SOGIPLL can create a virtual quadrature signal and attenuate the harmonic components in the same structure. Notice that the
294 Control of Power Electronic Converters and Systems
FIGURE 10.24 Structure of standard singlephase secondorder generalized integratorebased phaselocked loop.
frequency estimated by the PLL is fed back to the SOGI for adapting it to the grid frequency variations. The smallsignal modeling and control design of the singlephase SOGIPLL can be found in Ref. [52]. In addition, some attempts for frequencyfixed SOGIPLLs (FFSOGIPLLs) have been proposed in recent years [53]. In these PLLs, the resonance frequency of SOGIs is fixed at the nominal grid frequency. Fig. 10.25 shows the structure for implementing the FFSOGIPLLs, the b g /un to create baxis output of the nonadaptive SOGI is multiplied by u balanced twophase signals for the PLL input. The aaxis component has a phase difference with the grid voltage caused by SOGI, which results in a phase error in the PLL output, which can be corrected by a phase error compensator. Fig. 10.26 shows another approach for implementing the FFSOGIPLLs [54], which has employed two nonadaptive SOGIQSGs in the PLL. The first SOGIQSG operates as a prefilter, which may cause phase and amplitude errors under grid frequency drifts. The second SOGIQSG generates the same phase shift angle and amplitude to compensate the measured errors. More analysis and design of the FFSOGIPLLs can be found in Ref. [54].
FIGURE 10.25 Structure of singlephase integratorebased phaselocked loop.
frequencyfixed
secondorder
generalized
Phaselocked loops and their design Chapter  10
295
FIGURE 10.26 Extended structure of singlephase frequencyfixed secondorder generalized integratorebased phaselocked loop.
10.4.10 Synthesis circuit PLLs The SCPLLs are the simplest extended singlephase SRFPLL due to its easy implementation and characteristics [28]. The schematic diagram of standard SCPLL is shown in Fig. 10.27. In addition, when it is necessary to extract more information of grid voltage phase, amplitude, or more harmonic components, the structure of SCPLL can be extended as shown in Fig. 10.28. However, the computations will increase, especially considering the grid voltage harmonic effects. It is worth mentioning that the SCPLL demonstrates a very similar performance to the IPTPLL. One of the other attractive PLLs in singlephase applications is the EPLL, which is mathematically equivalent to the standard SCPLL [55]. However, the EPLL has modified the integrator output signal as the estimated frequency, which make a higher filtering capability and a more damped dynamic response in the frequency estimation. Furthermore, the extended EPLLs have been proposed in Refs. [38,56] to improve the ability of harmonic and DC offset components rejection. It should be mentioned that these strategies are also applicable to other singlephase and threephase PLLs, which could be applied to modern distributed power systems [57,58].
FIGURE 10.27 Schematic diagram of the standard synthesis circuitebased phaselocked loop.
296 Control of Power Electronic Converters and Systems
FIGURE 10.28 Extended schematic diagram of the synthesis circuitebased phaselocked loop.
10.4.11 Performance comparison and recommendation A summary of the mentioned singlephase PLLs is presented in Fig. 10.29 and a performance comparison between these singlephase PLLs can be summarized in Table 10.3. The signalphase PLL can be devided into two main categories.
FIGURE 10.29 Two main categories of the typical singlephase phaselocked loops.
Phaselocked loops and their design Chapter  10
297
TABLE 10.3 Performance comparison between singlephase phaselocked loops mentioned in Fig. 10.29. Features
PLLs type
Doublefrequency robustness
Harmonic robustness
Computational burden
Dynamic response
LPFPPLL [40]
Good
Medium
Low
Slow
MAFPPLL [41]
Good
Medium
Low
Slow
NFPPLL [43]
Good
Medium
Low
Slow
DFACPPLL [45]
Good
Good
Average
Average
MMPDPPLL [46]
Good
Good
Average
Average
TDPLL [47]
Good
Medium
Low
Fast
IPTPLL [40]
Good
Good
Average
Fast
GIPLL [52e54]
Good
Good
Average
Fast
Standard EPLL [55]
Good
Low
Low
Fast
SCPLL [28,59]
Good
Low
Low
Fast
Based on the performance comparison, the LPFPPLL [40], MAFPPLL [41], and NFPPLL [43] are not suggested due to their slow dynamic performance and limited harmonic filtering capacities. The DFACPPLL [45] and MMPDPPLL [46] provide some interesting and optimized features, but the overall design complexity and computational efforts of the PLLs are increased. In addition, the TDPLL and its extensions [47e50], SCPLL [28,59], and standard EPLL [55] are good options only when the grid voltage contains little harmonics since they have limited harmonic filtering capacity. In practice, the best possible choices are probably the IPTPLL [40] and GIPLLs [52e54] since they can provide a satisfactory compromise between the dynamic response, harmonic filtering capacity, and computational complexity, and they can be effectively customized for adverse grid conditions. Compare with the threephase PLLs, the singlephase PLL algorithm still presented the fast responses to frequency and phase disturbances, but they are more stiff to singlephase voltage sags and harmonics due to the low cutoff frequency of their designed filters. However, the threephase PLL algorithm
298 Control of Power Electronic Converters and Systems
is more efficient from the computational point of view, which is demanding about half the number of calculations on threephase detector stage than the singlephase PLL algorithm. And the threephase PLL may use three generated singlephase input signal to operate as normal working case.
10.5 Summary This chapter provides a comprehensive overview of the recent attempts for designing advanced threephase and singlephase PLLs to be applied in the power electronic area. The operating principle of both threephase and singlephase PLLs is carefully explained. Furthermore, the different features and applications of the PLLs are discussed. The research considers the enhancement of the PLLs filtering capacities, different grid disturbance rejections, control structure simplification, dynamic response performance, and proper compromise selections under different grid conditions. Finally, the performance comparison guidance is provided and that could be a quick reference for proper selections of appropriate PLLs for researchers and engineers.
References [1] H. de Bellescize, La Reception Synchrone, E. Chiron., June 1932. [2] F. Blaabjerg, R. Teodorescu, M. Liserre, A.V. Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (October 2006) 1398e1409. [3] A. Cataliotti, V. Cosentino, S. Nuccio, A phaselocked loop for the synchronization of power quality instruments in the presence of stationary and transient disturbances, IEEE Trans. Instrum. Meas. 56 (6) (December 2007) 2232e2239. [4] M. KarimiGhartemani, M. Iravani, Measurement of harmonics/interharmonics of timevarying frequencies, IEEE Trans. Power Deliv. 20 (1) (January 2005) 23e31. [5] G. Zhang, et al., ADALINE network based PLL for position sensorless interior permanent magnet synchronous motor drives, IEEE Trans. Ind. Electron. 31 (2) (Febuary 2016) 1450e1460. [6] T. Tran, T. Chun, H. Lee, et al., PLL based seamless transfer control between grid connected and islanding modes in gridconnected inverters, IEEE Trans. Power Electron. 29 (10) (October 2014) 5218e5228. [7] M. Ohrstrom, L. Soder, Fast protection of strong power systems with fault current limiters and PLLaided fault detection, IEEE Trans. Power Deliv. 26 (3) (July 2011) 1538e1544. [8] G. Chen, L. Zhang, R. Wang, et al., A novel SPLL and voltage sag detection based on LES filters and improved instantaneous symmetrical components method, IEEE Trans. Power Electron. 30 (3) (March 2015) 1177e1188. [9] S. Golestan, et al., Threephase PLLs: a review of recent advances, IEEE Trans. Power Electron. 32 (3) (March 2017) 1894e1907.
Phaselocked loops and their design Chapter  10 [10]
[11]
[12]
[13] [14]
[15] [16]
[17]
[18] [19] [20] [21]
[22]
[23]
[24]
[25]
[26]
299
M. KarimiGhartemani, M. Iravani, A method for synchronization of power electronic converters in polluted and variablefrequency environments, IEEE Trans. Power Syst. 19 (3) (August 2004) 1263e1270. P. Rodriguez, A. Luna, I. Candela, et al., Multiresonant frequencylocked loop for grid synchronization of power converters under distorted grid conditions, IEEE Trans. Ind. Electron. 58 (1) (January 2011) 127e138. Z. Xin, R. Zhao, F. Blaabjerg, et al., An improved flux observer for fieldoriented control of induction motors based on dual secondorder generalized integrator frequencylocked loop, IEEE J. Emerg. Sel. Top. Power Electron. 5 (1) (March 2017) 513e525. S. Golestan, J.M. Guerrero, J.C. Vasquez, DCoffset rejection in phaselocked loops: a novel approach, IEEE Trans. Ind. Electron. 63 (8) (August 2016) 4942e4946. M. Ciobotaru, V.G. Agelidis, R. Teodorescu, F. Blaabjerg, Accurate and lessdisturbing active antiislanding method based on PLL for gridconnected converters, IEEE Trans. Power Electron. 25 (6) (June 2010) 1576e1584. S. Golestan, et al., Moving average filter based phaselocked loops: performance analysis and design guidelines, IEEE Trans. Power Electron. 29 (6) (June 2014) 2750e2763. E. Robles, S. Ceballos, J. Pou, et al., Variablefrequency gridsequence detector based on a quasiideal lowpass filter stage and a phaselocked loop, IEEE Trans. Power Electron. 25 (10) (October 2010) 2552e2563. J.Y. Wang, J. Liang, F. Gao, et al., A method to improve the dynamic performance of moving average filterbased PLL, IEEE Trans. Power Electron. 30 (10) (October 2015) 5978e5990. I. Carugati, S. Maestri, P.G. Donato, et al., Variable sampling period filter PLL for distorted threephase systems, IEEE Trans. Power Electron. 27 (1) (January 2012) 321e330. S. Golestan, et al., MAFPLL with phaselead compensator, IEEE Trans. Ind. Electron. 62 (6) (June 2015) 3691e3695. F.D. Freijedo, et al., An optimized implementation of phase locked loops for grid applications, IEEE Trans. Instrum. Meas. 60 (9) (September 2011) 3110e3119. F. GonzalezEspin, et al., An adaptive control system for threephase photovoltaic inverters working in a polluted and variable frequency electric grid, IEEE Trans. Power Electron. 27 (10) (October 2012) 4248e4261. S. Eren, M. KarimiGhartemani, A. Bakhshai, Enhancing the threephase synchronous reference frame PLL to remove unbalance and harmonic errors, in: Proc. 35th Annu. Conf. IEEE Ind. Electron., November 2009, pp. 437e441. P. Rodriguez, R. Teodorescu, I. Candela, et al., New positivesequence voltage detector for grid synchronization of power converters under faulty grid conditions, in: IEEE Power Electronics Specialists Conference, 2006, pp. 1e7. M. Ciobotaru, R. Teodorescu, F. Blaabjerg, A new single phase PLL structure based on second order generalized integrator, in: IEEE Power Electronics Specialists Conference, 2006, pp. 1e6. P. Xiao, K.A. Corzine, G.K. Venayagamoorthy, Multiple reference framebased control of threephase PWM boost rectifiers under unbalanced and distorted input conditions, IEEE Trans. Power Electron. 23 (4) (July 2008) 2006e2017. X. Guo, W. Wu, Z. Chen, Multiplecomplex coefficientfilterbased phaselocked loop and synchronization technique for threephase grid interfaced converters in distributed utility networks, IEEE Trans. Ind. Electron. 58 (4) (April 2011) 1194e1204.
300 Control of Power Electronic Converters and Systems [27] W. Li, X. Ruan, C. Bao, et al., Grid synchronization systems of threephase gridconnected power converters: a complex vectorfilter perspective, IEEE Trans. Ind. Electron. 61 (4) (April 2014) 1855e1870. [28] S. Golestan, J.M. Guerrero, J.C. Vasquez, Singlephase PLLs: a review of recent advances, IEEE Trans. Power Electron. 32 (12) (December 2017) 9013e9030. [29] J. Svensson, M. Bongiorno, A. Sannino, Practical implementation of delayed signal cancellation method for phasesequence separation, IEEE Trans. Power Deliv. 22 (1) (January 2007) 18e26. [30] Y.N. Batista, H.E.P. de Souza, F.A.S. Neves, et al., Variable structure generalized delayed signal cancellation PLL (VSGDSCPLL) to improve convergence time, IEEE Trans. Ind. Electron. 62 (11) (November 2015) 7146e7150. [31] W. Liu, F. Blaabjerg, D. Zhou, S.F. Chou, Modified instantaneous power control with phase compensation and currentlimited function under unbalanced grid faults, IEEE J. Emerg. Sel. Top. Power Electron. (2020), https://doi.org/10.1109/JESTPE.2020.2984475. [32] L. Hadjidemetriou, E. Kyriakides, F. Blaabjerg, A robust synchronization to enhance the power quality of renewable energy systems, IEEE Trans. Ind. Electron. 62 (8) (August 2015) 4858e4868. [33] S. Golestan, M. Monfared, F.D. Freijedo, Designoriented study of advanced synchronous reference frame phaselocked loops, IEEE Trans. Power Electron. 28 (2) (Febuary 2013) 765e778. [34] A.V. Timbus, R. Teodorescu, F. Blaabjerg, et al., PLL algorithm for power generation systems robust to grid voltage faults, in: Proc. 37th IEEE Power Electron. Spec. Conf., June 2006, pp. 1e7. [35] B. Liu, F. Zhou, Y. Zhu, et al., A threephase PLL algorithm based on signal reforming under distorted grid conditions, IEEE Trans. Power Electron. 30 (9) (September 2015) 5272e5283. [36] F.A.S. Neves, et al., A spacevector discrete Fourier transform for unbalanced and distorted threephase signals, IEEE Trans. Ind. Electron. 57 (8) (August 2010) 2858e2867. [37] F.D. Freijedo, A.G. Yepes, O. Lopez, et al., Threephase PLLs with fast postfault retracking and steadystate rejection of voltage unbalance and harmonics by means of lead compensation, IEEE Trans. Power Electron. 26 (1) (January 2011) 85e97. [38] M. KarimiGhartemani, S.A. Khajehoddin, P. Jain, et al., Addressing DC component in PLL and notch filter algorithms, IEEE Trans. Power Electron. 27 (1) (January 2012) 78e86. [39] S. Golestan, J.M. Guerrero, G. Gharehpetian, Five approaches to deal with problem of DC offset in phaselocked loop algorithms: design considerations and performance evaluations, IEEE Trans. Power Electron. 31 (1) (January 2016) 648e660. [40] R.M. Santos Filho, et al., Comparison of three singlephase PLL algorithms for UPS applications, IEEE Trans. Ind. Electron. 55 (8) (August 2008) 2923e2932. [41] M.A. Perez, et al., A robust phaselocked loop algorithm to synchronize staticpower converters with polluted AC systems, IEEE Trans. Ind. Electron. 55 (5) (May 2008) 2185e2192. [42] I. Carugati, P. Donato, S. Maestri, et al., Frequency adaptive PLL for polluted singlephase grids, IEEE Trans. Power Electron. 27 (5) (May 2012) 2396e2404. [43] F.D. Freijedo, J. DovalGandoy, O. Lopez, et al., Robust phase locked loops optimized for DSP implementation in power quality applications, in: Proc. IEEE 34th Annu. Conf. Ind. Electron. Soc., November 2008, pp. 2390e2395.
Phaselocked loops and their design Chapter  10 [44] [45]
[46] [47] [48]
[49]
[50] [51]
[52]
[53] [54] [55] [56] [57]
[58] [59]
301
A. Elrayyah, Y. Sozer, M. Elbuluk, Robust phase lockedloop algorithm for singlephase utilityinteractive inverters, IET Power Electron. 7 (5) (May 2014) 1064e1072. S. Golestan, M. Monfared, F.D. Freijedo, et al., Design and tuning of a modified powerbased PLL for singlephase gridconnected power conditioning systems, IEEE Trans. Power Electron. 27 (8) (August 2012) 3639e3650. T. Thacker, et al., Phaselocked loop noise reduction via phase detector implementation for singlephase systems, IEEE Trans. Ind. Electron. 58 (6) (June 2011) 2482e2490. S.B. Kjaer, Design and Control of an Inverter for Photovoltaic Applications, Ph.D. dissertation, Inst. Energy Technol., Aalborg Univ, Aalborg, Denmark, 2005. S. Golestan, et al., Smallsignal modeling, stability analysis and design optimization of singlephase delaybased PLLs, IEEE Trans. Power Electron. 31 (5) (May 2016) 3517e3527. S. Golestan, et al., An adaptive quadrature signal generation based single phase phaselocked loop for gridconnected applications, IEEE Trans. Ind. Electron. 64 (4) (April 2017) 2848e2854. A. Ozdemir, I. Yazici, Fast and robust softwarebased digital phaselocked loop for power electronics applications, IET Gener. Transm. Distrib. 7 (12) (May 2013) 1435e1441. L. Hadjidemetriou, Y. Yang, E. Kyriakides, F. Blaabjerg, A synchronization scheme for single phase gridtied inverters under harmonic distortion and grid disturbances, IEEE Trans. Power Electron. 32 (4) (April 2017) 2784e2793. M. Ciobotaru, R. Teodorescu, F. Blaabjerg, A new singlephase PLL structure based on second order generalized integrator, in: Proc. 37th IEEE Power Electron. Spec. Conf., June 2006, pp. 1511e1516. F. Xiao, L. Dong, L. Li, X. Liao, A frequencyfixed SOGI based PLL for singlephase gridconnected converters, IEEE Trans. Power Electron. 32 (3) (March 2017) 1713e1719. Q. Guan, Y. Zhang, Y. Kang, et al., Singlephase phaselocked loop based on derivative elements, IEEE Trans. Power Electron. 32 (6) (June 2017) 4411e4420. M.K. Ghartemani, Enhanced PhaseLocked Loop Structures for Power and Energy Applications, WileyIEEE Press, New York, NY, USA, 2014. M.K. Ghartemani, S.A. Khajehoddin, P.K. Jain, et al., Problems of startup and phase jumps in PLL systems, IEEE Trans. Power Electron. 27 (4) (April 2012) 1830e1838. W. Liu, et al., An evaluation method for voltage dips in shipboard microgrid under quasibalanced and unbalanced voltage conditions, IEEE Trans. Ind. Electron. 66 (10) (October 2019) 7683e7693. W. Liu, et al., Power quality assessment in shipboard microgrids under unbalance and harmonic ac bus voltage, IEEE Trans. Ind. Appl. 55 (1) (January 2019) 7683e7693. F. Xiong, et al., A novel frequencyadaptive PLL for singlephase grid connected converters, in: Proc. of IEEE Energy Convers. Congr. Expo., September 2010, pp. 414e419.
This page intentionally left blank
Chapter 11
Stability and robustness improvement of power converters Xiongfei Wang, Yicheng Liao, Zichao Zhou Aalborg University, Aalborg, Denmark
11.1 Introduction There have been increasing deployments of converterbased power sources and loads in electric power systems. The stability and control of power converters tend to be more sensitive to the variabilities and uncertainties of the connected electrical systems. A number of power disruptions that are resulted from the instability of grideconverter interactions have been reported in recent years. Voltage source converters (VSCs) are commonly used with converterbased power sources and loads. Differing from synchronous generators, VSCs enable full control of electric power, and thus the control dynamics of VSCs play a critical role in grideconverter interactions. There have been numerous control schemes developed for gridconnected VSCs. The vector current control, among others, is still the mostly used approach to regulating the power exchanged between VSCs and the grid. The dynamic analysis and controller design for currentcontrolled VSCs have been extensively studied over the years. The focus of research has also been shifted from the reference tracking of control loops under various disturbances, such as the voltage harmonics and unbalances, to the stability and robustness of VSCs with a wide range of grid impedances, e.g., a low short circuit ratio (SCR) grid, and series/parallel resonant impedance. In this chapter, the stability and robustness improvements of currentcontrolled VSCs are addressed. The dynamic impacts of control loops, including the inner current loop, the phaselocked loop (PLL), and the outer DClink voltage control (DVC) loop, on the stability robustness of VSCs are systematically discussed. In Section 11.2, the smallsignal modeling of current control for a threephase VSC with L or LCL filter, where both the Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000160 Copyright © 2021 Elsevier Ltd. All rights reserved.
303
304 Control of Power Electronic Converters and Systems
converter and gridside current control are considered. The theoretical basis and assumption for representing the inner current loop by singleinput and singleoutput (SISO) transfer functions are elaborated. Then, the impedance modeling of the inner current control loop is discussed. Based on the impedance modeling of inner current loop, the frequencydomain passivitybased analysis is introduced to evaluate the stability robustness of current control, where the effect of time delay involved in the current control is illustrated. Then, three approaches for improving the passivity of inner current control are discussed, including the time delay reduction, the passivitybased design of passive filters, and the active damping control. Following the SISO impedance modeling of inner current loop, the dynamic impacts of PLL and the outer DVC loop are discussed in Section 11.3. Unlike the inner current control loop, which is generally designed with symmetrical dynamics, the DVC loop and PLL lead to asymmetrical control dynamics on the d and qaxes, which necessitate the use of multipleinput and multipleoutput (MIMO) model to analyze their dynamic effect. Further, the PLL and DVC loop are nonlinear and time varying, whose smallsignal dynamics are highly dependent on the operating (equilibrium) points of VSCs. The MIMO impedance modeling of VSCs, considering the inner current loop, the PLL, and the DVC loop, is illustrated step by step. It is shown that the DVC loop can destabilize the grideconverter interaction when VSCs operate in the rectifier mode, while the PLL can induce oscillations when VSCs operate in the inverter mode. Then, the generalized Nyquist stability criterion is applied to predict the stability of VSCs in weak grids. This is followed by the controller parameter tuning of the PLL and DVC for improving the stability robustness.
11.2 Stability and robustness improvement of current control Fig. 11.1 shows a singleline representation of a threephase VSC with L or LCL filter. The current control can be performed on either the converter side or
(A) Vdc
=
L1
(B)
i1 v
Vdc
≈
=
d
PWM
u
Gi i1
L2 e
i2 Cf
≈ d
θ d
L 1 i1
iref
(Id+jIq)e
jθ
PLL Id Iq
θ
v d
PWM
u
Gi i2
iref
jθ
(Id+jIq)e
PLL
e
Id Iq
FIGURE 11.1 Singleline representation of a threephase voltage source converter. (A) Converterside current control; (B) Gridside current control.
Stability and robustness improvement of power converters Chapter  11
305
the grid side. The current control performance is usually determined by a PLL and an innerloop current controller [1]. The PLL provides the phase reference of the voltage for the current control, which has slower dynamics than the innerloop current control, thus it is usually neglected for analyzing the current control dynamics [2]. The current control has faster dynamics, whose control bandwidth (BW) can be high up to onetenth of the converter switching frequency, yet still with enough stability margin [2]. Thus, the time delay introduced by converter digital control can have a significant impact on the current control dynamics, which also tends to result in harmonic instability issues in gridconnected applications [3]. This section derives the smallsignal model of current control loop first, then performs the stability analysis and introduces different ways to robustness enhancement.
11.2.1 Smallsignal modeling The smallsignal models of the VSC with the converterside current control and the gridside current control are derived, respectively, which are based on a linearization of the converter power stage.
11.2.1.1 Linearization of the converter power stage Fig. 11.2 gives the circuit diagram of a VSC with switches. The switching signals for different phases are defined as sa, sb, and sc, which can be either 1 or 0 to represent the on or off state of the upper switch or the lower switch during a normal operation. They are thus determined by the status of the switches, s1es6, i.e.,
idc
s1
s3
s5
ua ub
vdc
uc
i1a
L1
i1b
L1
i1c
L1
va vc Cf
s4
s6
L2 L2
ea
i2b i2c
L2
ec
i2a vb
Cf
Cf
s2
FIGURE 11.2 Switching model of the voltage source converter.
eb
306 Control of Power Electronic Converters and Systems
( sa ¼ sc ¼
1; s1 ¼ onXs4 ¼ off
0; s1 ¼ offXs4 ¼ on ( 1; s5 ¼ onXs2 ¼ off
( ; sb ¼
1; s3 ¼ onXs6 ¼ off 0; s3 ¼ offXs6 ¼ on
; (11.1)
0; s5 ¼ offXs2 ¼ on
Then the ACside voltage and DCside current through the switching modulation can be represented as 2 3 2 3 2 3 ua sa ia 6 7 6 7 6 7 (11.2) 4 ub 5 ¼ vdc 4 sb 5; idc ¼ ½ sa sb sc 4 ib 5: uc sc ic It can be seen from Eq. (11.2) that the converter system is a nonlinear and discontinuous system. By applying the averaging operator over the switching period [4], the switching ripples can be neglected. The duty cycles are defined as the averaged switching functions, i.e., Z t 1 da=b=c ¼ Sa=b=c ds: (11.3) Tsw tTsw Then a continuous dynamic model of the converter power stage is derived as
2
ua
3
2
da
3
2
6 7 6 7 4 ub 5 ¼ vdc 4 db 5; idc ¼ ½ da uc dc
db
ia
3
6 7 dc 4 ib 5:
(11.4)
ic
It is noted that the averaging operator determines the validity of the model till half of the switching frequency [4]. By linearizing the converter model, the smallsignal model for the ACside equivalent circuit can be derived as shown in Fig. 11.3, where the DCside dynamics can be neglected when Vdc is assumed as constant (vdc ¼ 0).
ua
ub dbVdc daVdc
uc
i1a
L1
i1b
L1
i1c
L1
dcVdc
va vc Cf
L2 L2
ea
i2b i2c
L2
ec
i2a vb Cf
eb
Cf
Dbvdc
Davdc
Dcvdc
Neglected with constant Vdc
FIGURE 11.3 Linearized ACside equivalent circuit of the voltage source converter.
Stability and robustness improvement of power converters Chapter  11
307
Under threephase balanced conditions, the threephase circuit in the abc frame can be transformed into a twophase system in the ab frame. If the converter current control is assumed symmetrical in the dq frame [5] or identical in the ab frame, thus, the twophase stationaryframe model can be further simplified as a singlephase system. Considering the converter control, the stationaryframe model can be represented by Fig. 11.4. The switching modulation with Vdc is canceled by the normalization with 1/Vdc in the converter control.
11.2.1.2 Smallsignal model of VSC with converterside current control For the converterside current control, the filters Cf and L2 can be excluded for the converter modeling. Consequently, only an Lfilter is considered and the capacitor voltage v is seen as the disturbance input. According to the circuit in Fig. 11.4, the plant model from the modulation voltage u to the converter output current i1 and the converter openloop output admittance can be derived, respectively, as i1 1 ; (11.5) Yui ðsÞ ¼ ¼ u v¼0 sL1 i1 1 Yol ðsÞ ¼ ¼ : (11.6) v u¼0 sL1 The pulse width modulation (PWM) can be modeled by a zeroorder hold block, which usually introduces a time delay of 0.5Ts [6], where Ts is the sampling rate. Then, considering a oneperiod calculation delay of digital controller, the total time delay is characterized as 1.5Ts, which can be represented as a transfer function (Gd) in series with the current controller, i.e., Gd ðsÞ ¼ esTd ¼ e1:5sTs :
L1 u
i1
v
i2
(11.7)
L2 e
Cf
Vdc
d PWM
1/Vdc
Controller
FIGURE 11.4 Stationaryframe model of the voltage source converter with converterside current control.
308 Control of Power Electronic Converters and Systems
v iref
Gi
Gd
u
Yol Yui
i1
Ti FIGURE 11.5 Closedloop smallsignal model of the voltage source converter with converterside current control.
Then the closedloop control diagram for the converter can be depicted in Fig. 11.5, where Gi represents the transfer function of the current controller, which utilizes a proportional þ resonant (PR) controller in the ab frame, i.e., Gi ðsÞ ¼ Kpi þ
Kri s high frequency z Kpi ; s2 þ u21
(11.8)
where u1 is the fundamental angular frequency. The resonant controller gain, Kri, is merely designed for the steadystate tracking performance at u1, thus it has negligible effect on the frequency response at the higher crossover frequency [2]. Based on the control diagram in Fig. 11.5, the loop gain and the closedloop output admittance of the current control can be derived, respectively, as Ti ðsÞ ¼ Gi ðsÞGd ðsÞYui ðsÞ; Yo ðsÞ ¼
Yol ðsÞ ; 1 þ Gi ðsÞGd ðsÞYui ðsÞ
(11.9) (11.10)
which can be utilized for controller design and stability analysis.
11.2.1.3 Smallsignal model of VSC with gridside current control For the gridside current control, the converter smallsignal model can be derived in a similar way. The only difference lies in the transfer functions of the plant model Yui and the openloop admittance model Yol. They are derived based on the LCL filter with the voltage e as the disturbance input, which are represented as i2 ZCf Yui ðsÞ ¼ ¼ ; (11.11) u e¼0 ZCf ZL1 þ ZCf ZL1 þ ZL1 ZL2 i2 ZCf þ ZL1 Yol ðsÞ ¼ ¼ ; (11.12) e u¼0 ZCf ZL1 þ ZCf ZL1 þ ZL1 ZL2 where ZL1, ZL2, and ZCf are the impedances of L1, L2, and Cf, respectively. Then, the closedloop smallsignal model can be derived in the same way as shown in Fig. 11.5, with v replaced by e and i1 replaced by i2.
Stability and robustness improvement of power converters Chapter  11
309
11.2.2 Passivitybased stability analysis For a gridconnected VSC, its output admittance plays a significant role in the grideconverter interaction. The overall system stability is contributed by minor loop gain composed by the grid impedance and the converter output admittance at their point of connection [7]. However, the external grid impedance profile is usually unknown, which makes it difficult to predict the system stability based on a specific minor loop gain. Therefore, the passivitybased stability analysis can be utilizeddthe interconnected system is passive and stable if the converter output admittance and the grid impedance are both passive [8]. Such a method provides a sufficient condition for system stability. The grid impedance is composed of passive components, thus it is always passive. Then the stability performance can be simply analyzed by the converter output admittance. The passivity condition implied that the converter output admittance dissipates energy, i.e., the real part of the admittance model is nonnegative at all frequencies. If the converter admittance has negative real part, it tends to result in negative damping to the stability of grid interaction. The passivitybased stability analysis for converterside current control and gridside current control is thus analyzed, respectively.
11.2.2.1 Passivitybased stability analysis for converterside current control To further investigate the passivity of the converter closeloop output admittance, Yo can be further derived as two admittances in series [9], i.e., Yo ðsÞ ¼
1 ; 1 1 þ Yol ðsÞ Yoa ðsÞ
(11.13)
where one admittance is the openloop output admittance Yol, which is passive, and the other is an active admittance, which is defined as Yoa. For the converterside current control, it is derived that Yoa is merely determined by the current controller and the time delay, i.e., Yoa ðsÞ ¼
Yol ðsÞ 1 ¼ : Gi ðsÞGd ðsÞYui ðsÞ Gi ðsÞGd ðsÞ
(11.14)
Neglecting the effect of the resonant current controller, the highfrequency response of Yoa is derived as Yoa ð juÞ
high frequency
z
1 juTd 1 e ¼ ½cosðuTd Þ þ j sinðuTd Þ: Kpi Kpi
(11.15)
310 Control of Power Electronic Converters and Systems
Converter side i1 Gciiref
PoC
Yol Y Yoa o
Grid side v i2 L 2 Cf
Lg vg
FIGURE 11.6 Equivalent circuit of the gridconnected converter with converterside current control.
It is seen that Yoa( ju) is a function of time delay, whose real part is determined by the function of cos(uTd). Consequently, the negative real part of Yoa locates in the frequency range of (1/(4Td), 3/(4Td)), i.e., ( fs/6, fs/2) with Td ¼ 1.5Ts. Since Yol is passive, the nonpassive region of the total converter admittance Yo should also locate in the nonpassive region of Yoa. Consequently, the equivalent circuit of a gridconnected converter can be represented by Fig. 11.6. It is noted that the converter admittance only considers the converterside filter L1; thus, the filters Cf and L2 are included in the model of the grid admittance, which is given by Yg ðsÞ ¼ sCf þ
1 : sðL2 þ Lg Þ
(11.16)
The converterside current controller design and the passivitybased analysis are carried out based on the circuit and control parameters provided in Table 11.1. Fig. 11.7A plots the Bode diagram of the current loop gain, where the current control BW is designed as 1/10 of the switching frequency. Fig. 11.7B plots the Bode diagrams of the converter output admittance Yo and the grid admittance Yg. It is seen that the negative real part of Yo appears in the frequency range ( fs/6, fs/2), since its phase response has exceeded 90 degrees. This nonpassive behavior of Yo makes the phase difference between the Yo and Yg exceed 180 degrees within the frequency range where Yo>Yg, thus the gridconnected converter system is unstable according to the Nyquist stability criterion [7]. A simulation verification is provided in Fig. 11.8 for this scenario. It is seen that highfrequency oscillations are observed in the converter output current waveforms. The oscillation frequency is estimated as 1.73 kHz, which agrees with the admittance magnitude crossover frequency shown in Fig. 11.7B.
Stability and robustness improvement of power converters Chapter  11
311
TABLE 11.1 Circuit and control parameters of the gridconnected converter. Parameter
Symbol
Value
Parameter
Symbol
Value
Grid voltage
Vg
380 VRMS
Grid frequency
f1
50 Hz
DClink voltage
Vdc
650 V
Power rating
P0
2 kW
Grid inductor (ESR)
Lg (Rg)
10 mH (0.1 U)
Converterside inductor (ESR)
L1 (RL1)
2.2 mH (0.1 U)
Filter capacitor
Cf
10 mF
Converterside inductor (ESR)
L2 (RL2)
1 mH (0.1 U)
Switching frequency
fsw
10 kHz
Sampling frequency/ period
fs/Ts
10 kHz/100 ms
P gain of converterside current control
Kpi
13.8 U
R gain of converterside current control
Kpr
8685 U/s
P gain of gridside current control
Kpi
5.8 U
R gain of gridside current control
Kpr
1094 U/s
P gain of gridside current control with Cf ¼ 1 mF
Kpi
18 U
R gain of gridside current control with Cf ¼ 1 mF
Kpr
1129 U/s
(A)
(B) 1.73 kHz Bandwidth of 1 kHz
>180°
fs/2
fs/6
fs/2
FIGURE 11.7 Converterside current control design and admittance analysis for the gridconnected converter. (A) Current control loop gain; (B) Admittancebased passivity and stability analysis.
312 Control of Power Electronic Converters and Systems
FIGURE 11.8 Simulation result of the gridconnected converter with converterside current control.
11.2.2.2 Passivitybased stability analysis for converterside current control For the gridside current control, the output admittance can also be seen as the openloop admittance Yol in series with the active admittance Yoa. The latter one is derived as Yoa ðsÞ ¼
Yol ðsÞ ZCf þ ZL1 ¼ ; Gi ðsÞGd ðsÞYui ðsÞ ZCf Gi ðsÞGd ðsÞ
(11.17)
whose highfrequency response can be approximated as Yoa ð juÞ
high frequency1
z
L1 Cf u2 sTd 1 L1 Cf u2 e ¼ ½cosðuTd Þ þ j sinðuTd Þ: Kpi Kpi (11.18)
It is found that the passivity of Yoa is not only dependent on the time delay but also related to the resonant frequency of L1 and Cf, i.e., fL1C. It is thus further derived that the negative real part of Yoa locates in the frequency range of ( fL1C, 1/(4Td)), i.e., ( fL1C, fs/6) with Td ¼ 1.5Ts. It is noted that if fL1C is larger than 1/(4Td), the nonpassive region should be (1/(4Td), fL1C). The equivalent circuit for a gridconnected converter with the gridside current control is shown in Fig. 11.9. Since the LCL filters are internal parameters of the converter, the external grid admittance is merely composed of Lg. Based on the same circuit parameters in Table 11.1, the current control loop gain is designed with control BW of 300 Hz, whose Bode diagram is shown in Fig. 11.10A. It is noted that the BW of the gridside current control cannot be designed as too large, because of the impact of the LCL resonance in the highfrequency range. It is seen that the BW of 300 Hz results in a gain margin of 4.5 dB and a phase margin of 8.7 degrees. The admittance plots for
Stability and robustness improvement of power converters Chapter  11
Converter side i2 Gciiref
PoC
Yol Y Yoa o
313
Grid side e Lg vg
FIGURE 11.9 Equivalent circuit of the gridconnected converter with gridside current control.
(A)
(B) Bandwidth of 300 Hz
fL1C
GM = 4.5 dB
1.19 kHz
>180°
PM = 8.7°
fs/2
fs/6
fs/2
FIGURE 11.10 Gridside current control design and admittance analysis for the gridconnected converter. (A) Current control loop gain; (B) Admittancebased passivity and stability analysis.
the converter and the grid are shown in Fig. 11.10B. It is seen that the converter admittance only has the negative real part within the frequency range of ( fL1C, fs/6), whose phase response has exceeded þ90 degrees. At the magnitude crossover frequency of 1.19 kHz, the phase difference between the converter admittance and the grid admittance has exceeded 180 degrees, which results in instability. The simulation result for the gridconnected converter with the gridside current control is provided in Fig. 11.11. When the converter is connected, an oscillation at the frequency of 1.18 kHz can be observed in the current waveform, which also agrees with the stability analysis provided in Fig. 11.10B.
11.2.3 Robustness enhancement Based on the admittance passivity analysis, the stability of the gridconnected converter can be improved by reducing the nonpassive frequency range. There are several approaches to enhance the system robustness by improving the converter admittance passivity.
314 Control of Power Electronic Converters and Systems
FIGURE 11.11 Simulation result of the gridconnected converter with gridside current control.
11.2.3.1 Time delay reduction For the converterside current control, it has been found that the nonpassive region of the converter admittance locates within (1/(4Td), 3/(4Td)), which is merely dependent on the time delay. Therefore, an effective approach to improving the converter passivity is to reduce the time delay, which could push the nonpassive region to a higher frequency range, thus the admittance phase difference over 180 degrees can be prevented at the magnitude crossover frequency. There are several ways to reduce the time delay: (a) Increasing the sampling frequency to twice as the switching frequency and the double PWM update within a switching period, where Td ¼ 1.5Ts still holds but the value of Td is reduced [6]. (b) Shifting the sampling instant toward the PWM update instant, by which the calculation delay is reduced as 0.5Ts, then the total time delay reduces to Td ¼ Ts [10]. Considering the two ways, with a fixed switching frequency of 10 kHz, the output admittances of the converter compared against the grid admittance are plotted in Fig. 11.12. It can be seen that the passivity at the admittance magnitude crossover frequency has been improved by reducing the time delay, which makes the admittance phase difference less than 180 degrees and thus contributes to stability. Simulation results for the two cases are provided in Fig. 11.13, where the converter can be stabilized. It is noted that the time delay reduction approach may not well apply to the gridside current control, since the nonpassive region is ( fL1C, 1/(4Td)). In such a case, a reduction of time delay can even widen the nonpassive region.
Stability and robustness improvement of power converters Chapter  11
< T1 ¼ ma sinðp=6 qÞTs p p (13.1) T2 ¼ ma sinðp=6 þ qÞTs ; < q > 6 6 : T0 ¼ Ts T1 T2 where T1 and T2 are the total dwell time of the two adjacent active vectors, while T0 is the dwell time of zero vector. Ts is the carrier period and q indicates the current reference position in each sector.
13.3.2.2 Selective harmonic elimination SHE is an offline modulation scheme, which can eliminate a specific number of loworder unwanted harmonics in the output current. Normally, the pause angles are precalculated and imported into the digital controller. Fig. 13.10
FIGURE 13.9 Space vector modulation of current source converter (CSC). (A) CSC diagram, (B) space vector diagram.
Highpower current source converters Chapter  13
377
FIGURE 13.10 Example of selective harmonic elimination modulation.
shows a typical SHE waveform; there are five pulses per halfcycle with five switching angles in the firstquarter period, where only two out of the five angles are independent as the PWM is halfwave symmetric and the fifth and seventh harmonics can be eliminated. Multilevel SHE was proposed in Ref. [19] for parallel CSC applications, where the pause angle design will be more flexible, and more orders of unwanted harmonics can be eliminated to achieve better output current. However, the pause angle calculation is much more complex due to a large number of freedoms.
13.3.2.3 Direct dutyratio pulse width modulation The DDPWM was first introduced in Ref. [18], which can directly produce the gating signals of CSC without any logic translation. Besides the dual relationship, the isomorphic relationship between VSC and CSC is presented in Ref. [20], which can sufficiently reveal the essential principle of DDPWM. It is verified that the 3phase CSC and single phase 3level VSC are isomorphic pairs and they shared the same modulation features. To achieve DDPWM, the six switches are identified as Spmax/Snmax, Spmid/ Snmid, and Spmin/Snmin according to the magnitude of threephase references (IA*, IB*, and IC*). The output phases are renamed as MaxMidMin phase instead of ABC phase. The maximum, medium, and minimum values (Imax and Imin) among the threephase references can be obtained as Imax ¼ max (IA*, IB*, IC*) and Imin ¼ min (IA*, IB*, IC*) by comparing their magnitudes. In MaxMidMin phase, Imax, Imid, and Imin are synthesized by Spmax/Snmax, Spmid/Snmid, and Spmin/Snmin, respectively. The waveforms of Imax and Imin with balanced threephase references are shown in Fig. 13.11A. As it can be seen, Imax is always bigger than zero, and Imin is always smaller than zero. Therefore, Imax can be synthesized by using Idc and zero, and Imin can be synthesized by using Idc and zero. That means the lower leg of Max phase Snmax and upper leg of Min phase Spmin are never turned on (Snmax ¼ Spmin ¼ 0). As one and only one switch among the three upper and lower switches should be on, which can be restricted by Spmax þ Spmid þ Spmin ¼ 1 and Snmax þ Snmid þ Snmin ¼ 1. The logical switch relations are expressed as 1; if Imax >¼ C1 1; if Imin < C2 ; Snmin ¼ (13.2) Spmax ¼ 0; if Imax < C1 0; if Imin >¼ C2
378 Control of Power Electronic Converters and Systems
FIGURE 13.11 Direct dutyratio pulse width modulation principle. (A) Current source converter diagram in MaxMidMin phase, (B) digital implement.
Highpower current source converters Chapter  13
379
where C1 is the upper carrier and C2 is the lower carrier. Once the gating signals of Spmax, Spmid, Spmin, Snmax, Snmid, and Snmin are obtained based on Eq. (13.2), they can be reassigned to ABC phase according to the magnitudes of threephase current references as shown in Fig. 13.13B. For example, when Imax ¼ IC*, Imid ¼ IA*, and Imin ¼ IB*, the switching signals can be reassigned as SCp ¼ Spmax/SCn ¼ Snmax; SAp ¼ Spmid/SAn ¼ Snmid; and SBp ¼ Spmin/ SBn ¼ Snmin.
13.3.2.4 Comparison of different CSC modulations The PWM output features such as total harmonic distortion (THD), harmonic spectrum, switching frequency, as well as CMV are highly dependent on the sequence design. The PWM sequences of different modulations are compared as shown in Table 13.1, where the numbers 1 to 6 represent the six active vectors and 7 to 9 mean the three zero states. The PWM sequence of SVM is very flexible and different segment forms can easily be achieved. Normally, the threesegment SVM is adopted for highpower CSC application in order to reduce switching losses. Both SHE and TPWM just utilize the active states, which result in a relatively low switching frequency. Among the three carrierbased SPWMs, the BTSPWM has 6 switching actions in each sampling period and the resulted switching frequency is the same as the carrier frequency. Both SSDPWM and DDPWM are fivesegment PWM and the switching frequency is twothird of the carrier frequency. All three SPWMs adopt two adjacent active vectors and one zero
FIGURE 13.12 Pulse width modulation waveforms and harmonic distribution with different modulations.
380 Control of Power Electronic Converters and Systems
FIGURE 13.13 Thirdorder CMV of single CSC. (A) Conventional SVM, (B) BTSPWM, (C) SSDPWM, (D) DDPWM.
TABLE 13.1 Switching sequence with different modulations. Interval
SVM
SHE/ TPWM
BTSPWM
SSDPWM
DDPWM
[0e30 degrees]
618
6e1
8168618
16861
61716
[30e60 degrees]
618
6e1
7217127
16861
16961
[60e90 degrees]
127
1e2
7217127
21712
12921
[90e120 degrees]
127
1e2
9329239
21712
21812
[120e150 degrees]
239
2e3
9329239
32923
23832
[150e180 degrees]
239
2e3
8438348
32923
32723
16 are active states; 7e9 are zero states.
vector to synthesize the current references. However, the active vector order and zero state selection are different. The DDPWM divides each sector into two subsectors equally due to different active vector orders. The output PWM waveforms and their harmonic distribution of different modulations are shown in Fig. 13.12. The output frequency ( f0) is 60 Hz, and
Highpower current source converters Chapter  13
381
the carrier frequency ( fc) for SPWMs as well as timer frequency ( ft) for SVM are both 1080 Hz. The switching frequency difference is easy to determine as the conventional threesegment SVM enjoys the smallest, while the switching frequency of BTSPWM is the highest. Compared to BTSPWM, other modulations have wider side band harmonics and their overall harmonic performances are similar. Besides the harmonic performance, the CMV is also an inherent feature and the dominant thirdorder component is determined by the modulation index (ma) and the displacement angle (4) between phase voltage and current as shown in Fig. 13.13. It shows that the DDPWM enjoys the smallest CMV among different carrierbased SPWMs. Moreover, the CMV excited by conventional SVM is high and redundant zero state replacement method was discussed in Ref. [21] to reduce CMV. Since the PWM sequence design of SVM is very flexible and the DDPWM enjoys inherent small CMV as well as modularity, this chapter will continue to focus on these two modulations.
13.4 Parallel CSC and circuit analysis Generally, the parallel CSC configurations can be classified into an independent DClink structure and a shared DClink structure. For independent DClink structure, the subDClink current can be controlled independently; thus, CMV and CC are the main issues. On the other hand, the DC current balance is the priority task for shared DClink stricture, since only the total DClink current can be regulated. This section mainly introduces the CM loop circuit and DClink circuit analysis for parallel CSC structure with shared DClink.
13.4.1 CM loop circuit of parallel CSC The introduction had addressed the possible CMV resonance issue for transformerless CSCfed MV drives, which also posed a great challenge in parallel CSC systems. To have a detailed analysis, the CM loop circuit for parallel CSC system will be derived from that of a single CSC. The possible CMV resonance point, CC, and CMC features will be investigated through the circuit analysis. Fig. 13.14A shows the typical CM loop of a backtoback CSC with an integrated DC choke on the DC link. The neutral points of input (Cfr) and output filters (Cfi) are connected through a damping resistor and thus creates a CM loop to flow the CMC. With the help of a CM choke, the CMV stress on the motor can be effectively reduced, which can be expressed as Vog ¼ icm $R ¼ Vcmi Vcmr VL
(13.3)
where Vog is the CMV stress of the motor, R is the damping resistor, and icm is the total CMC flowing through the CM loop. VL is the voltage drop on the CM choke. Vcmi and Vcmr are the CSI and CSRside CMVs, which are defined as
382 Control of Power Electronic Converters and Systems
FIGURE 13.14 Commonmode loop circuit of single current source converter system. (A) Commonmode loop, (B) equivalent circuit.
383
Highpower current source converters Chapter  13
the average value of the positive and negative DCbus voltage and their values are changed with the switching actions of CSI and CSR which are given as 2 3 Va 6 7 6 7 VpN þ VnN 1 7 ¼ $½ S1 þ S4 S3 þ S6 S5 þ S2 $6 Vcm ¼ 6 Vb 7 ; 2 2 4 5 (13.4) Vc 8 < 1; switch on Si ¼ i ¼ 1; 2; .; 6 : 0; switch off where VpN and VnN are the positive and negative DCbus voltage with respective to grid or motorside neutral points (o or g) which is represented by N. The six switching semiconductors are named as S1 to S6, where 1 means the switch is turned on and 0 means it is off. Va, Vb, and Vc are the threephase output voltages. Based on Eq. (13.4), the possible CMV values under different switching states can be summarized as in Table 13.2, where the CMVs generated by the three zero states are the same as the phase voltage, while the CMVs are caused by the 6 active states that are half of phase voltages. To get a deeper look at the CM loop features, the equivalent CM loop circuit of single backtoback CSC system is shown as in Fig. 13.14B. It consists of the CM choke, equivalent capacitance of input and output filters, and two CMV sources generated by CSR and CSI, respectively. Therefore, the CMC can be expressed as Eq. (13.5), icm ¼
Vcmr þ Vcmi R j=u=Ceq þ ju$Lcm
(13.5)
where u is the CMV dominant angular frequency, Lcm represents the CM choke, Ceq represents the equivalent capacitance. The resonance frequency in this circuit can be expressed as 8 1 > < Fr ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2p Lcm $Ceq (13.6) > : Ceq ¼ 3Cfr $Cfi =ðCfr þ Cfi Þ
TABLE 13.2 Commonmode voltage of single current source converter. Type
Vector
Switching states
Commonmode voltage
Zero
I0
[14], [36], [52]
Va, Vb, Vc
Active
I1eI6
[16], [12], [23] [34], [45], [56]
0.5Vc, 0.5Vb, 0.5Va 0.5Vc, 0.5Vb, 0.5Va
384 Control of Power Electronic Converters and Systems
It indicates that the CM resonance frequency is related to the actual value of system parameters. The excited CMC would be very large on the resonance point, which needs to be addressed. As discussed in Ref. [16], only zerosequence components are present in the CMV waveform and the dominant component is the thirdorder component. The rectifier is usually operated under the fixed grid frequency (60 Hz), and the dominant CMV component is 180 Hz. Based on the typical parameters of a CSCbased MV drive system, the LC resonance frequency is normally located around 30 Hz. This means that the variable speed drive can excite the CM resonance when the motor operates at around 10 Hz. Thus, the thirdorder component of the CMV generated by CSI is addressed to suppress the CM resonance. The equivalent CM loop circuit and LC resonance point analysis of a single CSC can be extended into parallel CSC system. The parameters of paralleled CSC modules are assumed to be the same and the detailed equivalent CM loop circuit is shown in Fig. 13.15. The system CMC is influenced by both gridside and loadside CMVs. Meanwhile, the CC caused by the difference of
FIGURE 13.15 Equivalent commonmode loop circuit of parallel current source converter system with shared DC link and a damping resistor.
Highpower current source converters Chapter  13
385
switching actions between parallel modules also needs to be considered. The expressions of CC and CMC are shown in 8 Vcmi N Vcmi 1 > > > < icrN ¼ icm1 icmN ¼ 1=2juL diff
> Vcmr þ 1=N$ðVcmi 1 þ Vcmi 2 þ /Vcmi N Þ > > : icm ¼ icm1 þ icm2 þ /icmN ¼ R j=uCeq þ juLcm (13.7) where icmi_1, icmi_2, $$$ icmi_N are the CMC flowing through CSI1, CSI2, $$$ CSIN. icrN is the CC between CSC1 and CSCN, which is proportional to the CMV difference between two CSIs. The current icm flowing to the damping resistor is the total system CMC, which is influenced by the CSR and CSIside CMV and can be analyzed separately. The equivalent CMV on the CSIside is (Vcmi_1 þ Vcmi_2 þ $$$ þ Vcmi_N)/N. Similar to the single CSC, the CM resonance also exists in parallel CSC system and the CM resonance frequency for NCSC parallel system is given as pﬃﬃﬃﬃ N Fr N ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ (13.8) 2p Lcm $Ceq It shows that the resonance frequency of parallel CSC system increases compared to single CSC. More importantly, the system CMV is normally increased with the operation frequency due to higher phase voltage, and the excited CMC under resonance point will increase correspondingly. Thus, it is also very important to suppress the inverterside CMV for a paralleled CSC system.
13.4.2 DClink circuit of parallel CSC For parallel CSC with shared DClink structure, since only the total DClink current can be controlled, the DC current balancing is a priority issue that needs to be addressed. Therefore, the equivalent DClink circuit should be studied to get a detailed understanding of the causes of current sharing error. Meanwhile, the CM loop circuit is also needed to be simultaneously considered when dealing with the current balance issue. The subDClink current flowing through the DC choke can be analyzed by considering the voltage stress on the choke, which influences the value of DClink current continuously. Obviously, the voltage stress on each subDC choke is changed with the rectifierside and inverterside switching state. Assume that the rectifierside and inverterside voltages can be replaced by a changeable voltage source, and the DClink circuit shared DClink parallel CSC system can be simplified as shown in Fig. 13.16. Vp and Vn are the rectifierside positive and negative DClink voltages. VipN and VinN (N ¼ 1, 2, $$$) represent the
386 Control of Power Electronic Converters and Systems
FIGURE 13.16 Equivalent DClink circuit of parallel current source converter system with shared DC link.
inverterside positive and negative DClink voltages, which are related to the inverterside switching states. ip and in are the total positive and negative DC current, i2N1 and i2N (N ¼ 1, 2, $$$) represent the positive and negative subDClink currents for CSIN, which can be expressed as Z 8 1 > > i ðVp VipN Þdt ¼ > 2N1 < Ldiff ; N ¼ 1; 2; / (13.9) Z > 1 > > ðVn VinN Þdt : i2N ¼ Ldiff The ideal DC current balance condition is that the positive and negative DC current of each CSI module is equal. Thus, the positive and negative DC current sharing errors are introduced to indicate the current balance performance, which can be expressed as Z 8 1 > > Di ðVipN Vip1 Þdt ¼ i i ¼ > pN 1 2N1 < Ldiff ; N ¼ 2; 3; / (13.10) Z > 1 > > ðVinN Vin1 Þdt : DinN ¼ i2 i2N ¼ Ldiff where DipN and DinN represent the positive and negative DC current sharing error between CSI1 and CSIN. Obviously, the values of current sharing
Highpower current source converters Chapter  13
387
errors can be adjusted with VipN and VinN (N ¼ 1, 2, $$$). Therefore, the switching sequence of each CSI should be designed comprehensively to regulate both positive and negative DC current sharing errors to converge to zero.
13.5 DC current balance and CMV reduction methods The CMV puts a great challenge on transformerless parallel CSCfed drive system and the DC current balance needs to be further considered in a shared DClink structure. Since the increased number of switching states can be generated through a parallel connection, it is more flexible to design a proper PWM sequence to achieve smaller CMV. To achieve DC current balance, the PWM sequence needs to be properly designed. Due to the flexibility of PWM sequence design, the SVMbased methods to solve the CMV and DC current balance issues will be introduced first. Then, the interleaved DDPWM will be investigated for multiple parallel CSC systems due to inherent modularity.
13.5.1 SVMbased methods For parallel CSC system, both interleaved SVM and multilevel SVM can be implemented to achieve multilevel output, and the redundant switching states can be utilized to deal with the CMV issue. Specifically, the interleaved SVM can replace the redundant zero states to achieve a minimized CMV, while multilevel SVM can make use of all redundant switching states to balance DC current and reduce CMV, but results in a heavier computational burden.
13.5.1.1 Interleaved SVM The digital implementation of SVM can be fulfilled by comparing the calculated dwell time and sawtooth timer. The PWM sequence can be flexibly designed to achieve multiple goals such as harmonic optimization and switching time minimization. Different types of SVMbased PWM design such as 3segment D, 4segment, and 4segment D were proposed in Ref. [21], which result in different harmonic and CMV performance. The 3segment SVM is shown in Fig. 13.17; two adjacent active states (In, Inþ1) and one zero state are used to synthesize the current reference. The interleaved SVM can easily be implemented by shifting the timer while sharing the same dwell time. For example, the timer can be shifted by Ts/2 for 2CSC parallel system, where the PWM sequence of each CSC also shifted almost Ts/2 since the dwell time stays constant during adjacent sampling intervals. Therefore, both CSCs enjoy 3segment PWM sequence and multilevel output can be achieved. There are three redundant zero states for single CSC, and the conventional SVM selects the proper zero state to minimize the switching frequency. To reduce the CMV, an average value reduction (AVR)based SVM was proposed in Ref. [14] to minimize the average value of CMV during one sampling
388 Control of Power Electronic Converters and Systems
FIGURE 13.17 Interleaved space vector modulation for parallel current source converter (CSC) application with two units. (A) Pulse width modulation sequence, (B) zero state replacement.
period instead of the switching frequency with redundant zero state replacement. The average value of CMV (CMVave) in each sampling period can be defined as Eq. (13.11), CMVave ¼ jT1 $ CMVact1 þ T2 $ CMVact2 þ T0 $ CMVzero j
(13.11)
where CMVact1, CMVact2, and CMVzero represent the CMV produced by adjacent active and zero states in one sampling period, respectively. T1, T2, and T0 are the corresponding dwell times. Fig. 13.18 shows comparative results of interleaved conventional 3segment SVM and 3segment AVR SVM for 2CSC parallel system under certain case (ma ¼ 0.8, 4 ¼ 0 , f0 ¼ 60 Hz, ft ¼ 1080 Hz). 5level current output can be achieved with interleaved SVM. The equivalent switching frequency is doubled. The output THD of both interleaved conventional 3segment SVM and 3segment AVR SVM is 38.30% since the replaced zero state does not influence the output PWM. However, the CMV can be effectively suppressed from 0.48 to 0.07 p.u by adopting an interleaved AVR SVM.
Highpower current source converters Chapter  13
389
FIGURE 13.18 Output current pulse width modulation (PWM) and commonmode voltage (CMV) waveforms of 2CSC parallel system (ma ¼ 0.8, 4 ¼ 0 , f0 ¼ 60 Hz, ft ¼ 1080 Hz). (A) Interleaved conventional 3segment space vector modulation (SVM), (B) interleaved 3segment average value reduction SVM.
Due to the modularity of interleaved modulation, other AVR SVMbased CMV reduction methods developed for single CSC can easily be extended to parallel CSC system. The thirdorder CMVs of different interleaved AVR SVMs are shown in Fig. 13.19. Compared to 3segment AVR SVM and 4segment AVR SVM, the 4segment AVR SVM D and 3segment AVR SVM D can effectively suppress the CMV in the low modulation index region.
13.5.1.2 Multilevel SVM Multilevel SVM has been well developed in 2CSC parallel systems where 19 current vectors and 81 switching states are available as shown in Fig. 13.20A. These vectors can be divided into four types, named as zero, small, medium, and large vectors based on their length. Different from interleaved SVM, the multilevel SVM strategy is implemented by combining the switching states of the paralleled CSC module together. The PWM sequences of parallel module are synchronized with each other. To implement multilevel SVM modulation, three adjacent vectors are selected to synthesize the reference considering the triangle where the reference is located.
390 Control of Power Electronic Converters and Systems
FIGURE 13.19 Thirdorder commonmode voltage (CMV) with different interleaved AVR SVM. (A) 3segment AVR SVM, (B) 4segment AVR SVM, (C) 4segment AVR SVM D, (D) 3segment AVR SVM D.
FIGURE 13.20 Multilevel space vector modulation with two parallel units. (A) Space vector diagram, (B) redundant switching state replacement.
Based on the equivalent CM loop circuit analysis, it shows that the system equivalent CMV is the average value of CMV caused by each CSC module. Different switching state combinations can cause different kinds of CMV values, which are summarized in Table 13.3. Each switching state is represented by four digits which show the turnedon devices, the first two digits show the onstate switches in CSC1, and the last two represent the onstate switches in CSC2, respectively. For example, one possible switching state [12,16] of medium vector IM1 indicates that S1, S6 in CSC1 and S1, S2 in CSC2 are turned on. The types of CMV values are significantly increased
Highpower current source converters Chapter  13
391
TABLE 13.3 DC current influence and commonmode voltage with different switching states of 2CSC parallel system. Switching states
Commonmode voltage
DC current influence
[14;14], [36;36], [52;52]
Va, Vb, Vc
Dip: х; Din: х
[14;36], [36;52], [52;14]
0.5Vc, 0.5Va, 0.5Vb
If Vab/Vbc/Vca>0, Dip Y; Din [
[16;34], [32;56], [54;12]
0.5Vc, 0.5Va, 0.5Vb
If Vab/Vbc/Vca>0, Dip Y; DinY
IL1
[16;16]
0.5Vc
Dip: х; Din: х
Medium
IM1
[16;12]
0.25Va
Dip: х; if Vbc>0, Din [
Small
IS1
[16;14]
0.25(3Va þ Vb)
Dip: х; if Vab>0, Din Y
[16;36]
0.25(Va þ 3Vb)
Din: х; if Vab>0, Dip Y
[16;52]
0.25Vc
If Vca>0, Dip [; if Vbc>0, Din [
[12;56]
0.25Vc
If Vca>0, Dip [; if Vbc>0, Din Y
Type
Vector
Zero
I0
Large
compared to single CSC due to the abundance of switching states. Therefore, it is more flexible to choose proper switching states to reduce the CMV with improved modulation schemes. After determining the three vectors to form the current reference, the specific switching states need to be further determined since some vectors have redundant switching states. The switching state selection for conventional multilevel SVM is to minimize the switching frequency. To reduce the CMV, multilevel AVR SVM can be achieved to deal with the CMV with these redundant switching states. The CMVave produced in each sampling period with multilevel SVM can be expressed as CMVave ¼ jT1 $ CMV1 þ T2 $ CMV2 þ T3 $ CMV3 j
(13.12)
where CMV1, CMV2, and CMV3 are the CMVs produced by the three adjacent vectors Ix, Iy, and Iz, respectively. The possible values are shown in Table 13.3. The CMVave generated by every possible redundant switching states is compared and the switching states which produce the minimized average CMVave are selected; therefore, the CMV can be suppressed.
392 Control of Power Electronic Converters and Systems
FIGURE 13.21 Output current pulse width modulation (PWM) and commonmode voltage (CMV) waveforms of 2CSC parallel system (ma ¼ 0.8, 4 ¼ 0 , f0 ¼ 60 Hz, ft ¼ 2160 Hz). (A) Conventional 5level space vector modulation (SVM), (B) 5level average value reduction SVM.
To compare the PWM and CMV features of conventional multilevel SVM and multilevel AVR SVM, Fig. 13.21 shows comparative results of conventional 5level SVM and 5level AVR SVM for 2CSC parallel system. Since the switching frequency of conventional 5level SVM for 2CSC parallel system is half of 3level SVM adopted for single CSC, the timer frequency is set as 2160 Hz to guarantee the same switching frequency. As it can be seen, the same 5level current output can be achieved for both conventional 5level SVM and 5level AVR SVM (THD: 38.31%), since replaced redundant switching states of AVR SVM do not influence the output PWM. Different from conventional multilevel SVM, the PWM sequence of multilevel AVR SVM is designed to reduce the CMVave through Eq. (13.12) instead of switching frequency. As a result, the CMV can be effectively suppressed from 0.48 to 0.06 p.u by adopting 5level AVR SVM. It is worth to mention that the switching frequency and computational burden proposed method are increased due to numerous redundant switching state replacement, especially the parallel module number increased.
Highpower current source converters Chapter  13
393
FIGURE 13.22 Thirdorder commonmode voltage (CMV) of 2CSC parallel system. (A) Conventional 5level space vector modulation (SVM), (B) 5level average value reduction SVM.
Fig. 13.22 shows the thirdorder CMV caused by conventional 5level SVM and proposed 5level AVR SVM adopted for 2CSC parallel system under different displacement angles and modulation indexes. Compared to conventional 5level SVM, the proposed 5level AVR SVM can effectively reduce the thirdorder CMV in the whole modulation index range. As a result, the CMV resonance can be suppressed. Besides the CMV reduction methods, the DC current balance for shared DClink structure also needs to be considered. According to current sharing error in Eq. (13.10), the subDClink currents are influenced by the linetoline voltage and switching states, which can be summarized as given in Table 13.3, where symbol “ⅹ” means no influence, “Y” means DC current decrease, and “[” means increase. It shows that the large vectors have no influence on the DC current as the turnon devices of each CSI are the same, which means the inverterside DClink voltage of each CSI is the same. Therefore, the DC currents will stay constant under these switching states. The same conclusion is also applied for some zero vectors with the same switching states for each CSI. The rest switching states can result in different inverterside DClink voltages, and the DC currents can be adjusted by them according to the signs of the linetoline voltages. Based on the above analysis, a general multilevel SVMbased DC current balance strategy is shown in Fig. 13.23, where the desired switching
FIGURE 13.23 DC current balance strategy under 2CSC multilevel converters.
394 Control of Power Electronic Converters and Systems
states can be selected to adjust the DC current by considering the inverterside linetoline voltage and the symbols of Dip and Din. Once the symbols of Dip and Din are detected, the proper switching states can be chosen to make them close to zero based on Table 13.3. According to Table 13.3, the small vectors can adjust both positive and negative currents, which is key to minimize CMV while considering the DC current balance. When selecting the switching state of small vectors, the CMVave produced by all possible switching states will be compared, the proper switching state which produces the smallest CMVave will be selected, and as a result, the DC current balance and CMV reduction can be achieved simultaneously. For example, when the reference located in Sector 1 Area 3, the 3 adjacent vectors IL1, IM1, and IS1 are adopted and the PWM sequence is shown as in Fig. 13.24. The medium vector IM1 can only adjust Din, while the small vector IS1 can adjust both Dip and Din; thus, the redundant switching states can be selected to reduce CMVave.
13.5.2 Carriershifted SPWMbased methods Since the number of redundant switching states has increased significantly due to parallel connection, for example, 729 switching states are available in a 3CSC parallel system. Multilevel SVMbased redundant switching state selection will be very complex to deal with the CMV and DC current balance issue when the module number is increased. Therefore, most of multilevel SVMbased methods are focusing on 2CSC parallel system. On the other hand, the carrierbased SPWM enjoys inherent scalability, modularity, and easytoimplement features for parallel CSC application, which can generate the gating signals by simply comparing the carriers and reference without realtime reference location identification and dwell time calculation. The low computational burden feature makes it very suitable for a modular CSC system, especially for high switching frequency applications where the admitted
FIGURE 13.24 Space vector sequence and DC current influence of 2CSC parallel system in Sector 1 Area 3.
Highpower current source converters Chapter  13
395
processing time is limited. Among different carrierbased SPWMs, the DDPWM enjoys inherent small CMV feature; thus, this section will focus on an interleaved DDPWMbased CMV reduction and DC current balance methods. The principle of interleaved SPWM has been introduced in Ref. [22], where the interleaved DDPWM adopted for 2CSC parallel system with shared DC link is analyzed as an example. Similar to the DDPWM applied in single CSC, the maximum, middle, and minimum values among the threephase reference currents are first obtained. The six switches in each CSC module are identified as Spmax1(2)/Snmax1(2), Spmid1(2)/Snmid1(2), and Spmin1(2)/Snmin1(2) as shown in Fig. 13.25A, and they are used to synthesize the references Imax, Imid, and Imin, respectively. The phase voltages are marked as Vx, Vd, and Vn. C11 and C12 are the upper carrier and lower carrier of CSC1, while C21 and C22 are the upper carriers and lower carrier of CSC2, the carrier shifting angle between the two CSCs is 180 degrees. As a result, the PWM sequence of each CSC can be shifted 180 degrees as shown in Fig. 13.25B. A general interleaved DDPWM for NCSC (N 3) parallel system with 360 degrees/N shifting angle was proposed in Ref. [22], where 2Nþ 1 level current can be guaranteed with better output quality. Fig. 13.26 shows the current PWM and CMV waveforms with interleaved DDPWM for 2CSC parallel system. Similar to the interleaved SVM, the interleaved DDPWM enjoys small CMV while keeping good THD performance (40.64%). The magnitude of thirdorder CMV is 0.12 p.u which is achieved inherently without redundant switching state replacement. Therefore, the computational burden can be reduced, which makes it very suitable for high switching frequency applications and modular design.
FIGURE 13.25 Pulse width modulation (PWM) sequence of interleaved DDPWM. (A) 2CSC parallel system with shared DClink, (B) interleaved DDPWM.
396 Control of Power Electronic Converters and Systems
FIGURE 13.26 Output current pulse width modulation (PWM) and commonmode voltage (CMV) waveforms with interleaved DDPWM of 2CSC parallel system (ma ¼ 0.8, 4 ¼ 0 , f0 ¼ 60 Hz, fc ¼ 1080 Hz).
FIGURE 13.27 Thirdorder commonmode voltage (CMV) of 2CSC parallel system. (A) Interleaved DDPWM, (B) interleaved AVR DDPWM.
The magnitude of thirdorder CMV of interleaved DDPWM under different modulation indexes and displacement angles is shown in Fig. 13.27A, which can verify that DDPWM can effectively reduce the CMV under most conditions (highmodulation index or small displacement angle range). To further reduce the CMV, the redundant zero state replacement adopted for interleaved SVM can also be applied for interleaved DDPWM as shown in Fig. 13. 27B. However, the actual CMV value is small due to low base value in the low modulation index range. Therefore, the interleaved DDPWM is competent for most of drive applications. As introduced in Eq. (13.10), the subDClink currents are influenced by the inverterside linetoline voltage and switching states. The DC current influence under different switching states of multilevel SVM is summarized in Table 13.4. Similar analysis can be conducted with interleaved DDPWM. The switching state for DDPWM is defined as [Spmax, Spmid, Snmid, Snmin], as Spmax, Spmid and Snmid, Snmin are complementary. Therefore, there are four different switching states ([0,1,0,1], [0,1,1,0], [1,0,0,1], and [1,0,1,0]) for one CSC, which are named as S1, S2, S3, and S4, respectively.
Highpower current source converters Chapter  13
397
TABLE 13.4 DC current influence with interleaved DDPWM. S1
S2
S3
S4
S1
Dip: х; Din: х
Dip: х; Din: [
Dip: [; Din: х
Dip: [; Din: [
S2
Dip: х; Din: Y
Dip: х; Din: х
Dip: [; Din: Y
Dip: [; Din: х
S3
Dip: Y; Din: х
Dip: Y; Din: [
Dip: х; Din: х
Dip: х; Din: [
S4
Dip: Y; Din: Y
Dip: Y; Din: х
Dip: х; Din: Y
Dip: х; Din: х
The positive DCbus voltages under the four switching states are Vd, Vd, Vx, and Vx, while the negative DCbus voltages are Vn, Vd, Vn, and Vd in MaxMidMin phase. Their values are equal to phase voltages Va, Vb, Vc in ABC phase. When considering a 2CSC parallel system, there are 16 different switching state combinations. The DC current influence with different combinations is shown as in Table 13.4 by assuming Vx > Vm and Vm > Vn. Otherwise, the DC current influences are opposite. The positive DC difference Dip and negative DC current difference Din are not changed when the switching states of two CSCs are the same. Based on the PWM sequence of interleaved DDPWM, the PWM sequence of each CSC is 5segment symmetrical and they are shifted 180 degrees. Based on the DC current influence summarized in Table 13.4, the excited DVp and DVn values are positive and negative linetoline voltage alternately in each carrier period. That means the resulting Dip and Din will keep going up and down around zero in each carrier period. Meanwhile, both DVp and DVn are repetitive in each 120 degrees interval as shown in Fig. 13.28. It indicates that their integral results are exactly the same in each 120 degrees interval; therefore, the DC current error changed value is fixed in each 120 degrees interval. According to the above analysis, the active DC current balance method can be simply fulfilled by taking use of the repetitive features of the voltage differences and shows the results of voltage difference and resulted current sharing error waveforms by reversing the gating signals of the two CSCs. It indicates that the voltage difference waveforms are reversed too, which can keep the current sharing error equal to zero in every two 120 degrees intervals.
13.5.3 Case study results Experiments are conducted on a 2CSC parallel CSC system (SGCT module number: 5SHZ 0860F0005) with shared DC link. Two CSIs are paralleled with a shared output filter and the parameters are listed in Table 13.5. The shared DClink current is supported by a DC source and the carrier frequency is
398 Control of Power Electronic Converters and Systems
FIGURE 13.28 DC current sharing error of 2CSC parallel system with interleaved DDPWM.
TABLE 13.5 Experimental parameters for a case study. Parameters
Experiment value
Power rating
10 kW/208 V
DClink current
4A
Carrier frequency
1080 Hz
Differentialmode inductor
10 mH
Output filter capacitance
120 mF
Load resistance
5.76 U
Load inductance
5 mH
1080 Hz. Multilevel SVMbased DC current balance and CMV reduction method are verified on a parallel CSC with a shared DC link. Then, different interleaved SPWMs are compared on a 2CSC parallel system with shared DC link to verify the superior performance of DDPWM. The experimental results of parallel CSC with shared DC link by adopting multilevel SVM are shown in Fig. 13.29. The output frequency is 10 Hz and the modulation index is 0.8. At the time of 1s, the modulation method is switched from multilevel SVMbased DC current balance modulation to the
Highpower current source converters Chapter  13
399
FIGURE 13.29 Experimental results of 2CSC parallel system with shared DC link. (A) DClink current and commonmode voltage (CMV) waveforms, CMV FFT results of (B) DC current balance method, (C) DC current balance with CMV reduction method.
DC current balance with CMV reduction method. The positive subDClink currents i1 and i3 are constant around 2 A, which shows good DC current sharing performance. More importantly, the CMV reduced effectively when the modulation strategy is switching to simultaneous DC current balance and CMV reduction method. The FFT results show that the dominant component of CMV is the third order, which reduced from 6.7 V to only 1.2 V while keeping the DC current balanced. Thus, the experimental results verify that the proposed method can achieve DC current balance and CMV reduction simultaneously. To further verify the performance with different carrierbased SPWMs, three interleaved SPWMs (BTSPWM, SSDPWM, and DDPWM) are implemented on a 2CSC parallel system with shared DC link as shown in Fig. 13.30. The output frequency is 60Hz and the interleaving angle is 180 degrees for all of the three modulations. The excited CMVs for interleaved discontinuous BTSPWM and SSDPWM are similar in terms of magnitude as well as FFT spectrum. The thirdorder component of CMV excited by interleaved discontinuous BTSPWM is 11.20 and 10.81 V when modulation index is 0.6 and 0.8, respectively, where they are 11.12 and 10.86 V for interleaved SSDPWM. In terms of DC current balance, all of the three SPWMs can achieve good DC current sharing. Moreover, compared to discontinuous BTSPWM and SSDPWM, the CMV magnitude and thirdorder component can be effectively reduced with DDPWM. The thirdorder component is reduced to 0.58 V when the modulation index is 0.6, while it is reduced to 2.14 V when the modulation index is 0.8. It is consistent with theoretical analysis in Section 13.3 and the effectiveness of interleaved DDPWM can be verified.
400 Control of Power Electronic Converters and Systems
FIGURE 13.30 Experimental results of different carrierbased sinusoidal pulse width modulation for 2CSC parallel system with shared DC link. (A) BTSPWM, (B) SSDPWM, (C) DDPWM.
13.6 Conclusions CSCs have been widely adopted in highpower applications, such as HVDC and industrial drives, for a long history. Parallel CSC configuration can further improve the system power rating, reliability, and also output quality.
Highpower current source converters Chapter  13
401
Compared to bulky isolation transformers, transformerless CSC topology by adopting a CM choke can effectively reduce the cost. This chapter mainly studies the transformerless parallel CSC structures and improved modulation and control methods to deal with their potential challenges, such as CMV, CC, and DC current balance. An SVMbased simultaneous DC current balance and CMV reduction method is proposed for parallel CSC system with shared DC link, which can reduce the CMV effectively while keeping a balanced DC current. To achieve a true modular design, an interleaved DDPWM is presented for use with multiple CSC parallel system, which enjoys small CMV as well as having scalability. The analysis and effectiveness of the proposed methods are verified through experiment results. The parallel configurations associated with the proposed methods have the potential benefits of improving the system power rating and power quality, scale down the CM choke and the damping resistor as well as output filter, which are key indicators to improve performance and reduce the cost and size of the CSCfed drive system.
References [1] [2] [3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
B. Wu, HighPower Converters and AC Drives, WileyIEEE Press, 2006, pp. 189e218. N.M. Kirby, L. Xu, M. Luckett, W. Siepmann, HVDC transmission for large offshore wind farms, Power Eng. J. 16 (3) (June 2002) 135e141. S. Nishikata, F. Tatsuta, A new interconnecting method for wind turbine/generators in a wind farm and basic performances of the integrated system, IEEE Trans. Ind. Electron. 57 (2) (February 2010) 468e475. B. Wu, J. Pontt, J. Rodriguez, S. Bernet, S. Kouro, Currentsource converter and cycloconverter topologies for industrial mediumvoltage drives, IEEE Trans. Ind. Electron. 55 (7) (2008) 2786e2797. J.M. Erdman, R.J. Kerkman, D.W. Schlegel, G.L. Skibinski, Effect of PWM inverters on AC motor bearing currents and shaft voltages, IEEE Trans. Ind. Appl. 32 (2) (MarchApril 1996) 250e259. R.E. TorresOlguin, A. Garces, M. Molinas, T. Undeland, Integration of offshore wind farm using a hybrid HVDC transmission composed by the PWM currentsource converter and linecommutated converter, IEEE Trans. Energy Convers. 28 (1) (2013) 125e134. M. Popat, B. Wu, F. Liu, N. Zargari, Coordinated control of cascaded currentsource converter based offshore wind farm, IEEE Trans. Sustain. Energy 3 (3) (July 2012) 557e565. L. Ding, Y.W. Li, Mixed SeriesParallel Connected Current Source Converters with Interleaved SPWM, IEEE Energy Conversion Congress and Exposition (ECCE), Baltimore, MD, USA, 2019, pp. 640e645, 2019. B. Sahan, A.N. Vergara, N. Henze, A. Engler, P. Zacharias, A single stage PV module integrated converter based on a lowpower currentsource inverter, IEEE Trans. Ind. Electron. 55 (7) (July 2008) 2602e2609. L. Tang, G. Su, Boost Mode Test of a CurrentSourceInverterFed Permanent Magnet Synchronous Motor Drive for Automotive Applications, IEEE 12th Workshop on Control and Modeling for Power Electronics (COMPEL), Boulder, CO, 2010, pp. 1e8, 2010.
402 Control of Power Electronic Converters and Systems [11] F. Xu, B. Guo, L.M. Tolbert, F. Wang, B.J. Blalock, An allSiC threephase buck rectifier for highefficiency data center power supplies, IEEE Trans. Ind. Appl. 49 (6) (NovemberDecember 2013) 2662e2673. [12] T. Friedli, S.D. Round, D. Hassler, J.W. Kolar, Design and performance of a 200kHz allSiC JFET current DClink backtoback converter, IEEE Trans. Ind. Appl. 45 (5) (Septemberoctober 2009) 1868e1878. [13] M. Guacci, M. Tatic, D. Bortis, J.W. Kolar, Y. Kinoshita, H. Ishida, Novel ThreePhase TwoThirdModulated BuckBoost Current Source Inverter System Employing DualGate Monolithic Bidirectional GaN eFETs, IEEE 10th International Symposium on Power Electronics for Distributed Generation Systems (PEDG), Xi’an, China, 2019, pp. 674e683, 2019. [14] L. Ding, Z. Quan, Y.W. Li, Commonmode voltage reduction for parallel CSCfed motor drives with multilevel modulation, IEEE Trans. Power Electron. 33 (8) (August 2018) 6555e6566. [15] L. Ding, Y.W. Li, Simultaneous DC current balance and commonmode voltage control with multilevel current source inverters, IEEE Trans. Power Electron. 33 (11) (November 2018) 9188e9197. [16] X. Wang, B.T. Ooi, Unity PF currentsource rectifier based on dynamic trilogic PWM, IEEE Trans. Power Electron. 8 (3) (July 1993) 288e294. [17] Z. Bai, X. Ruan, Z. Zhang, A generic sixstep direct PWM (SSDPWM) scheme for current source converter, IEEE Trans. Power Electron. 25 (3) (March 2010) 659e666. [18] N. Choi, K. Lee, B. Han, A Novel Carrier Based PWM for Current Source Converter, IEEE 7th International Power Electronics and Motion Control Conference ECCE Asia, Harbin, China, 2012. June 25, pp. 1945e1950, 2012. [19] J. Espinoza, G. Jos, J. Guzmn, L. Morn, R. Burgos, Selective harmonic elimination and current/voltage control in current/voltage source topologies: a unified approach, IEEE Trans. Ind. Electron. 48 (1) (2001) 71e81. [20] Y. Li, L. Ding, Y. Li, Isomorphic relationships between voltagesource and currentsource converters, IEEE Trans. Power Electron. 34 (8) (August 2019) 7131e7135. [21] L. Ding, Y. Lian, Y.W. Li, Multilevel current source converters for high power medium voltage applications, CES Trans. Electric. Machine. Syst. 1 (3) (September 2017) 306e314. [22] L. Ding, Y. Li, Simultaneous DC current balance and CMV reduction for parallel CSC system with interleaved carrierbased SPWM, IEEE Trans. Ind. Electron. 67 (10) (October 2020) 8495e8505.
Chapter 14
Parallel operation of power converters and their filters Ghanshyamsinh Gohil University of Texas at Dallas, Richardson, TX, United States
14.1 Introduction Threephase pulse width modulated (PWM) voltage source converter (VSC) is widely used in many DC/AC and AC/DC power conversion applications, including variable speed motor drives, renewable energy and distributed generation systems, uninterrupted power supply, solidstate transformer, active power filters, fast battery charger for the electric vehicle, etc. The parallel operation of VSC is desired in many of these applications to achieve system cost reduction through modularity. A parallel connection enables the use of the standard VSC module, high availability, and low cost of maintenance and small inventory. Moreover, the flexibility offered by the parallel connection of the VSCs is highly desirable in many applications that require capacity expansion over time. The parallel connection of VSC is also ubiquitous in critical installations, where redundancy is required to ensure minimum downtime. In highcurrent applications, processing entire power using a single VSC, may not be preferred because of the economic and technical challenges associated with the highcurrent power semiconductor modules and passive components. Therefore, VSCs are often connected in parallel to divide the total current among parallel VSCs. The power conversion system (PCS) with parallel VSCs for AC/DC and AC/ AC conversion applications is shown in Fig. 14.1. The parallel backtoback connected VSCs, shown in Fig. 14.1B, have separate DC buses. However, a common DCbus connection is also possible. In a PCS with parallel VSCs, proper current sharing between the VSCs is highly desirable. However, the filter impedance and semiconductor device parameter mismatch, application of different voltage vectors, and deadtime effect make current sharing very challenging. As a result, circulating current flows between the parallel VCSs due to the existence of the closed path, leading to the underutilization of the system
Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000019 Copyright © 2021 Elsevier Ltd. All rights reserved.
403
404 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 14.1 Power conversion system with parallel voltage source converters. (A) Parallel voltage source converters with a common DC and AC buses in AC/DC conversion applications, (B) AC/AC conversion system using parallel backtoback connected voltage source converters.
capacity. Moreover, the components (semiconductors and filter) present in the circulating current path experience additional losses due to the flow of the circulating current, which in turn increases stress and may reduce the lifetime of these components. The circulating current may have lowfrequency and highfrequency components. The lowfrequency circulating current can be reduced by using an appropriate control scheme, whereas modifications in the modulation schemes, carrier signal synchronization, and passive filters are employed for the highfrequency circulating current reduction. The modeling of the circulating current is discussed in Section 2. Various control schemes to achieve desired load sharing and to control the circulating current are discussed in Section 3. In the conventional PCS with parallel VSCs, control efforts are made to apply the same voltage vector to all the parallel VSCs to ensure minimum
Parallel operation of power converters and their filters Chapter  14
405
circulating current. However, the harmonic quality can be improved at the expense of large highfrequency circulating current by applying different vectors to the parallel VSCs. This is typically achieved by interleaving the carrier signals of the parallel VSCs. The concept of harmonic quality improvement of the parallelconnected PWM VSCs by interleaving the carrier signals has been first proposed in Ref. [1]. Various interleaving techniques and their impact on the harmonic performance and circulating current will be presented in Section 4. When VSCs are connected in parallel, the circulating current flows between the VSCs due to the control asymmetry and the impedance mismatch. When the carriers are interleaved, the switched output voltages of the interleaved parallel legs are phase shifted. As a result, the instantaneous voltage difference exists between parallel VSCs, which further increases the highfrequency circulating current. Therefore, the circulating current should be suppressed to realize the full potential of the interleaved carriers in parallelconnected VSCs. The highfrequency circulating current introduced by the carrier interleaving can be eliminated by providing galvanic isolation [2] or can be reduced to a reasonable limit by introducing an impedance in the circulating current path. Various inductive components that are used to offer large impedance to the highfrequency circulating current, including coupled inductor (CI) [1,3], commonmode (CM) inductor [4], and integrated inductor, are discussed in details in Section 5. The main aim of this chapter is to provide an understanding of the parallel operation of the VSCs. The challenges associated with the load sharing and solutions for the circulating current reduction are described in detail. Harmonic performance improvement using interleaved operation and techniques for the highfrequency circulating current suppression for parallel interleaved operation are illustrated.
14.2 Circulating current modeling Because of the filter impedance and semiconductor device parameter mismatch, application of different voltage vectors, and deadtime effect, the VSC current may deviate from the desired set point. As a result, the circulating current flows between the VSCs, leading to additional losses in the semiconductor and magnetic components present in the circulating current path. The circulating current for the parallel VSCs with a common DCbus arrangement (c.f. Fig. 14.1A) and separate DCbus arrangement (c.f. Fig. 14.1B) is analyzed in this section.
14.2.1 Parallel converters with a common DC bus For the PCS with parallel VSCs sharing a common DC link shown in Fig. 14.1A, the dynamic behavior of the kth VSC current (k ˛ {1,2,.,N})
406 Control of Power Electronic Converters and Systems
with singlephase inductors having inductance Lk and equivalent series resistance rk can be expressed as 2 3 2 3 2 3 2 32 3 2 3 iak vak o Lk 0 0 rk 0 0 iak vao 6 7 6 7 6 7 6 76 7 6 7 6 vbk o 7 ¼ 6 0 Lk 0 7 d 6 ibk 7 þ 6 0 rk 0 76 ibk 7 þ 6 vbo 7 (14.1) 4 5 4 5 dt 4 5 4 54 5 4 5 v ck o i ck ick 0 0 Lk 0 0 rk vco where vako ; vbko , and vcko represents pole voltages (with respect to the arbitary reference point o) of phase a; b, and c, respectively. vao ; vbo , and vco are the load voltages (with the respect to the reference point o) of phase a; b, and c, respectively. The dynamic behavior of the phase x (where x ˛ {a,b,c}) currents of all the parallel VSCs can be represented as 1 1 1 d CV P ¼ Cr I þ CL I þ V N N N dt
(14.2)
where the V P is the pole voltage vector v xN o T
V P ¼ ½ vx1 o vx2 o/
(14.3)
I is a VSC current vector I ¼ ½ i x1
ix2 /ixN T
(14.4)
and V ¼ ½ Vxo
Vxo /
Vxo T
(14.5)
and 2
ðN 1Þ 6 6 1 6 C¼6 6 « 4 1 2
6 6 6 Cr ¼ 6 6 4
1 1
« 1/
« ðN 1Þ
3 7 7 7 7 7 5
ðN 1Þr1
r2 /
rN
r1 «
ðN 1Þr2 / «
rN «
r1
r2 /
ðN 1ÞrN
ðN 1ÞL1
L2 /
LN
L1
ðN 1ÞL2 /
LN
« L1
« L2 /
« ðN 1ÞLN
6 6 6 Cr ¼ 6 6 4 2
1/ ðN 1Þ/
(14.6)
3 7 7 7 7 7 5
(14.7)
3 7 7 7 7 7 5
(14.8)
Parallel operation of power converters and their filters Chapter  14
407
The load current of a particular phase is the sum of all VSC currents of that phase and it is given as ix ¼
N X
ixk ; where 1 k N
(14.9)
k¼1
where ixk is the current through phase x leg of kth VSC. For the parallelconnected VSCs, the VSC current ixk can be split into two components: 1. The component contributing to the load current ixk ;l 2. The circulating current ixk ;c Therefore, the VSC current can be represented as ixk ¼ ixk ;l þ ixk ;c
(14.10)
where ixk ;l is the load current component of the VSC current ixk and ixk ;c is the circulating current component of the VSC current. The circulating current components ixk ;c do not contribute to the resultant line current. Therefore, (14.9) can be rewritten as ix ¼
N X
ixk ;l; where 1 k N
(14.11)
ixk ;c ¼ 0; where 1 k N
(14.12)
k¼1
and N X k¼1
14.2.1.1 Impact of the modulator mismatch In a modular PCS with parallel VSCs, each VSC uses a separate control unit with its own modulator. The mismatch in the reference signals as well as the asynchronous carrier signals leads to the circulating current. A simplified analysis is performed to evaluate the impact of the modular mismatch. A symmetrical filter impedance is assumed to decouple the impact of the impedance mismatch on the circulating current. Under this assumption L1 ¼ L2 ¼ LN ¼ Lf
(14.13)
r1 ¼ r2 ¼ rN ¼ rf
(14.14)
and
408 Control of Power Electronic Converters and Systems
Using Eqs. (14.2), (14.13), and (14.14), the dynamics of the circulating current is given as 1 d CV P ¼ Rf I c þ Lf I c N dt
(14.15)
where I c ¼ ½ ix1 ;c ix2 ;c/ ixN ;c 3 2 Lf 0/ 0 7 6 60 L/ 07 f 7 6 Lf ¼ 6 7 6 « « « 7 5 4 0 0/ Lf
(14.16)
(14.17)
and 2 r 6 f 60 6 Rf ¼ 6 6« 4 0
0/ rf / « 0/
3 0 7 07 7 7 «7 5 rf
(14.18)
Since the internal resistance of filter inductor is very small compared to the inductive reactance, resistance of the filter can be neglected. By neglecting the resistance of the VSC current part, the circulating current for the PCS with two parallel VSCs can be expressed as Z 1 ix1 ;cðtÞ ¼ ix2 ;cðtÞ ¼ ðvx1 o vx2 oÞdt (14.19) 2Lf From Eq. (14.21), it clear that any difference in the pole voltages would lead to the flow of the circulating current. Several factors, including mismatch in the reference voltages, asynchronous carriers, and semiconductor devices, contribute to the pole voltage mismatch. The circulating current of all three phases can also be represented in terms of the CM current, and the CM current of kth VSC can be represented as Z N Z ia þ ibk þ ick N 1 1 X icmk ðtÞ ¼ k vcmk dt vcmj dt ¼ (14.20) NLf NLf j¼1 3 jsk
Parallel operation of power converters and their filters Chapter  14
409
For two parallel VSCs, the CM circulating current can be represented as Z 1 ðvcm1 vcm2 Þdt icm1 ðtÞ ¼ icm2 ðtÞ ¼ (14.21) 2Lf
14.2.1.2 Impact of the impedance mismatch The pole voltages of parallel VSCs are assumed to be equal to analyze the impact of the impedance mismatch on the circulating current. Under this assumption, the voltage drop across the filter impedance can be represented as rk ixk þ Lk
dixk ¼ vxk o vxo dt
(14.22)
For the two parallel VSCs with equal pole voltages, the voltage drop across the filter impedance is given as r 1 i x1 þ L 1
dix1 dix ¼ r 2 i x2 þ L 2 2 dt dt
(14.23)
Only considering the fundamental frequency component, the VSC currents can be represented using phasors (represented by bold symbols) as ðr1 þ juL1 Þix1 ¼ ðr1 þ juL2 Þix2
(14.24)
and the phasor representation of the circulating current is ix1 ;c ¼ ix2 ;c ¼
ðr2 r1 Þ þ juðL2 L1 Þ ix2 2ðr1 þ juL1 Þ
(14.25)
Therefore, it is evident from Eq. (14.25) that the filter impedance mismatch would lead to the circulating current flow even when the pole voltages are perfectly matched.
14.2.2 Parallel converters with separate DC bus For the PCS with separate DC bus as shown in Fig. 14.1B, the CM current flows between the parallelconnected backtoback VSCs. The circulating current behavior of the PCS with two backtoback connected parallel VSCs with separate DC links is analyzed. The analysis can be simplified by assuming symmetrical impedances (Lg1 ¼ Lg2 ¼ Lg and Ll1 ¼ Ll2 ¼ Ll ). The CM circulating current can be expressed as Z 1 icm1 ðtÞ ¼ vcml1 vcmg1 vcml2 þ vcmg2 dt (14.26) 2 Ll þ Lg where icm1 is the CM circulating current of first backtoback connected VSCs and icm1 ¼ icm2 . vcmlk and vcmgk are the CM voltages of the frontend and backend converter of kth VSC.
410 Control of Power Electronic Converters and Systems
14.3 Circulating current control The circulating current can be avoided by providing galvanic isolation between the parallel VSCs. The galvanic isolation is often achieved using a bulky line frequency transformer, which adds to the cost and increases the size of the PCS. On the other hand, many gridconnected applications use a transformer between the converter system and a grid for voltage matching. Also, in some applications, the grid codes demand galvanic isolation. In such applications requiring parallel VSCs to meet high current requirements, using a transformer with multiple isolated primary windings [2] is a good solution as it avoids the use of any additional components and control efforts to avoid the circulating current. The circulating current reduction and proper load sharing between parallel VSCs can be achieved by incorporating a suitable control strategy. One such control strategy is to use communication lines between the parallel VSCs for exchanging information and reference commands to achieve load sharing. Several implementation using communication among the module have been reported, including centralized and distributed control [5e9]. The dependence on the communication channel is a major drawback of these schemes, as the modularity is compromised. Moreover, redundant communication lines are required in critical installations, which increase the cost and complexity of the system. The dependency on communication can be avoided by using droop control techniques [10,11]. The droop control scheme is motivated by the droop control scheme of the synchronous generators in the conventional power grid. The active and reactive power sharing between the parallel VSCs is achieved by adjusting frequency and voltage amplitude set points of each VSC output voltage, defined by their droop equations. The selfregulating action of the droop control makes it communication free. However, a large deviation in the voltage and frequency is inevitable if accurate load sharing is required. Therefore, droop control is often complimented by having additional control information using a slow and noncritical communication channel between the VSCs. As discussed in Section 2, the circulating current flows between the parallel VSCs due to the difference in the pole voltage. Since the circulating current between the parallel VSCs can also be represented as a CM circulating current, it can be reduced by modifying the CM voltage added to the reference signal used for the carrier comparisonbased modulator. For the space vector implementation, this can be achieved by adjusting the duty cycle of the zero voltage vector. In this section, control schemes for the circulating current reduction are discussed.
Parallel operation of power converters and their filters Chapter  14
411
FIGURE 14.2 Parallel voltage source converters with central controller that implements outer voltage/power loop and provides current references for each individual converter inner current loop realization.
14.3.1 Current sharing schemes In its simplest form, the current sharing scheme can be realized by having a central controller that implements outer voltage or power loop and provides current references for each individual VSC [10], as shown in Fig. 14.2. Each of the VSCs has its own inner current control loop, which processes the current error, obtained by subtracting the measured VSC current from the reference obtained from the central controller [5]. In this scheme, VSC current and load current measurements along with the central controller are required, which compromises the modularity and faulttolerant operation. The central controller and total load current measurement can be avoided by employing circular chain control [12], where VSC controllers are connected in the circular chain, as shown in Fig. 14.3. Each VSC has its inner current loop and outer voltage or power loop. The inner current loop tracks the inductor current of the previous module in the circular chain, leading to an equal current sharing. A mastereslave control technique is shown in Fig. 14.4, where the master VSC operates in the voltagecontrolled mode and provides current references for the currentcontrolled slave VSCs [7]. In the mastereslave control scheme with the dedicated master VSC, the failure of the dedicated master VSC compromises the availability of the complete PCS. The PCS can be made resilient by incorporating the status line to decide the master VSC [13]. In this scheme, one of the slave VSCs takes over if the master VSC fails. The resiliency can also be improved by adopting auto mastereslave control, where the VSC with the highest load current is automatically assigned as a master VSC [14].
412 Control of Power Electronic Converters and Systems
FIGURE 14.3 Circular chain control scheme for parallel voltage source converters.
14.3.2 Droop control scheme All the current sharing schemes discussed so far require critical communication links between the VSCs and a failure of the communication links compromises the reliability of the PCS. The critical communication link can be avoided by employing a droop control scheme [10]. The droop control of the parallel VSCs is similar to frequency/voltage droop control of the synchronous generator in the conventional power grid. For the VSC connected to the common AC bus using a filter, the active and reactive power of the inverter are EV V2 EV P¼ cosf sin f sin q (14.27) cosq þ Z Z Z EV V2 EV cosf sin f cos q (14.28) Q¼ sin q þ Z Z Z where E is the amplitude of the fundamental frequency component of the VSC voltage, V is the amplitude of the common ACbus voltage, Z is the filter
Parallel operation of power converters and their filters Chapter  14
413
FIGURE 14.4 Mastereslave control scheme for parallel voltage source converters.
impedance, q is the phase angle of the filter impedance, and f is the phase angle. For the VSC with an inductive filter (X[R), the active and reactive power expressions can be approximated as EV EV sin fz f X X
(14.29)
EV V 2 VðE VÞ cos f z X X X
(14.30)
P¼ Q¼
where X is the inductive reactance of the filter. From Eqs. (14.29) and (14.30), it is evident that the active power is proportional to the phase angle f and the reactive power is proportional to the voltage amplitude difference E V. Therefore, the active and reactive power of the VSC are regulated by setting appropriate frequency (phase angle) and amplitude of the fundamental frequency component of the VSC output voltage, respectively [11]. Inspired by the synchronous generator control in the conventional power grid, the frequency and amplitude set points of the fundamental frequency component
414 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 14.5 Droop control scheme for the power sharing among parallel voltage source converters. (A) Control block diagram for reference voltage derivation. (B) Droop control.
of the VSC voltage can be derived using the P u and Q V droop schemes, as shown in Fig. 14.5, where the control laws can be expressed as u ¼ un kp ðP Pn Þ
(14.31)
E ¼ En kq ðQ Qn Þ
(14.32)
where kp and kq are the frequency and voltage amplitude droop coefficients, respectively. un is the frequency at no load and En is the voltage amplitude at no load. As it is evident from Eqs. (14.35) and (14.36) that the droop control law offers negative feedback, and good power sharing between parallel VSCs can be achieved by selecting a sufficiently high value of droop coefficients. However, having a high value of the droop coefficients leads to the larger variation in the frequency and voltages. The decoupling between the active and reactive power is only valid for the inductively coupled AC sources, and therefore the droop control laws given by Eqs. (14.35) and (14.36) work satisfactorily with good power sharing only in case of the VSC having inductive filters. However, the closedloop output impedance of the VSC depends on the control strategy. Moreover, the line impedance could be resistive in the lowvoltage distribution system, leading to inaccurate power sharing [15]. In such cases, the active and reactive power can be expressed as P¼
EV V 2 VðE VÞ cos f z R R R
(14.33)
Parallel operation of power converters and their filters Chapter  14
Q¼
EV EV sin fz f R R
415
(14.34)
The droop control laws can be adapted as u ¼ un þ kp ðQ Qn Þ
(14.35)
E ¼ En kq ðP Pn Þ
(14.36)
The power sharing accuracy can also be improved by adjusting the output impedance of the VSC by employing virtual output impedance control [16]. The desired virtual output impedance Zo can be inserted by modifying the voltage reference as vvsc ¼ E sinðutÞ io Zo ðSÞ
(14.37)
where io is the output current. Moreover, it is also possible to design virtual impedance that exhibits inductive behavior at the fundamental frequency and a resistive behavior at the harmonic frequencies [17]. As a result, good power sharing in the case of linear and nonlinear loads can be obtained. Proper load sharing can be achieved by combining both the droop control laws and virtual output impedance control [18], as shown in Fig. 14.6. The droop control laws are used to derive voltage references, whereas the virtual output impedance control is used to achieve a symmetrical impedance. The circulating current controller is also incorporated, which modifies the reference current generated using the outer voltage controller. The VSC currents (iabc1 and iabc2 ) are transformed into the rotating reference frame, and the circulating current controller is implemented in the rotating reference frame, with the control objective to drive the difference between the VSC currents to zero. Although each of the VSCs has a circulating current controller, it is evident from Eq. (14.21) that controlling the pole voltage of only one VSC can effectively reduce the circulating current.
14.3.3 Zero vector dwell time control For two parallel VSCs, the circulating CM current behavior can be derived using Eq. (14.2) as vcm1 vcm2 ¼ ðr1 þ r2 Þicm þ ðL1 þ L2 Þ
dicm dt
(14.38)
where icm is the CM circulating current and it is expressed as icm ¼ icm1 ¼ icm2
(14.39)
416 Control of Power Electronic Converters and Systems
FIGURE 14.6 Load and circulating current control of two parallel converters using a combination of droop control laws and virtual output impedance control.
From Eq. (14.38), it is evident that the lowfrequency CM circulating current depends on the difference of the average value of the CM voltages. Therefore, the lowfrequency CM circulating current can be reduced through the proper control of the average CM voltage over a switching period. Using phaseleg averaging technique, the average of the pole voltage over a switching period vxk o can be defined as 1 c vxk o ¼ dxk Vdc ¼ dxk Vdc (14.40) 2 where dxk is the duty cycle of the x phase leg of kth VSC. Note that the pole voltage is measured with respect to the DCbus midpoint o. The duty cycle is defined as d xk ¼
Ton v 1 ¼ x þ Ts Vdc 2
(14.41)
where Ton is the on time of the upper switch in the VSC leg, Ts is the switching period, and v x is the reference used for the carrier comparison. The reference
Parallel operation of power converters and their filters Chapter  14
417
voltage waveform is often derived by adding common voltage vz in the reference sinusoidal voltage waveform, i.e., the reference waveform of the phase x of kth VSC leg is v xk ¼ vxk þ vzk
(14.42)
Using Eqs. (14.40) and (14.42), the average of the CM voltage over a switching period can be obtained as vcmk ¼ vzk
(14.43)
Therefore, the CM circulating current can be reduced by controlling vzk . The average value of the CM voltage can also be controlled by selecting the appropriate value of the zero vector dwell times. The twolevel (2L) VSC has eight voltage vectors defined by the combination of the switch states. ! ! These states generate six active vectors ( V1 V6 ) and two zero vectors ! ! ( V0 , V7 ), as shown in Fig. 14.7. The threephase reference signals can be ! represented by a complex reference vector Vs+ . Based on the magnitude ! ! ( Vs+ ) and angle (js ) of the sampled Vs+ , two adjacent active voltage vectors and zero vectors are commonly applied to synthesize the reference vector. The respective dwell time of the active vectors is chosen to maintain the voltesec balance.
FIGURE 14.7 Basic space vector sectors and states in complex ab plane, used for the modulation of the voltage source converter.
418 Control of Power Electronic Converters and Systems
! ! ! ! Let T1 , T2 , and Tz be the dwell time of V1 , V2 , and V0 / V7 , respectively, and they are given by ! 2 Vs+ Ts sinð60+ js Þ (14.44) T1 ¼ pﬃﬃﬃ 3 Vdc ! 2 Vs+ T2 ¼ pﬃﬃﬃ Ts sinðjs Þ 3 Vdc
(14.45)
(14.46) Tz ¼ Ts T1 T2 ! ! The dwell time of the zero voltage vectors V0 and V7 is given by Kz Tz and ð1 Kz ÞTz , respectively, where 0GKz G1. Different modulation possibilities exist with variation in the parameter Kz. For example, Kz ¼ 0:5 results in the conventional space vector modulation (SVM). By changing the value of Kz between zero and one with a frequency three times higher than the frequency of the reference signal, several discontinuous pulse width modulation (DPWM) schemes can be realized. If the value of Kz is changed from zero to one in the middle of sector 1 (c.f. 7), the reference signals for 60 degrees clamped DPWM (DPWM1) are generated. Since both the zero voltage vectors do not contribute to any voltesec toward the reference vector synthesizing, they offer an additional degree of freedom ! that can be used to control the average CM voltage. The application of V0 ! and V7 results into the CM voltage of V2dc and V2dc , respectively. Therefore, by adjusting the value of the Kz , the desired average CM voltage can be achieved to reduce the CM circulating current [19], as shown in Fig. 14.8. The circulating current controller is implemented for only one of the VSCs in two parallel VSCbased PCS. All threephase currents are measured, and the CM current is obtained. A proportional integrator (PI) is used to regulate the CM voltage to zero. The PI controller output is Kz , which is then fed to the modulator to determine the dwell time of the voltage vectors ! ! V0 and V7 .
FIGURE 14.8 Circulating current controller implementation for two parallel voltage source converters. Circulating current controller is only implemented for one of the VSCs, as there is only one CM circulating current in two parallel converter system.
Parallel operation of power converters and their filters Chapter  14
419
14.4 Harmonic performance improvement through interleaved operation The harmonic distortion and CM voltage of the PCS with parallel VSCs can be significantly reduced by interleaving operation. Besides, interleaving operation also reduces the DClink capacitor ripple [20]. As a result, the size of the harmonic filter components and DClink capacitor can be reduced.
14.4.1 Modulation of parallel interleaved converters Conventionally the interleaved operation is realized using the phaseshifted carrier signals. Based on the phaseshift angle, the interleaving operation can be classified into two categories: 1. Symmetrical carrier interleaving 2. Asymmetrical carrier interleaving Symmetrical carrier interleaving is commonly used, where the carrier interleaving angle g is set to g¼
360+ N
(14.47)
where N is the number of parallel VSCs. The impact of the asymmetrical interleaving angle on the harmonic performance has been investigated in Refs. [21,22], where it is demonstrated that the asymmetrical carrier interleaving could lead to improved harmonic performance under some operating conditions. Two parallel VSCs are shown in Fig. 14.9A. Carrier signals of both the VSCs are phase shifted by 180 degrees, as shown in Fig. 14.9B. As a result, the switched output voltages of the parallel VSC legs (referred to as pole voltages) are also phase shifted, as shown in Fig. 14.10A and Fig. 14.10B. The resultant voltage is the average of the individual switched output voltage of the parallel legs of that phase and demonstrates threelevel voltage waveforms, as shown in Fig. 14.10C. The linetoline voltage across the phase a and phase b is also shown in Fig. 14.10E. It demonstrates a fivelevel voltage waveform. As a result, compared to the 2LVSC, superior harmonic performance can be achieved. The harmonic performance of PCS with parallel interleaved VSCs is strongly influenced by the modulation scheme, interleaving angle, and the number of parallel VSCs.
14.4.2 Symmetrically interleaved converters In addition to the desired fundamental frequency component, the pole voltages have undesirable harmonic frequency components due to the modulation.
420 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 14.9 Parallel interleaved voltage source converters. (A) Power conversion system with two parallel interleaved voltage source converters (VSC1 and VSC2), (B) Interleaving implementation using phaseshifted carrier signals. Pole voltages of each voltage source converter and their average voltage for phase a are shown.
(A)
(B)
(C)
(D)
(E) FIGURE 14.10 Simulated voltage waveforms for two parallel interleaved voltage source converters. The carrier signals of the two voltage source converters are phase shifted by 180 degrees. (A) Switched output voltage of leg a1 , (B) Switched output voltage of leg a2 , (C) Resultant voltage of phase a, (D) Resultant voltage of phase b, (E) Linetoline voltage Vab . The voltage waveforms are normalized with respect to Vdc =2.
Parallel operation of power converters and their filters Chapter  14
(A)
(B)
(C)
(D)
421
FIGURE 14.11 Representative modulation waveforms for modulation index M ¼ 1. (A) Centeraligned space vector modulation (SVM), (B) 60 degrees clamp discontinuous PWM (DPWM1), (C) 30 degrees lagging clamp discontinuous PWM (DPWM2), and (D) 30 degrees clamp discontinuous PWM (DPWM3).
Every PWM scheme has a unique harmonic spectrum, which dictates the load current quality and harmonic filter sizing. The harmonic performance of the following four conventional PWM schemes is evaluated: l l l l
Centeraligned SVM 60 degrees clamp discontinuous PWM (DPWM1) 30 degrees lagging clamp discontinuous PWM (DPWM2) 30 degrees clamp discontinuous PWM (DPWM3)
The representative modulation waveform for each of these PWM schemes is shown in Fig. 14.11. The harmonic performances of these PWM schemes are compared by evaluating the normalized weighted total harmonic distortion (NWTHD), which is defined as sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ N P M ðvh =hÞ2 NWTHD ¼
h¼2
vf
(14.48)
where vf is the fundamental component and vh is the magnitude of the hth harmonic component.
422 Control of Power Electronic Converters and Systems
The harmonic components can be represented as the summation series of sinusoids, characterized by the carrier index variable m and the baseband index variable n [23] and it is given as
f ðtÞ ¼ þ
N A00 X þ ½A0n cosðn½u0 t þ q0 Þ þ B0n sinðn½u0 t þ q0 Þ 2 n¼1 N X ½Am0 cosðm½uc t þ qc Þ þ Bm0 sinðm½uc t þ qc Þ
(14.49)
m¼1
þ
N X
N X
½Amn cosðm½uc t þ qc þ n½u0 t þ q0 Þ
m¼1 n¼Nns0
þ Bmn sinðm½uc t þ qc þ n½u0 t þ q0 Þ The hth harmonic component is defined in terms of m and n, and it is given as uc þn (14.50) h¼m u0 where u0 is the fundamental frequency and uc is the carrier frequency. The harmonic coefficients Amn and Bmn in (14.49) are evaluated for each 60 degrees sextant using the double Fourier integral. These coefficients are evaluated for a single VSC (noninterleaved VSC) under SVM scheme for phase a as
1 pﬃﬃﬃ h i 3 p p p 3p np C B C B sin ðm þ nÞ M Jn q M þ 2cos Jn q C B6 4 2 4 6 C B C B pﬃﬃﬃ C B 1 mp np np 3p 3 p B þ sin ns0 C cos sin J0 q M J0 q M C B n 2 2 6 4 4 C B C B C B N X 1 p p p C B sin ½m þ k cos ½n þ k sin ½n þ k C Bþ C B nþk 2 2 6 C B k¼1 C B 4Vdc B ksn C Amnva1o ¼ 2 B C p ﬃﬃ ﬃ C B qp C B J q 3p M þ 2cos ½2n þ 3k p J q 3p M C B k k 4 6 4 C B C B C B N X C B 1 p p p C Bþ sin ½m þ k cos ½n k sin ½n k C B 2 2 6 C B k¼1 n k C B ksn C B C B p ﬃﬃ ﬃ C B 3p p 3p C B Jk q M A @ Jk q M þ 2cos ½2n 3k 4 6 4 0
(14.51)
Parallel operation of power converters and their filters Chapter  14
423
where Vdc is the DClink voltage, M is the modulation index, and q ¼ m þ nðu0 =uc Þ. The expressions contain Jy ðzÞ, which represents the Bessel functions of the first kind of the order y and argument z. The double summation term in (14.49) is the ensemble of all possible frequencies, formed by taking the sum and the difference between the carrier harmonics, the fundamental waveform, and its associated baseband harmonics. The harmonic coefficients for phase b are also obtained to evaluate the harmonic performance of the linetoline voltage vab . The harmonic spectrum of the linetoline voltage of a single VSC with the SVM scheme is shown in Fig. 14.12 for the modulation index of M ¼ 1. The major harmonic components are concentrated around the carrier frequency harmonics, as shown in Fig. 14.12A. To evaluate the harmonic performance, the harmonic coefficients of the average pole voltage of phase a are calculated for two symmetrically interleaved VSCs under SVM as
(A)
(B)
FIGURE 14.12 Harmonic spectrum of the linetoline voltage under space vector modulation scheme with modulation index M ¼ 1 and pulse ratio uc =u0 ¼ 33. The pulse ratio is defined as the ratio of the carrier frequency to the fundamental frequency. (A) Linetoline voltage of the single voltage source converter (noninterleaved), (B) Linetoline voltage of two symmetrically interleaved voltage source converters.
424 Control of Power Electronic Converters and Systems 1 p p 3p np pﬃﬃ3ﬃp C Bp C B cos m sin n M Jn q M þ 2cos Jn q C B 4 2 2 4 6 C B6 C B C B N p p X B 1 p p C C Bþ cos m sin k cos ½n þ k sin ½n þ k C B 2 2 2 6 C B k¼1 n þ k C B ksn C B C B p ﬃﬃ ﬃ C B 4Vdc B 3p 3p p C Amnvao ¼ 2 B Jk q M þ 2cos ½2n þ 3k C M Jk q C B qp 4 4 6 C B C B N X C B 1 p p p p C Bþ cos m sin k cos ½n k sin ½n k C B n k 2 2 2 6 C B k¼1 C B ksn C B C B pﬃﬃﬃ C B C B 3p p 3p C B Jk q M þ 2cos ½2n 3k J M q k A @ 4 6 4 0
(14.52)
Using Eq. (14.52), the harmonic coefficients of the linetoline voltages are given as 0
p ph p p cos m sin n 1 cos 2n 6 2 2 3 np pﬃﬃ3ﬃp 3p Jn q Jn q M þ 2cos M 4 6 4 N p p h X 1 p h p cos m sin k cos n þ k sin n þ k þ nþk 2 2 2 6
1
C B C B C B C B C B C B C B C B C B C B C B C B C B C B k¼1 C B ksn C B h
C B C B 3p p C B M 1 cos ½n þ 3k J q k C B 4 3 C 4Vdc B C p ﬃﬃ ﬃ Amnvab ¼ 2 B pi C B h p p 3 p qp C B þ2 cos ½2n þ 3k cos ½n 3k cos n Jk q M C B 6 6 6 4 C B C B C B N p p X 1 p p C B C B cos m sin k cos ½n k sin ½n k þ C B n k 2 2 2 6 C B k¼1 C B ksn C B C B
h C B 3p p C B M 1 cos ½n 3k q J k C B 4 3 C B C B p ﬃﬃ ﬃ C B h i 3p p p p C B A @ þ2 cos ½2n 3k M cos ½n þ 3k cos n Jk q 4 6 6 6 (14.53)
Parallel operation of power converters and their filters Chapter  14
0 B B B B B B B B B B B Bþ B B B B B B B B 4Vdc B B Bmnvab ¼ 2 B B qp B B B B B B Bþ B B B B B B B B B B B B @
425
1 p p pi p C cos m sin n sin 2n C 6 2 2 3 C C pﬃﬃﬃ C C 3p 3p np C M Jn q M þ 2 cos Jn q C 4 4 6 C C N X 1 p p p p C C cos m sin k cos ½n þ k sin ½n þ k nþk 2 2 2 6 C C k¼1 C ksn C C C C 3p p C Jk q M sin ½n þ 3k C 4 3 C C p ﬃﬃ ﬃ p C 3p p C Jk q M 2cos n sin ½n 3k C 4 6 6 C C C N X 1 p p p p C C cos m sin k cos ½n k sin ½n k nk 2 2 2 6 C C k¼1 C ksn C C C 3p p C Jk q M sin ½n 3k C 4 3 C C pﬃﬃﬃ C p p 3p C A Jk q M 2cos n sin ½n þ 3k 6 6 4 (14.54)
Using Eqs. (14.53) and (14.54), the harmonic spectrum of the linetoline voltage of two symmetrical interleaved VSCs is obtained as shown in Fig. 14.12B. The odd multiple of the carrier frequency harmonics and their sidebands are highly reduced, leading to significant improvements in the harmonic quality.
14.4.3 Harmonic performance evaluation 14.4.3.1 Two symmetrically interleaved VSCs Using the approach illustrated in the section 14.4.2, the NWTHD is obtained for different modulation schemes and the NWTHD variation with the modulation index for different modulation schemes for two symmetrical interleaved VSCs is shown in Fig. 14.14. The carrier frequency is taken to be the same in all cases; thus, the number of commutations in DPWM schemes is 2/3 times than the number of commutation in the SVM. The DPWM schemes demonstrate superior harmonic performance compared to the SVM in the entire operating modulation range. At low modulation indices range, all the DPWM
426 Control of Power Electronic Converters and Systems
(A)
(B)
FIGURE 14.13 Harmonic spectrum of the linetoline voltage under 60 degrees clamp discontinuous (DPWM1) modulation scheme with modulation index M ¼ 1 and pulse ratio uc = u0 ¼ 33. (A) Linetoline voltage of the single voltage source converter (noninterleaved), (B) Linetoline voltage of two symmetrically interleaved voltage source converters.
FIGURE 14.14 Harmonic performance of various pulse width modulation schemes for two symmetrically interleaved voltage source converters. The carrier frequency is the same for all the pulse width modulation schemes.
schemes have a similar harmonic distortion, far more superior than the SVM. For modulation indices higher than 0.6, DPWM1 has the lowest harmonic distortion compared to other methods. This surprising result can be explained using the harmonic energy distribution characteristic of the SVM and DPWM. For the noninterleaved VSC, it is well known that the SVM leads to the lowest harmonic distortion. This is due to the cancelation of the odd harmonics around the first carrier frequency in the linetoline voltage and distribution of the harmonic energy into the outer sideband harmonics and sidebands of the second carrier frequency harmonic group, as shown in Fig. 14.12A. On the other hand, the odd harmonic cancelation around the first carrier frequency group does not happen in
427
Parallel operation of power converters and their filters Chapter  14
DPWM1, and the harmonic energy is mostly distributed with significantly low rolloff in the magnitude of sidebands of the first harmonic carrier group, as shown in Fig. 14.13A. For two symmetrically interleaved VSCs, the sidebands of the first carrier frequency harmonic group are significantly reduced due to the interleaving, whereas the sidebands of the second carrier frequency harmonic group remain unaffected, leading to superior harmonic performance of DPWM1, as shown in Fig. 14.13B. Moreover, the switching losses in the discontinuous modulation schemes are lower compared to the SVM, leading to an improved efficiency. For the applications where the unity power factor operation is required, the use of DPWM1 is advantageous for two symmetrically interleaved VSCs due to low NWTHD and lowest switching losses.
14.4.3.2 Three symmetrically interleaved VSCs A harmonic performance comparison of different modulation schemes for three symmetrically interleaved VSCs is shown in Fig. 14.15B. The SVM demonstrates superior harmonic performance at low and high modulation indices. However, it is important to note that the carrier frequency is taken to be the same for all the PWM schemes for this comparison. Since the switching losses in the discontinuous modulation schemes are lower than the SVM due to fewer switch transitions, the carrier frequency of the discontinuous modulation schemes can be adjusted to achieve the same number of switch transitions as that of the SVM. The harmonic performance of various modulation schemes with the adjusted carrier frequency is shown in Fig. 14.15B. The SVM offers low harmonic distortion at low modulation indices, whereas discontinuous modulation schemes have a superior harmonic performance at high modulation indices.
(A)
(B) 0.4
0.6
0.4 0.2 0.2
0 0
0.2
0.4
0.6
0.8
1
0 0
0.2
0.4
0.6
0.8
1
FIGURE 14.15 Harmonic performance of various pulse width modulation schemes for three symmetrically interleaved voltage source converters (interleaving angle is 120 degrees). (A) The carrier frequency is the same for all the pulse width modulation schemes, (B) The carrier frequency of the discontinuous modulation scheme is 3/2 times compared to the SVM.
428 Control of Power Electronic Converters and Systems
14.4.4 Nearest three vector modulation Since interleaving of the parallel 2L VSCs leads to the multilevel resultant voltage, parallel VSCs can be treated as a multilevel converter and can be modulated using the multilevel modulation schemes. The switched output ! pﬃﬃﬃ. voltages of phase a of each of the VSCs (for Vs+ ¼ 3 3 8 and js ¼ 20+ ) are shown in Fig. 14.16A. The carrier signals of the three VSCs are symmetrically phase shifted by an interleaving angle of 120 degrees. The resultant switched output voltages of all the phases are also shown in Fig. 14.16B, which demonstrate fourlevel voltage waveforms. The switching sequences
(A)
(B)
FIGURE 14.16 Switched output voltages of three symmetrically interleaved voltage source ! pﬃﬃﬃ. converters for Vs+ ¼ 3 3 8 and space vector angle js ¼ 20+. Each of the VSCs is modulated using twolevel space vector modulation. (A) Switched output voltage of phase a of all three VSCs, (B) Resultant switched output voltages of all three phases.
Parallel operation of power converters and their filters Chapter  14
429
220 210 310 311 310 210 220 are employed, where each of the digits represents the voltage level of the resultant output voltage of the phase a, b, and c, respectively (e.g., 210 represents that voltage levels of phase a, phase b, and phase c are level 2, level 1, and level 0, respectively). As the three ! parallel VSCs give fourlevel voltage output, the Vs+ can be projected on the space vector diagram of the fourlevel converter, as shown in Fig. 14.17, where the reference space voltage vector is located in the triangle Ds . It is well known that the lowest harmonic distortion can be achieved by using the nearest three vectors (NTVs) [24], which are located on the vertices of the triangle Ds (210e310e311). However, that is not the case when the symmetrical interleaving is used (an additional voltage vector 220 is also employed), as shown in Fig. 14.16B. Therefore, it is evident that the PS PWM is not an optimal solution for the modulation of the parallel VSCs. This issue can be mitigated by using asymmetrical interleaving. It is shown in Ref. [21] that the interleaving angle strongly influences harmonic performance. For a given modulation scheme, the lowest harmonic performance can be achieved by varying the interleaving angle with the modulation index. Further improvement can be achieved by selecting an optimal combination of switching sequences and interleaving angle to synthesize the ! reference space vector Vs+ [25]. In this scheme, several combinations of the switching sequences and the phaseshift are used in one fundamental cycle of the reference space voltage vector. This would substantially increase the complexity and make it very difficult to implement.
! pﬃﬃﬃ. FIGURE 14.17 Projection of the reference space voltage vector Vs+ ¼ 3 3 8:20+ in the first sextant of the space vector diagram of the fourlevel converter, realized using interleaved operation of three parallel voltage source converters.
430 Control of Power Electronic Converters and Systems
For the parallelconnected 2LVSCs, multilevel voltage waveforms can be achieved by interleaving the carrier signals. Therefore, the parallelconnected 2LVSCs can be treated as a multilevel converter. For the multilevel converter, the harmonic profile of the synthesized voltage can be improved by using the NTV [24]. For the carrier comparison implementation, NTV can be achieved using the phase disposition (PD) modulator. The harmonic performance of the parallel VSCs with the symmetrical interleaving and with PD modulator is evaluated and shown in Fig. 14.18. However, the conventional implementation of PD PWM implementation cannot be readily used for the modulation of interleaved VSCs, as it introduces a DC component in the circulating current during band transition [26]. As a result, the magnetic components present in the circulating current path may saturate. This issue can be addressed by introducing additional switchings during the band transition, as also presented in Ref. [26,27]. The load current waveform for the PDmodulated parallel VSCs is shown in Fig. 14.19B. The load current ripple is significantly lower compared to the ripple in the load current with symmetrical interleaving, as shown in Fig. 14.19A.
14.5 Circulating current suppression in parallel interleaved converters When VSCs are connected in parallel, the circulating current flows between the VSCs due to the control asymmetry and the impedance mismatch. When the carriers are interleaved, the switched output voltages of the interleaved parallel legs are phase shifted. As a result, the instantaneous voltage difference exists between parallel VSCs, which further increases the highfrequency circulating current. Therefore, the circulating current should be suppressed to realize the full potential of the interleaved carriers in parallelconnected VSCs. The highfrequency circulating current introduced by the carrier
FIGURE 14.18 Harmonic performance of three parallel interleaved converters under nearest three vector modulation. The NWTHD of various modulation schemes with the conventional phaseshifted carrier signals are also shown for the comparison.
Parallel operation of power converters and their filters Chapter  14
431
(A)
(B)
FIGURE 14.19 Load current waveform for three parallel converters. (A) Symmetrically interleaved VSCs with three 120 degrees phaseshifted carrier signals, (B) Phase disposition modulation [26] scheme for parallel interleaved VSCs.
432 Control of Power Electronic Converters and Systems
interleaving can be eliminated by providing galvanic isolation [2] or can be reduced to a reasonable limit by introducing an impedance in the circulating current path. Various inductive components that are used to offer large impedance to the highfrequency circulating current, including CI [1,3], CM inductor [4], and integrated inductor, are discussed in this section.
14.5.1 Galvanic isolation The circulating current can be avoided by providing galvanic isolation between the parallel VSCs using the multiple winding line frequency transformer [2], as shown in Fig. 14.20. This solution is preferred in applications requiring a transformer for voltage matching and for providing isolation. The harmonic filter inductor Lf represents a series combination of the leakage inductance of the transformer and external inductor (if leakage inductance is not sufficient). For the three parallel VCSs, the phaseshifted voltages are applied to the isolated primary windings, leading to the cancelation of the major harmonic component that is concentrated around the first and second carrier frequency harmonics in the magnetic flux. As a result, the induced electromotive force (EMF) has major harmonic components that are concentrated around the third carrier frequency component. As a result, the load current has very low harmonic distortion, as it is shown in Fig. 14.21B. On the contrary, the individual VSC current has a major harmonic current concentrated around the first and second carrier harmonic frequency, as shown in Fig. 14.21A. This is due to the fact that these components of the switched output voltage appear across the Lf (since these harmonic components are negligible in the induced EMF). As a result, these harmonic components in the individual VSC currents are only limited by the Lf . Therefore, significant harmonic frequency components are present in the individual VSC currents if a sufficiently large value of Lf is not selected. This current will flow through the primary windings of the transformer and external filter inductors, leading to significant winding losses.
FIGURE 14.20 Three parallel interleaved voltage source converters. A line frequency isolation transformer with multiple windings is used for providing galvanic isolation.
Parallel operation of power converters and their filters Chapter  14
(A)
433
(B)
FIGURE 14.21 Simulated current waveform for three parallel interleaved VSCs with isolated transformer windings (see Fig. 14.20). Currents are normalized to the rated current. (A) Individual converter leg current ia1 , (B) Load current ia .
14.5.2 Coupled inductor The circulating current between the parallel interleaved VSCs can be suppressed by introducing a high impedance in the circulating current path without affecting the impedance in the load current path. This can be achieved by using CI. It suppress the circulating current by providing magnetic coupling between the parallel interleaved legs of the corresponding phases. The magnetic structure of the CI in one of the phases in a threephase system for N parallel interleaved VSCs is shown in Fig. 14.22A. It consists of N magnetically coupled coils. The start terminal of the coils is connected to the AC terminals of the VSC legs of phase a (a1 , a2 , aN ), whereas the end terminals are connected to form a common connection node ac .
(A)
(B)
FIGURE 14.22 Circulating current suppression in parallel interleaved voltage source converter using coupled inductor. (A) Magnetic structure of the coupled inductor, (B) Equivalent electric circuit of the phase a of N parallel interleaved voltage source converters with the coupled inductor.
434 Control of Power Electronic Converters and Systems
14.5.2.1 Equivalent electric circuit The phaseshifted pole voltages are applied between the starting terminals of the coils and fictitious midpoint of the DC bus o, which can be expressed as V P ¼ RI þ L
d IþV dt
(14.55)
where the V P is the pole voltage vector V P ¼ ½ va1 o va2 o /
vaN o T
(14.56)
I is a VSC current vector I ¼ ½ ia1
ia2 /iaN T
(14.57)
and V ¼ ½ Vac o Vac o/
Vac o T
(14.58)
L and R represent the inductance and resistance matrix, respectively. 3 2 La1 a1 La1 a2 / La1 aN 7 6 7 6L 6 a2 a1 La2 a2 / La2 aN 7 (14.59) L¼6 7 6 « « « 7 5 4 LaN a1 LaN a2 / LaN aN 2 r 6 a1 6 0 6 R¼6 6 « 4 0
0/ ra2 / « 0/
3 0 7 0 7 7 7 « 7 5 raN
(14.60)
Let the average of the switched output voltages of phase x of N parallel VSCs be Vxv o and it is represented as V xv o ¼
n 1 X Vx o; where 1 < k N N k¼1 k
(14.61)
where Vxk o represents the pole voltage of phase x of kth VSC with respect to the common reference point o. Assuming an equal line current sharing between the parallel VSCs, the common component of the leg current is obtained as ixk ;l ¼
ix N
(14.62)
Parallel operation of power converters and their filters Chapter  14
435
Using Eq. (14.9) and (14.62), the current through the coils can be represented as ixk ¼
ix þ ixk ;c N
(14.63)
where 1Nx is the load current component of VSC current and ixk ;c is the circulating current component. Since the common current component is in phase for all the coils of the CI, the flux produced by this component (referred to as the common flux component) does not link with other coils and completes its path through the air (c.f. fa1 ;l in Fig. 14.22A). This leakage flux has a dominant fundamental frequency component and may lead to significant eddy current losses in the laminated magnetic structure (due to the flux component which is orthogonal to the lamination layers) and the metal enclosures [28], especially in highcurrent applications. The voltage of the common node (vxc o) is the average of all the pole voltages (see Eq. 14.61). Therefore, the phaseshifted harmonic components present in the pole voltage will be canceled out in vxc o. As a result, the phaseshifted harmonic components only appear across the CI coils. The flux induced by these phaseshifted harmonic frequency components of the pole voltages will link with other coils of the CI due to the availability of a low reluctance path through other limbs of the CI. Considering a symmetrical magnetic structure, the inductances can be represented as Laj aj ¼ Ls for all 1 j N
(14.64)
Laj ak ¼ Lm
(14.65)
and
for all 1 j N; 1 k N; and jsk Neglecting the resistance in Eq. (14.55) and segregating the load current and circulating current components of the VSC current yields n 1 X Ls ðN 1ÞLm dia (14.66) Vxk o vac o ¼ N k¼1 N dt Using Eqs. (14.61) and (14.66), the behavior of the load current component can be described as vav o vac o ¼ LfCI
dia dt
(14.67)
where LfCI is the inductance offered to the load current and it is given as Ls ðN 1ÞLm LfCI ¼ (14.68) N
436 Control of Power Electronic Converters and Systems
The behavior of the circulating current components can be described as d V P ¼ Lc I a;c þ Vav o dt
(14.69)
where I a;c ¼ ½ ia1 ;c ia2 ;c . ian ;c T 3 2 Ls Lm / Lm 7 6 6 L Ls / Lm 7 7 6 m Lc ¼ 6 7 6 « « « « 7 5 4 Lm Lm / Ls
(14.70)
(14.71)
Since a high value of the mutual inductance Lm can be achieved using the CI, the circulating current can be effectively suppressed.
14.5.2.2 Impact of the modulation scheme The PWM scheme has a strong influence on the core losses. Therefore, the design of the CI, especially the thermal design, is strongly impacted by the selection of the PWM scheme. The impact of the PWM scheme on the core losses of the CI for two parallel VSCs has been analyzed. By neglecting the leakage flux, the flux linkage in the CI is given as Z la ðtÞ ¼ la1 ðtÞ þ la2 ðtÞ ¼ ðva1 O va2 OÞdt (14.72) The differential voltage that appears across the CI (va1 O va2 O) determines the flux linkage, which entirely depends on the PWM scheme. The flux density in the CI can be obtained as Z 1 ðva1 O va2 OÞdt Ba ðtÞ ¼ (14.73) 2Nc Ac where Nc is the number of turns in a coil, and Ac is the core crosssectional area. The variation in the peak flux linkage with the modulation index for different PWM schemes is plotted in Fig. 14.23 [29]. The maximum value of the peak flux linkage is the same in all PWM schemes. However, the flux linkage pattern is different. As a result, the core losses would be different in each of the schemes, which is an important factor in determining the size and efficiency of the CI. The impact of the modulation schemes on the CI core losses is shown in Fig. 14.24. The core losses are obtained using the improved generalized Steinmetz equation. The core losses are normalized with respect to the core losses in the case of the SVM. When compared to the SVM, all the discontinuous PWM schemes lead to lower core losses at low modulation
Parallel operation of power converters and their filters Chapter  14
437
FIGURE 14.23 Variation of the maximum peak flux linkage with the modulation index. The flux linkage is normalized with respect to the Vdc Ts .
FIGURE 14.24 The core losses in the coupled inductor for different PWM schemes. The core losses are normalized with respect to that of the SVM. The carrier frequency is taken to be the same in all cases.
indices. For high modulation indices, the use of DPWM1 leads to the highest losses, followed by DPWM2. The core losses in the case of DPWM3 are lowest in the entire modulation index range.
14.5.3 Commonmode inductor Although the CI is very effective in suppressing the circulating current, it has a serious limitation when it comes to modularity. Moreover, unequal current sharing between the parallel VSCs is not possible without overdesigning or saturating the CI. The circulating current (which mainly appears as a CM current) can be suppressed using the CM inductor [4], and it is the preferred solution when unequal load sharing between the parallel VSCs and modularity is desired. The CM inductor is realized using three coils that are wound in the same direction. The coils are connected to the threephase AC terminals of the VSC, as shown in Fig. 14.25. Each VSC also uses differential mode (DM) inductor for suppressing the harmonics in the load current.
438 Control of Power Electronic Converters and Systems
FIGURE 14.25 Schematic illustration of the parallel interleaved voltage source converters. Each converter is connected to the common AC connection using a series combination of differential and commonmode inductors.
For the system shown in Fig. 14.25, the kth VSC currents can be represented as 3 2 3 2 3 2 3 2 Lsc þ Lsd Lmc Lmd Lmc Lmd iak vao vak o 7 6 7 6 7 6 7 6 6 vbk o 7 ¼ 6 Lmc Lmd Lsc þ Lsd Lmc Lmd 7 d 6 ibk 7 þ 6 vbo 7 (14.74) 5 4 5 dt 4 5 4 5 4 v ck o Lmc Lmd Lmc Lmd Lsc þ Lsd ick vco where the subscript k represents the kth VSC. Lsc and Lmc are the self and mutual inductances of the CM inductor, whereas Lsd and Lmd are the self and mutual inductances of the DM inductor. vxk o and ixk represent pole voltage and VSC current of phase x of kth VSC, vxo is the voltage at the load terminal. The VSC current can be represented as i xk ¼
ix þ icmk N
(14.75)
where ix is the load current of phase x and icmk is the CM current of kth VSC and represented as icmk ¼
iak þ ibk þ ick 3
(14.76)
Parallel operation of power converters and their filters Chapter  14
439
The dynamic behavior of the CM current of kth VSC can be represented as vcmk ¼ ðLs 2LM Þ
dicmk þ vcm dt
(14.77)
where Ls ¼ Lsc þ Lsd and Lm ¼ Lmc Lmd , vcmk is the CM voltage of kth VSC vcmk ¼
vak o þ vbk o þ vck o 3
(14.78)
vao þ vbo þ vco 3
(14.79)
and vcm ¼
By assuming a strong coupling between the DM inductor, the flux linkage in the CM inductor can be given as Z N Z N1 1 X lcmk ðtÞ ¼ vcmk dt vcmj dt (14.80) N N j¼1 jsk
The volume of the CM inductor is proportional to the maximum value of the CM flux linkage, which strongly depends on the PWM scheme. The maximum value of the peak CM flux linkage as a function of the modulation index for two parallel VSCs is plotted in Fig. 14.26 [30]. For the SVM, the maximum value of the peak flux linkage increases as the modulation index decreases. Therefore, for the applications demanding operation over the full modulation range, the CM inductor has to be designed for the maximum flux linkage, which occurs at low modulation indices. On the other hand, the CM is subjected to maximum flux linkage for a modulation index M ¼ 2=3 if DPWM1 is employed. The maximum value of the peak flux linkage in DPWM is almost 33% lower than the SVM, leading to a smaller CM inductor.
FIGURE 14.26 Comparison of the maximum values of the peak CM flux linkage as a function of the modulation index. The flux linkage is normalized with respect to Vdc Ts .
440 Control of Power Electronic Converters and Systems
14.5.4 Integrated inductor The circulating current between the parallel VSCs can be suppressed by introducing impedance in the circulating current path. This can be achieved using the CI. Therefore, for the interleaved operation of parallel VSCs, two distinct magnetic components may be required: l l
Inductor for the circulating current suppression. Line filter inductor Lf (commonly referred to as a boost inductor) for improving the line current quality.
The volume of the inductive components can be reduced by integrating both of these functionalities into a single magnetic component, as shown in Fig. 14.27 [31]. The simplified arrangement of the magnetic structure is shown in Fig. 14.28. It consists of three CIs for a threephase system. The CIs of all the three phases are magnetically coupled using the top and bottom bridge yokes. The necessary air gaps are inserted between the cells and the bridge yokes. The top and bottom yokes provide low reluctance for the magnetic coupling between the phases. Considering a symmetrical cell structure, the inductances can be represented as Laj bj ¼ Lbj cj ¼ Lcj aj ¼ Lm for all 1 j N
(14.81)
Laj bk ¼ Lbj ck ¼ Lcj ak y0 for all 1 j N; 1 k N; and jsk
(14.82)
FIGURE 14.27 Magnetic structure of the proposed integrated threephase inductor for N number of parallelconnected VSCs (N ¼ 4 in this illustration).
Parallel operation of power converters and their filters Chapter  14
441
FIGURE 14.28 Simplified magnetic structure of the integrated inductor shown in Fig. 14.27.
Lxj xk ¼ Lm1 for all 1 j N; 1 k N; and jsk
(14.83)
The ve sign is used to represent the Lm and Lm1 . Neglecting the leakage flux, the selfinductance of each of the coils is given as Laj aj ¼ Lbj bj ¼ Lcj cj ¼ ðN 1ÞLm1 þ 2Lm for all 1 j N
(14.84)
Using these inductance values and averaging the pole voltages of each of the phases gives 2 2 3 3 2 3 vav o vac o 2Lm Lm Lm ia 6 7 16 7d6 7 6 vbv o vbc o 7 ¼ 6 Lm 2Lm Lm 7 6 ib 7 (14.85) 4 5 N4 5 dt 4 5 vcv o vcc o Lm Lm 2Lm ic For the threephase threewire system, ia þ ib þ ic ¼ 0 and the inductance offered to the resultant line current is given as Lf ¼
v x v o v xc o 3 ¼ Lm N dix =dt
(14.86)
The behavior of the circulating current can be described by subtracting the average pole voltage from the pole voltages of the corresponding phases and further simplification of those equations gives d V Sx ¼ Lc I x;c þ Vxv o dt
(14.87)
where V S x ¼ ½ v x1 o v x2 o .
v xN o T
(14.88)
442 Control of Power Electronic Converters and Systems
FIGURE 14.29 Equivalent electrical circuit of the parallel interleaved VSCs with the integrated inductor.
I x;c ¼ ½ ix1 ;c 2 6 6 6 Lc ¼ 6 6 4
ix2 ;c
.
ixN ;c T
ðN 1ÞLm1
Lm1
/
Lm1
Lm1
ðN 1ÞLm1
/
Lm1
« Lm1
« Lm1
« « / ðN 1ÞLm1
(14.89) 3 7 7 7 7 7 5
(14.90)
Using (14.86) and (14.87), the electrical equivalent circuit is obtained and it is shown in Fig. 14.29. Here, xv is the virtual common point and the potential of this point with respect to the midpoint of the DC link is the averaged pole voltage vxv o. The potential difference of vxv o vxc o appears across the line filter inductor Lf , as shown in Fig. 14.29. The advantage offered by the integrated inductor in terms of the size reduction is demonstrated in Ref. [31] by comparing the volume of the integrated inductor with the stateoftheart solution of using three CIs and the threephase line filter inductor, where 15% reduction in the active material usage was demonstrated for a specific PCS with three interleaved VSCs.
14.6 Summary The parallel connection of the VSCs is used in multiple applications to realize modular, reliable, and costeffective PCS. To achieve the full potential of the parallel operation of VSCs, it is very important to reduce the circulating current to a reasonable level. In a PCS with parallelconnected VSCs, the circulating current may flow between the VSCs because of the filter impedance and semiconductor device parameter mismatch, application of different
Parallel operation of power converters and their filters Chapter  14
443
voltage vectors, and deadtime effect, etc. The flow of the unwanted circulating current leads to additional losses in the active and passive components present in the circulating current path. The impacts of the application of different applied voltage vectors and impedance mismatch are analyzed. Moreover, the correlation between the circulating current that flows between the VSC legs of a particular phase and the CM circulating current is derived. Various control schemes for load sharing and circulating current reduction are discussed. The control scheme with the central controller can achieve good load sharing. However, it requires fast communications between the parallel VSCs, which compromises modularity and degrades the redundancy. The droop control scheme can be used to avoid critical communication lines and to achieve good load sharing. The virtual output impedance control can be incorporated to alleviate the impedance mismatch between the parallel VSCs. The droop control, combined with the virtual output impedance control and circulating current controller, can achieve good load sharing and minimize circulating current. The lowfrequency component of the CM circulating current can be reduced by controlling the average CM voltage using a closedloop circulating current controller. The average value of the CM voltage in a switching period can be controlled by appropriately sharing the zero vector dwell time between two zero voltage vectors. The harmonic performance of the parallel VSCs can be improved by interleaving control. The harmonic performance of various conventional PWM schemes with the symmetrical interleaving is evaluated, where it is shown that the DPWM1 modulation scheme demonstrates superior harmonic performance and low switching losses for the two parallel VSCs, whereas for three parallel VSCs, the continuous SVM has low harmonic distortion. In fact, parallelconnected 2L VSCs can be treated as a multilevel converter. Therefore, superior harmonic performance can be achieved by modulating parallel VSCs as a single multilevel converter using an NTV (PD) modulation scheme. However, the conventional PD modulator cannot be used for parallel VSCs as it may saturate the filter inductors. A detailed discussion on the modified PD modulator, suitable for the parallel VSCs, has been provided. In the interleaving control, different voltage vectors are applied to the parallel VSCs to improve the harmonic performance. The application of the different voltage vectors gives rise to the highfrequency circulating current. Various filter arrangements for the highfrequency circulating current suppression are discussed. The impact of the PWM scheme on the peak flux linkage and core losses is also analyzed. It is also shown that the size of the passive components can be reduced through the magnetic integration of the circulating current filter and line filter inductor.
444 Control of Power Electronic Converters and Systems
References [1] K. Matsui, Y. Murai, M. Watanabe, M. Kaneko, F. Ueda, A pulsewidthmodulated inverter with parallel connected transistors using currentsharing reactors, IEEE Trans. Power Electron. 8 (2) (April 1993) 186e191. [2] H. Akagi, A. Nabae, S. Atoh, Control strategy of active power filters using multiple voltagesource pwm converters, IEEE Trans. Ind. Appl. IA22 (3) (May 1986) 460e465. [3] I.G. Park, S.I. Kim, Modeling and analysis of multiinterphase transformers for connecting power converters in parallel, in: Proc. 28th Annual IEEE Power Electronics Specialists Conference, vol. 2, June 1997, pp. 1164e1170. [4] L. Asiminoaei, E. Aeloiza, P.N. Enjeti, F. Blaabjerg, Shunt activepower filter topology based on parallel interleaved inverters, IEEE Trans. Ind. Electron. 55 (3) (March 2008) 1175e1189. [5] A.P. Martins, A.S. Carvalho, A.S. Araujo, Design and implementation of a current controller for the parallel operation of standard upss, Proc. 21st Annual Conf. IEEE Ind. Electron. 1 (November 1995) 584e589. [6] J. Holtz, K. Werner, Multiinverter ups system with redundant load sharing control, IEEE Trans. Ind. Electron. 37 (6) (December 1990) 506e513. [7] J.F. Chen, C.L. Chu, Combination voltagecontrolled and currentcontrolled pwm inverters for ups parallel operation, IEEE Trans. Power Electron. 10 (5) (September 1995) 547e558. [8] X. Sun, Y.S. Lee, D. Xu, Modeling, analysis, and implementation of parallel multiinverter systems with instantaneous averagecurrent sharing scheme, IEEE Trans. Power Electron. 18 (3) (May 2003) 844e856. [9] Y.J. Cheng, E.K.K. Sng, A novel communication strategy for decentralized control of paralleled multiinverter systems, IEEE Trans. Power Electron. 21 (1) (January 2006) 148e156. [10] T. Kawabata, S. Higashino, Parallel operation of voltage source inverters, IEEE Trans. Ind. Appl. 24 (2) (March 1988) 281e287. [11] M.C. Chandorkar, D.M. Divan, R. Adapa, Control of parallel connected inverters in standalone ac supply systems, IEEE Trans. Ind. Appl. 29 (1) (January 1993) 136e143. [12] T.F. Wu, Y.K. Chen, Y.H. Huang, 3c strategy for inverters in parallel operation achieving an equal current distribution, IEEE Trans. Ind. Electron. 47 (2) (April 2000) 273e281. [13] H. Van Der Broeck, U. Boeke, A simple method for parallel operation of inverters, in: Proc. 20th International Telecommunications Energy Conference, October 1998, pp. 143e150. [14] Y. Pei, G. Jiang, X. Yang, Z. Wang, Automasterslave control technique of parallel inverters in distributed ac power systems and ups, in: Proc. IEEE 35th Annual Power Electronics Specialists Conference, vol. 3, June 2004, pp. 2050e2053. [15] A. Engler, N. Soultanis, Droop control in lvgrids, in: Proc. International Conference on Future Power Systems, November 2005, p. 6. [16] J.M. Guerrero, L. Garcia de Vicuna, J. Matas, M. Castilla, J. Miret, Output impedance design of parallelconnected ups inverters with wireless loadsharing control, IEEE Trans. Ind. Electron. 52 (4) (August 2005) 1126e1135. [17] S.J. Chiang, J.M. Chang, Parallel control of the ups inverters with frequencydependent droop scheme, in: Proc. IEEE 32nd Annual Power Electronics Specialists Conference, vol. 2, June 2001, pp. 957e961, vol. 2. [18] B. Wei, J.M. Guerrero, J.C. Va´squez, X. Guo, A circulatingcurrent suppression method for parallelconnected voltagesource inverters with common dc and ac buses, IEEE Trans. Ind. Appl. 53 (4) (July 2017) 3758e3769.
Parallel operation of power converters and their filters Chapter  14 [19]
[20]
[21]
[22]
[23] [24] [25] [26]
[27]
[28]
[29]
[30]
[31]
445
Z. Ye, D. Boroyevich, J.Y. Choi, F.C. Lee, Control of circulating current in two parallel threephase boost rectifiers, IEEE Trans. Power Electron. 17 (5) (September 2002) 609e615. S.K.T. Miller, T. Beechner, J. Sun, A comprehensive study of harmonic cancellation effects in interleaved threephase vscs, in: Proc. IEEE Power Electronics Specialists Conference, June 2007, pp. 29e35. D. Zhang, F. Wang, R. Burgos, R. Lai, D. Boroyevich, Impact of interleaving on ac passive components of paralleled threephase voltagesource converters, IEEE Trans. Ind. Appl. 46 (3) (May 2010) 1042e1054. J.S. Siva Prasad, G. Narayanan, Minimization of grid current distortion in parallelconnected converters through carrier interleaving, IEEE Trans. Ind. Electron. 61 (1) (January 2014) 76e91. D.G. Holmes, T.A. Lipo, Pulse Width Modulation for Power Converters: Principles and Practice, WileyIEEE Press, Hoboken, NJ, 2003. N. Celanovic, D. Boroyevich, A fast spacevector modulation algorithm for multilevel threephase converters, IEEE Trans. Ind. Appl. 37 (2) (March 2001) 637e641. X. Mao, A.K. Jain, R. Ayyanar, Hybrid interleaved space vector pwm for ripple reduction in modular converters, IEEE Trans. Power Electron. 26 (7) (July 2011) 1954e1967. G. Gohil, L. Bede, R. Teodorescu, T. Kerekes, F. Blaabjerg, Fluxbalancing scheme for pdmodulated parallelinterleaved inverters, IEEE Trans. Power Electron. 32 (5) (May 2017) 3442e3457. B. Cougo, G. Gateau, T. Meynard, M. BobrowskaRafal, M. Cousineau, Pd modulation scheme for threephase parallel multilevel inverters, IEEE Trans. Ind. Electron. 59 (2) (February 2012) 690e700. I. Trintis, G. Gohil, M. Wurzer, M. Franzen, S.L. Pallesgaard, S. Munk Nielsen, P. Carne Kjaer, Line reactor for parallelinterleaved high power inverters, in: Proc. 19th European Conference on Power Electronics and Applications (EPE’17 ECCE Europe), September 2017, pp. 1e10. G. Gohil, L. Bede, R. Maheshwari, R. Teodorescu, T. Kerekes, F. Blaabjerg, Parallel interleaved vscs: influence of the pwm scheme on the design of the coupled inductor, in: Proc. 40th Annual Conference of the IEEE Industrial Electronics Society, October 2014, pp. 1693e1699. G. Gohil, R. Maheshwari, L. Bede, T. Kerekes, R. Teodorescu, M. Liserre, F. Blaabjerg, Modified discontinuous pwm for size reduction of the circulating current filter in parallel interleaved converters, IEEE Trans. Power Electron. 30 (7) (July 2015) 3457e3470. G. Gohil, L. Bede, R. Teodorescu, T. Kerekes, F. Blaabjerg, An integrated inductor for parallel interleaved threephase voltage source converters, IEEE Trans. Power Electron. 31 (5) (May 2016) 3400e3414.
This page intentionally left blank
Chapter 15
Advanced power control of photovoltaic systems Ariya Sangwongwanich, Jinkui He, Yiwei Pan Aalborg University, Aalborg, Denmark
15.1 Introduction Conventional control of photovoltaic (PV) system aims at maximizing the PV power production with the maximum power point tracking (MPPT) control. This control method is mandatory for maximizing the energy harvesting of the PV system and thereby minimize the levelized cost of energy of the whole PV plant [1]. However, as the penetration level of the PV systems increases, the variability of the PV power generation, which varies with the weather condition (e.g., solar irradiance and temperature), may impose more and more gridintegration challenges. For instance, overloading of the grid during the peak PV power generation periods (e.g., midday) may occur more frequently in the distribution network with a high PV penetration [2]. Grid voltage fluctuation due to the intermittent PV power generation is another concern, which can occur during a cloudy day [3]. Moreover, the system operator may also face a challenge related to frequency regulation capability, since a majority of PV systems cannot easily be dispatched [4,5]. The above concerns have driven new requirements to the control functionality of PV systems in the grid code. Most of the grid codes, especially for countries with high penetration level of PV systems, have been revised and included new requirements in terms of advanced control functionalities of PV systems [6e10]. Instead of always operating with MPPT control, the PV systems are expected to provide certain flexibility in the power regulation. For instance, active power control functionalities such as absolute power constraint, delta power constraint, and power ramprate constraint have been defined in some grid codes, and the PV systems need to fulfill these requirements [8]. In order to achieve these requirements, the PV systems should be able to regulate its output power to a certain limit during the operation (depending on the constraints), which can be enabled through the constant power generation (CPG) control strategies. Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000123 Copyright © 2021 Elsevier Ltd. All rights reserved.
447
448 Control of Power Electronic Converters and Systems
15.2 Overview of PV inverter control In general, two main fundamental tasks should be achieved by the gridconnected PV inverters: (1) the MPPT control to extract the maximum available power from the PV panels, and (2) the injection of grid current with high power quality. For the grid current control, a detailed discussion has been provided in Chapter 8, and thus, only the most popular control structures of PV inverters will be discussed. For the MPPT control, two of the most popular MPPT methods, the perturb and observe (P&O) MPPT and fractional open circuit voltage MPPT, are introduced to exemplify the operating principle of the MPPT control.
15.2.1 Control structure Although many topologies have been proposed and applied to the PV applications, they can generally be divided into two categories: the singlestage and doublestage configurations [11], as shown in Fig. 15.1, where the control diagrams are also illustrated. For the singlestage configuration, the inverter is responsible for both of the aforementioned tasks, which means that the
FIGURE 15.1 Configurations of photovoltaic (PV) inverter systems: (A) the singlestage PV system and (B) the doublestage PV system, where ginv and gdc are the gate signals for the inverter and the DCeDC converter, respectively, POC is the point of connection, and Cdc denotes for the DClink capacitance.
Advanced power control of photovoltaic systems Chapter  15
449
dynamics of the PV power point tracking will affect the dynamics of the power control of the inverter. On the other hand, the interaction between these two control tasks can be decoupled with the doublestage configuration, where the MPPT is performed on the DCeDC stage, and the inverter is in charge of power injection. It can be observed in Fig. 15.1 that the output of the MPPT controller can be different variables depending on the configuration of the , power reference P , or the duty system, being the PV voltage reference VPV PV cycle d . For the inverter control, it can be realized with two cascaded loops: the power/voltage outer loop and the current inner loop [12], as shown in Fig. 15.2. A simple PI regulator can be adopted for the outer loop to regulate the real power or DC voltage according to the reference from the MPPT controller, and generate a current reference id for the inner loop. The inner loop can be designed under various reference frames, i.e., the natural abc, the stationary ab, and the rotating dqreference frames. In the dq frames, PI regulators are always employed for the inner loop, and in the abc and ab frames, PR regulators are recommended to ensure the zero steadystate errors for the inner loop, as shown in Fig. 15.2A and B, respectively. For the threephase systems, the control is typically in the ab and dqreference frames, with the help of Clarke and Park transformations, as it has been discussed in
(A) id
* − P* or VPV +
PI
vd
id* + −
+ + vd*
PI
vα*
dq
αβ
* vinv
///
*
Q + −
PI
iq* +
Q
+
PI
−
vq* +
iq
vq
vβ*
αβ
abc
Modulator
P or VPV
ginv
θg
Power/voltage outer loop Current inner loop
(B)
Q* + −
PI PI
id*
iq*
Q
Power/voltage outer loop
dq αβ θg
iα* + − iβ* + − iβ
PR PR
* + + vα
vβ*
+ +
αβ
* vinv
///
abc
Modulator
* − P* or VPV +
vα
iα
P or VPV
ginv
vβ
Current inner loop
FIGURE 15.2 Control structure of photovoltaic inverter systems: (A) in the rotating dq frames using PI controllers, and (B) in the stationary ab frames using PI and PR controllers. Here, vinv is the modulation index for the inverter.
450 Control of Power Electronic Converters and Systems
Chapter 8. In singlephase systems, the inner loop control can also be designed under the ab and dqreference frames, with the introduction of an imaginary AC coordinate [13]. Moreover, it can be seen in Fig. 15.1 that the phase angle of the grid voltage is also required for the gridconnected control of the inverter, which can be extracted by a phaselocked loop. The dualloop control and grid synchronization have been discussed in Chapters 8 and 9.
15.2.2 MPPT algorithm Since the irradiance and temperature vary throughout a day, the MPPs of PV panels will also change accordingly. Therefore, to extract the maximum power from PV panels regardless of the variation of environmental conditions, the MPPT control should be implemented in the PV inverters. Despite the fact that hundreds of MPPT algorithms have been proposed, they can generally be divided into two categories: the extremumseeking algorithms and the characteristicbased algorithms [14]. In general, there is always a tradeoff between efficiency (e.g., tracking accuracy) and complexity (e.g., parameterization) for different MPPT algorithms. For the extremumseeking algorithms, the optimal operating point is online searched, while for the characteristicbased algorithms, the operating point is adjusted mainly according to initial characteristics (e.g., according to the datasheet of PV panels). In the following, two typical MPPT methods are introduced to exemplify these two kinds of algorithms.
15.2.2.1 Perturb and observe MPPT One of the simplest extremumseeking MPPT algorithms is the P&O MPPT method, which is also the most popular method in practical applications. The flowchart of the P&O algorithm is shown in Fig. 15.3. As shown in Fig. 15.3, the PV power is online calculated and compared with the power value of the previous MPPT period. If the power increases, the MPPT controller will continue the perturbation following the same direction. If the power decreases, the perturbation direction will be reversed. To better illustrate the operating principle of the P&O algorithm, in Fig. 15.4, the operating points of the PV panels are alphabetically marked starting from the initial reference value to the steady state. It can be clearly observed from the figure that the operating point will oscillate around the MPP in a steady state. This means that a certain amount of power in proportion to the perturbation stepsize will be lost because the operating point cannot stay exactly at the MPP all the time [14,15]. Besides, this oscillation may also introduce interharmonics to the grid [16]. Moreover, in fastchanging environmental conditions, the P&O algorithm may loose its MPPT performance [17]. To implement the P&O algorithm, voltage and current sensing circuits are both needed, which will also increase
Advanced power control of photovoltaic systems Chapter  15
451
Initialization * Set Pk−1 = 0; set VPV = V0* ; set vstep; set Inc_dir = +1
Pk−1 = Pk; Sample VPV and IPV; Pk = VPV ∙ IPV; Y
Pk − Pk−1 = 0 N Y
Pk − Pk−1 > 0 N Inc_dir = Inc_dir ∙ (−1) * * VPV = VPV + Inc_dir ∙ vstep
Wait until the next MPPT period
FIGURE 15.3 Flowchart of the P&O MPPT algorithm, where Pk denotes the measured power on the kth MPPT period. V0 and vstep are the initial voltage reference and perturbation stepsize for MPPT, respectively. Inc_dir is the increment direction for the MPPT perturbation, with its value being þ1 or 1.
PMPP
b a
c MPPd f e
vstep VOC
V0* VMPP
FIGURE 15.4 Illustration of the P&O MPPT algorithm on the P/V curve of PV panels, where VMPP and PMPP refer to the voltage and power on the MPP, respectively, and VOC refers to the open circuit voltage of PV panels.
452 Control of Power Electronic Converters and Systems
the cost and volume of the system. However, despite these limitations, the P&O is still a simple and effective method to maximize the power utilization of the PV panels.
15.2.2.2 Fractional open circuit voltage MPPT The fractional open circuit voltage MPPT is one of the simplest characteristicbased MPPT algorithms, which adopts a fraction of the open circuit voltage of PV panels as the MPP voltage. This is because the ratio between the MPP voltage and the open circuit voltage is approximately consistent under different environmental conditions. In practice, the ratio is generally within 70%e80%, and the exact fraction value can be determined according to the datasheet of the PV panels. The flowchart of the fractional open circuit voltage MPPT is shown in Fig. 15.5. As shown in Fig. 15.5, this method periodically measures the open circuit voltage of PV panels, and adjusts the voltage reference as a fraction of it. Inevitably, during the time interval of the open circuit voltage sampling, no power will be generated by the PV panels. Therefore, there is a tradeoff between the MPPT accuracy and the power loss caused by the open circuit voltage measurement. In practice, the open circuit voltage MPPT is measured every hundreds of milliseconds, and the PV panels should keep open circuited for tens of microseconds to ensure a reliable sampling result. Moreover, there are still inaccuracies of this method in
Initialization * = V0*; Set the fraction value Fv; Set VPV
Stop the operation of the converter
Measure the open circuit voltage (VOC) of PV panels
* = Fv ∙ VOC Set VPV
Enable the voltage control of the converter
Wait until the next MPPT period
FIGURE 15.5 Flowchart of the fractional open circuit voltage maximum power point tracking (MPPT) algorithm, where Fv is the designed fractional value.
Advanced power control of photovoltaic systems Chapter  15
453
tracking the MPP, since the fraction of the open circuit voltage cannot coincide exactly with the MPP voltage. Due to this, there will also be energy losses. However, compared with the P&O algorithm, there will be no oscillation in the steady state, which will bring about better interharmonic performances [18]. Once the open circuit voltage is obtained, the voltage reference can be quickly set as a fraction of the open circuit voltage, thus the dynamic performance is better than the P&O method. Besides, since only the voltage sensing circuit is required for this method, the fractional open circuit MPPT method is more costeffective.
15.3 Requirement of advanced control functionality Gridconnected PV systems are becoming one of the essential renewable energy sources in today’s power grid. With the main purpose of increasing renewable power production, both the number of installation and system capacity of PV plants are expected to be increased even further in the near future [6,11]. In order to cope with this transition, the design, control, and operation of gridconnected PV systems should comply with the currently active grid requirements, defined as the grid codes, which include not only the power quality requirements of gridconnected PV systems but also the advanced control functionalities they should provide [11]. In the following, the main requirements of advanced control functionalities defined in the current grid codes are reviewed, including both general requirements and active power control requirements.
15.3.1 Grid code 15.3.1.1 Requirements under normal grid conditions In general, the gridconnected PV systems should meet certain fundamental demands, such as the power quality demand within the normal power production range. Also, the DC content of the supplied AC current should not exceed 0.5% of the nominal current, and the total harmonic distortion level should be lower than 5%. According to the current active grid regulations, the gridconnected PV systems must be able to withstand frequency and voltage deviations in the point of connection while reducing their active power as little as possible. For instance, when the normal operating voltage is within Uc 10% and the frequency range is 47.00e52.00 Hz, the overall requirements for active power production are as shown in Fig. 15.6 [8]. In the Uc to Uc þ10% voltage range, the active power is limited to the nominal output. In the Umin to Uc voltage range, the active power is limited by the potential nominal current.
454 Control of Power Electronic Converters and Systems Vpoc Vmax Vc+10% Active power = Nominal power
Active power = nominal current × Vc Vc10% Vmin 47
48
49 50 Frequency (Hz)
51
52
FIGURE 15.6 Requirements of active power production in the event of frequency and voltage deviations [8].
15.3.1.2 Requirements under abnormal grid conditions A grid is not always ideal, where the grid voltage and frequency may exceed the prescribed limits, which may negatively affect the surrounding equipment (e.g., influencing equipment lifetime) or even challenge the entire system stability (e.g., during grid disturbances and also low SCR). The gridconnected PV systems thus should succeed in riding through grid disturbances and help restore the stability of the power grid. Due to power line short circuit, starting of large motors, etc., the grid voltage amplitude may drop to a certain level (10%e90% of the nominal voltage or even to zero) for a short time (several milliseconds to several seconds), which is referred to as voltage sags. To maintain the stability of the power grid, firstly, the PV inverters should not be disconnected from the grid in case of a large inrush transient current, which usually occurs at the beginning of a sudden and deep voltage sag [19]. Furthermore, the PV systems should provide their maximum voltage support by injecting a controlled amount of additional reactive current to the grid. This is known as the lowvoltage ridethrough (LVRT) capability. Fig. 15.7 shows an overview of the LVRT requirements in different countries, where the PV systems should remain connected to the grid when the grid voltage is above the curves in Fig. 15.7. In contrast, the disconnection of these power sources during voltage sags may further lead to the instability of the grid. 15.3.2 Active power control requirement With the increasing installation of gridconnected PV plants, the grid may face overloading issues during peak power generation periods (e.g., noon hours
Advanced power control of photovoltaic systems Chapter  15 Germany
100
Voltage level (% of nominal)
455
80 Normal operatoin
Denmark
60 China 40
May be disconnected
20
0 0.15
0.5
1
1.5 Time (s)
2
3
FIGURE 15.7 Lowvoltage ridethrough requirements for photovoltaic system in different countries [8,20,21].
with high solar radiation) [22]. To maintain the stability of the power network, modern grid codes in many countries dictate that these gridtied PV systems should be able to control the active power according to the following remote power constraints: l
l
l
Absolute power constraint (to protect the power grid against overloading during peak power generation periods) Delta power constraint (to establish a regulating power reserve for other ancillary services, e.g., frequency control) Ramprate constraint (to prevent the changes in active power from negatively impacting the grid stability)
Fig. 15.8 depicts an overview of the active power constraint requirements. It should be noted that the active power control functions may comply with the requirements of frequency response/control, as shown in Fig. 15.8D. It is clear that for enabling gridfriendly systems, the gridconnected PV systems should not solely maximize the energy harvesting but also be active in grid regulation with the various integrated advanced control functions.
15.4 Constant power generation control strategy In order to achieve the active power control requirements, the PV system needs to be able to regulate its power production to a certain level during the operation. This control strategy is referred to as the CPG control [22,23], which will be implemented with the PV system in Fig. 15.1B and demonstrated in this section. The main objective of the CPG control strategy is to limit the PV output power to a certain powerlimit level. This requirement can
456 Control of Power Electronic Converters and Systems
(A)
(B) Available power
Active Power
Active Power
Available power
Extracted power
Extracted power
Time
(C)
Time
(D) Start to reduce active power at a certain slope rate
Extracted power
Active Power
Active Power
Available power
50 fR Time
Frequency (Hz)
FIGURE 15.8 Active power control functions for gridconnected photovoltaic systems: (A) absolute power constraint, (B) delta power constraint, (C) ramprate constraint, and (D) frequency response/control ( fR: grid frequency) [8].
be achieved by regulating the operating point of the PV array below the MPP, as it is illustrated in Fig. 15.9. In principle, there are two possible operating pointsdCPPL and CPPR for a certain powerlimit level at a certain solar irradiance and temperature condition. Therefore, the main task of the CPG strategy is to regulate the operating point of the PV arrays at one of these two CPPs. However, the solar irradiance and temperature condition change dynamically during the operation, and thus the PeV characteristic curve. Thus, the CPG strategy also needs to follow the change in the PeV curve and track the CPP under dynamic conditions. In general, the demands of the CPG strategy are as follows: l
l
During steady state (e.g., constant solar irradiance), the CPG strategy should keep the operating point of the PV array at the CPP with minimum deviation, in order to minimize the power loss. During transient (e.g., changing solar irradiance), the CPG strategy should track the CPP and follow the change in the PeV curve. Moreover, it should also ensure a smooth transition, e.g., between MPPT and CPG operation.
Advanced power control of photovoltaic systems Chapter  15
Constant Power Point (CPPL)
Plimit
Constant Power Point (CPPR)
High dPpv/dvpv
PV power Ppv (kW)
Maximum Power Point (MPP)
457
dPpv dvpv Low dPpv/dvpv PV voltage vpv (V)
FIGURE 15.9 Possible operating points of the photovoltaic (PV) system in the powerevoltage curve of the PV arrays during the CPG operation (i.e., CPP) at a certain powerlimit and solar irradiance condition (CPPL and CPPR are the operating points at the left and right side of the MPP, respectively).
15.4.1 Direct power control (PCPG) One possible way to achieve the CPG operation is by directly regulating the PV output power through the closedloop power control. In this approach, the reference PV current obtained from the MPPT algorithm is multiplied with the measured PV voltage in order to determine the reference PV power in the MPPT mode. Then, a saturation block is employed to limit the reference PV power to a certain level according to the powerlimit setpoint. In this way, it can be ensured that the reference PV power will be kept at the powerlimit level once the available power is higher than the powerlimit, and thereby achieving the CPG operation. Afterward, the modified PV power reference is used in the closedloop control, where a PI controller is employed to regulate the PV output power following the control structure in Fig. 15.10. The PV power reference with this control method can be summarized as in the following: PMPPT ; when PMPPT Plimit Ppv ¼ (15.1) Plimit ; when PMPPT > Plimit where PMPPT is the maximum available power (according to the MPPT operation), and Plimit is the powerlimit level during the CPG operation.
vpv ipv
MPPT
iMPPT
Saturation block
PMPPT
P*pv
Plimit
PI
d*
Ppv
FIGURE 15.10 Control diagram of the CPG strategy based on Direct Power Control method (PCPG).
458 Control of Power Electronic Converters and Systems
The performance of CPG operation based on the PCPG method is demonstrated in Fig. 15.11A, where different powerlimit levels are considered during the test. It can be seen from the results that the PV output power can be limited accurately according to the setpoint with an error of only 0.61% during the operation (e.g., when Plimit ¼ 80%). It also achieves a smooth transition between the MPPT and CPG operating modes, as it can be observed from the operating trajectory in the PeV curve of the PV arrays in Fig. 15.11B. In this case, the operating point of the PV arrays is regulated at the right side of the MPP (i.e., CPPR) during the CPG operation.
15.4.2 Currentlimiting control (ICPG) The powerlimiting operation during the CPG mode can also be achieved through the regulation of the PV output current. According to the IeV characteristic of the PV array, there is an operating region where the PV voltage is
PV power Ppv (kW)
(A) Available power 3.0
Ppv (Plimit = 80 %)
2.0
Error = 0.61 %
1.0
Ppv (Plimit = 50 %)
0.0
0
Ppv (Plimit = 20 %) 120 180 Time (seconds)
60
240
300
PV power Ppv (kW)
(B) 3.0
Ideal MPPT
Plimit = 80 % 2.0
Experiments with CPG control
CPG operation
1.0 MPPT operation 0.0 100
200
300 PV voltage vpv (V)
400
FIGURE 15.11 Experimental results of CPG strategy based on the Direct Power Control method (PCPG): (A) Photovoltaic (PV) output power and (B) operating trajectory in powervoltage curve of the PV arrays.
Advanced power control of photovoltaic systems Chapter  15
12 PV current ipv (A)
Irradiance level: 1000 W/m2 2 700 W/m2 500 W/m
CPP: Constant Power Point
10
459
8 6 ilimit
4
CPPR
2 0
Area = Plimit 0
50
100
150 200 PV voltage vpv (V)
250
300
350
FIGURE 15.12 Operational principle of the CPG strategy based on CurrentLimiting Control method (ICPG).
almost constant, which is located at the right side of the MPP, as it is illustrated in Fig. 15.12. Therefore, the regulation of the PV current in this region can effectively regulate the PV output power, which corresponds to the rectangular area under the CPPR shown in Fig. 15.12. In this case, the reference PV current needs to be limited according to 8 > < iMPPT ; when PMPPT Plimit ipv ¼ Plimit (15.2) > : v ; when PMPPT > Plimit pv which can be implemented with a saturation block, as illustrated in Fig. 15.13. The reference PV current is then regulated with a PI controller to determine the duty cycle of the DCeDC converter. According to (15.2), the saturation block will not be activated during the MPPT operation (e.g., PMPPT Ppv). Thus, the modification of the ICPG method will not affect the performance of the MPPT algorithm during normal operation. The performance of the ICPG method is demonstrated in Fig. 15.14A under different powerlimit levels. According to the results, the CPG operation
ipv vpv Plimit
Saturation block
MPPT
iMPPT
ilimit
i*pv
PI
d*
ipv
FIGURE 15.13 Control diagram of the CPG strategy based on CurrentLimiting Control method (ICPG).
460 Control of Power Electronic Converters and Systems
can be achieved according to the required powerlimit level with the ICPG method. It can further be seen from the case when Plimit ¼ 80% that the operating mode transition between the MPPT and CPG is achieved smoothly with no overshoot. As it can be seen from Fig. 15.14B, the operating point of the PV arrays is regulated at the right side of the MPP (i.e., CPPR) during the CPG operation, which is the operating region where the PV voltage is almost constant. In that case, the tracking error during the CPG mode is only 0.36%. However, there is a risk of unstable operation during a decreasing solar irradiance condition, where the operating point of the PV arrays may fall into short circuit condition, as it is shown in Fig. 15.14B. This is mainly due to the steepness of the powerecurrent characteristic of the PV arrays, which may also occur during the MPPT operation [24].
15.4.3 Perturb and observeebased control (P&OCPG) A P&Obased control algorithm, which has been employed in the MPPT operation, can also be implemented for the CPG control strategy. However, in
PV power Ppv (kW)
(A) Available power 3.0
Ppv (Plimit = 80 %)
2.0
Error = 0.36 %
1.0
Ppv (Plimit = 50 %)
0.0
0
Ppv (Plimit = 20 %) 120 180 Time (seconds)
60
240
300
PV power Ppv (kW)
(B) 3.0
Ideal MPPT
Plimit = 80 % 2.0
Risk of Experiments with shortcircuit CPG control condition
CPG operation
1.0 MPPT operation 0.0 100
200
300 PV voltage vpv (V)
400
FIGURE 15.14 Experimental results of CPG strategy based on the CurrentLimiting Control method (ICPG): (A) Photovoltaic (PV) output power and (B) operating trajectory in powervoltage curve of the PV arrays.
Advanced power control of photovoltaic systems Chapter  15
461
this case, the control algorithm should track the CPP in order to achieve the powerlimit constraint, instead of tracking the MPP like in normal MPPT operation. In fact, the operating point of the PV array needs to be continuously perturbed toward one of the CPPs, e.g., Ppv ¼ Plimit, as it is illustrated in Fig. 15.15. After a number of iterations, the operating point of the PV array will reach and oscillate around the corresponding CPP during steady state. In this CPG strategy, it is possible to regulate the PV output power at either the CPPL or CPPR, depending on the perturbation direction. It should also be noted that the operating point at the right side of the MPP, i.e., CPPR, will result in a larger power oscillation during steady state due to the high slope of the PeV curve (i.e., large dPpv/dt). On the other hand, operating the PV array at the CPPL may be restricted in a singlestage PV inverter due to the minimum DClink voltage requirement. The control structure of the P&OCPG method is shown in Fig. 15.16, where the reference PV voltage vpv can be calculated as vMPPT ; when PMPPT Plimit vpv ¼ (15.3) vpv vSTEP ; when PMPPT > Plimit if the operating point is regulated at the CPPL, or vMPPT ; when PMPPT Plimit vpv ¼ vpv þ vSTEP ; when PMPPT > Plimit
(15.4)
if the operating point is regulated at the CPPR. MPP
PV power Ppv
MPPT operation CPG operation Ppv = Plimit
MPP
CPPL
CPPR
MPP Irradiance level: 2 1000 W/m 700 W/m22 500 W/m PV voltage vpv
FIGURE 15.15 Operational principle of the CPG strategy based on the P&O control method.
vpv ipv Plimit
MPPT/CPG
v*pv
PI
d*
vpv
FIGURE 15.16 Operational principle of the CPG strategy based on the P&O control method.
462 Control of Power Electronic Converters and Systems
The performance of the P&OCPG method when the operating point of the PV arrays is regulated at CPPR and CPPL is demonstrated in Figs. 15.17 and 15.18, respectively. According to the results in Fig. 15.17, the operating point at the CPPR results in a larger power oscillations. Consequently, the tracking error during the CPG operation is 1.22%. On the other hand, regulating the operating point at the CPPL during the CPG operation results in a lower power oscillation, where the tracking error is only 0.37%, as shown in Fig. 15.18. Nevertheless, there is a small overshoot in the PV power during the operating mode transition (e.g., from MPPT to CPG) when employing P&OCPG method for both the operating points. This is due to the fact that the P&OCPG method requires a number of iterations until the operating point of the PV array reaches the corresponding CPP. Notably, the performance of the P&OCPG algorithm can be improved by employing an adaptive stepsize during dynamic and steadystate conditions, as it has been demonstrated in Refs. [25,26].
PV power Ppv (kW)
(A) Available power
3.0
Ppv (Plimit = 80 %)
2.0
Error = 1.22 %
1.0
Ppv (Plimit = 50 %)
0.0
0
Ppv (Plimit = 20 %) 120 180 Time (seconds)
60
240
300
PV power Ppv (kW)
(B) 3.0
Ideal MPPT Plimit = 80 % Experiments with 2.0 CPG control
CPG operation
1.0 MPPT operation 0.0 100
200
300 PV voltage vpv (V)
400
FIGURE 15.17 Experimental results of CPG strategy based on the Perturb and Observe method (P&OCPG) operating at the right side of the MPP (i.e., CPPR): (A) Photovoltaic (PV) output power and (B) operating trajectory in powerevoltage curve of the PV arrays.
Advanced power control of photovoltaic systems Chapter  15
PV power Ppv (kW)
(A) 3.0 Available power Error = 0.37 %
1.0
Ppv (Plimit = 50 %)
0
Ppv (Plimit = 20 %) 120 180 Time (seconds)
60
(B) PV power Ppv (kW)
Ppv (Plimit = 80 %)
2.0
0.0
3.0
463
240
300
Ideal MPPT
Plimit = 80 % 2.0
CPG operation
1.0
Experiments with CPG control
MPPT operation 0.0 100
200
300 PV voltage vpv (V)
400
FIGURE 15.18 Experimental results of CPG strategy based on the Perturb and Observe method (P&OCPG) operating at the left side of the MPP (i.e., CPPL): (A) Photovoltaic (PV) output power and (B) operating trajectory in powerevoltage curve of the PV arrays.
15.5 Benchmarking of constant power generation control strategy The performance of different CPG strategies is compared through experimental results. Two different operating conditions during cloudyday and clearday conditions are considered where a powerlimit level of Plimit ¼ 50% is employed, as it is shown in Figs. 15.19 and 15.20, respectively. The performance benchmarking is carried out by considering dynamic responses, steadystate responses, tracking error, stability, and complexity.
15.5.1 Dynamic responses The dynamic responses of the CPG strategies can be evaluated by considering the operation during a cloudy day, as shown in Fig. 15.19. During this operating condition, the solar irradiance and thus available PV power fluctuate with
3.5
(B)ICPG
PCPG
PV power Ppv (kW)
3.0
Tracking error = 0.66 % (During stable operation)
Available power
2.5 2.0 Plimit=1.5 kW
1.5
Plimit=1.5 kW Unstable (Opencircuit condition)
Ppv
1.0 0.5 0
(C) 3.5
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
Unstable (Shortcircuit condition)
Ppv
24
0
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
24
(D)P&OCPG (CPPL)
P&OCPG (CPPR)
3.0
PV power Ppv (kW)
Tracking error = 0.63 % (During stable operation)
Available power
Tracking error = 5.21 % (During stable operation)
Available power
Tracking error = 3.47 % Available power
2.5 2.0 Plimit=1.5 kW
1.5
Plimit=1.5 kW Unstable (Opencircuit condition)
1.0 Ppv
0.5 0
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
Ppv
24
0
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
24
FIGURE 15.19 Experimental results of CPG strategies under a cloudyday condition with powerlimit level of Plimit ¼ 50% based on (A) direct power control, (B) currentlimiting, (C) Perturb and Observe at the CPPR, and (D) perturb and observe at the CPPL.
464 Control of Power Electronic Converters and Systems
(A)
(A)
3.5
(B) ICPG
PCPG
Available power
2.5 2.0
Ppv
Plimit=1.5 kW
1.5 1.0
Unstable (Shortcircuit condition)
Available power Ppv
Plimit=1.5 kW
Tracking error = 0.80 %
Tracking error = 0.68 %
0.5 0
(C)
3.5
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
24
0
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
24
18
20
22
24
(D) P&OCPG (CPPL)
P&OCPG (CPPR)
PV power Ppv (kW)
3.0 Available power
2.5
Available power Ppv
2.0 Plimit=1.5 kW
1.5
Ppv
Plimit=1.5 kW
1.0
Tracking error = 1.59 %
Tracking error = 0.40 %
0.5 0
2
4
6
8
10 12 14 Time (hours)
16
18
20
22
24
0
2
4
6
8
10 12 14 Time (hours)
16
465
FIGURE 15.20 Experimental results of CPG strategies under a clearday condition with the powerlimit level of Plimit ¼ 50% based on (A) the direct power control, (B) currentlimiting, (C) perturb and observe at the CPPR, and (D) perturb and observe at the CPPL.
Advanced power control of photovoltaic systems Chapter  15
PV power Ppv (kW)
3.0
466 Control of Power Electronic Converters and Systems
a fastchanging rate. This challenges the dynamic performance of the CPG strategy, which needs to keep the PV output power constant at the powerlimit regardless of the variation in the available PV power condition. According to the experimental results in Fig. 15.19, it can be observed that both the CPG strategies based on the direct power control (PCPG) and currentlimiting control (ICPG) can achieve a fast dynamic performance where there is no overshoot in the PV output power during the fluctuation of the available PV power. Therefore, the PV output power can be limited to 1.5 kW during the entire operation. In contrast, the P&Obased CPG method presents a large power overshoot during the fluctuation of the available power. This slow dynamic response is due to the fact that the control algorithm requires a number of iterations to regulate the operating point at the CPP. This is applied for both the operating point at the CPPR and CPPL, as shown in Fig. 15.19C and D, respectively. In that case, the powerlimit constraint is violated during the fastchanging of the solar irradiance level.
15.5.2 Steadystate responses The steadystate performance of the CPG strategies can be evaluated during the clearday condition, where the available PV power changes smoothly during the daily operation. According to the experimental results in Fig. 15.20, all of the CPG strategies are capable of limiting the PV output power according to the powerlimit level during the operation with a very small deviation. Only the P&Obased CPG method presents a power oscillation during the operation when the operating point is regulated at the CPPR, as it is shown in Fig. 15.20C. This is mainly due to a large dPpv/dvpv at the CPPR, where even a small perturbation in the PV voltage dvpv can result in a large PV power change dPpv.
15.5.3 Tracking error The tracking error is a quantitative measure of the accuracy of the CPG strategies. It is calculated from the difference between the actual PV output power and the powerlimit level during the operation and then divided by the total energy yield (i.e., PpvdPlimit/Epv). For instance, the dynamic performance of the CPG strategy can be evaluated quantitatively by considering the tracking error during the cloudyday condition. It can be seen from the results in Fig. 15.19C that a large error occurs when the P&Obased CPG method is employed. On the other hand, the tracking error during the clearday condition can reflect the steadystate response of the CPG strategies. In all cases, the tracking error during steady state is minimal. The largest error occurs when the P&Obased CPG method regulates the operating point at the CPPR, which is in an agreement with the power oscillation in the timedomain waveform of the PV output power (see Fig. 15.20C).
Advanced power control of photovoltaic systems Chapter  15
467
15.5.4 Stability It is important for the CPG strategy to ensure a stable operation and continuously deliver power to the grid. However, instability may occur for the CPG strategies during the fast decreasing solar irradiance condition. Due to the rapid change in the powervoltage characteristic of the PV arrays, the operating point during the CPG operation may fall into either open circuit or short circuit conditions. For instance, the short circuit condition may occur when the currentlimiting CPG strategy is employed, as it has been demonstrated in Fig. 15.19B. On the other hand, the operating point of the PV arrays may also fall into an open circuit condition when the direct power control or the P&Obased CPG strategies operating at CPPR are employed during a fast decreasing solar irradiance condition, as it is shown in Fig. 15.19A and C. Therefore, only the P&Obased operating at CPPL is robust against the stability issue during the fastchanging solar irradiance condition.
15.5.5 Complexity The complexity of the control algorithm plays a major role in the real implementation of the CPG strategy. When comparing all the CPG strategies, the currentlimiting CPG strategy offers the simplest control structure where only one saturation block needs to be added to the original MPPT controller. On the other hand, the P&Obased CPG strategy requires a modification at the MPPT algorithm level, while the direct power control CPG strategy needs both the modification of the MPPT algorithm (to be able to provide the reference PV power) and an additional saturation block to limit the PV power reference. The performance benchmarking of the CPG strategies under different aspects are further summarized in Table 15.1, where þ indicates less tracking error, better stability, and less complexity of the CPG strategy.
TABLE 15.1 Benchmarking of the CPG strategies. CPG strategy
Dynamic responses
Steadystate responses
Tracking error
Stability
Complexity
PCPG
þþ
þ
þ
ICPG
þþ
þ
þ
þþ
P&OCPG at CPPR
P&OCPG at CPPL
þþ
þþ
Note: the more þ, the less tracking error, better stability, and less complexity.
468 Control of Power Electronic Converters and Systems
15.6 Summary With an increasing installation of PV systems, a more gridfriendly control strategy is required to address the challenging issues related to the overloading of the grid, grid voltage fluctuation, and frequency regulation capability. This chapter has presented an advanced control of PV systems by means of CPG control. Three different approaches to implement the CPG control strategy have been discussed and their performance has been benchmarked in terms of dynamic responses, steadystate responses, tracking error, stability, and complexity. According to the evaluation results, the direct power and the currentlimiting control strategies can achieve fast dynamic responses, which results in low tracking errors. However, the P&Obased control strategy has a superior steadystate performance and is more robust, e.g., against unstable operation during the fast decreasing solar irradiance conditions. Therefore, it is a suitable method to be implemented in practical applications.
References [1] F. Blaabjerg, Z. Chen, S.B. Kjaer, Power electronics as efficient interface in dispersed power generation systems, IEEE Trans. Power Electron. 19 (5) (September 2004) 1184e1194. [2] T. Stetz, J. von Appen, F. Niedermeyer, G. Scheibner, R. Sikora, M. Braun, “Twilight of the grids: the impact of distributed solar on Germany’s energy transition, IEEE Power Energy Mag. 13 (2) (March 2015) 50e61. [3] A. Woyte, V.V. Thong, R. Belmans, J. Nijs, Voltage fluctuations on distribution level introduced by photovoltaic systems, IEEE Trans. Energy Convers. 21 (1) (March 2006) 202e209. [4] S. Shivashankar, S. Mekhilef, H. Mokhlis, M. Karimi, Mitigating methods of power fluctuation of photovoltaic (PV) sources  a review, Renew. Sustain. Energy Rev. 59 (2016) 1170e1184. [5] B. Kroposki, Can solar save the grid? IEEE Spectr. 53 (11) (November 2016) 42e47. [6] Y. Yang, P. Enjeti, F. Blaabjerg, H. Wang, Widescale adoption of photovoltaic energy: grid code modifications are explored in the distribution grid, IEEE Ind. Appl. Mag. 21 (5) (September 2015) 21e31. [7] Y.K. Wu, J.H. Lin, H.J. Lin, Standards and guidelines for gridconnected photovoltaic generation systems: a review and comparison, IEEE Trans. Ind. Appl. 53 (4) (JulyAugust 2017) 3205e3216. [8] Energinet.dk, Technical regulation 3.2.2 for PV power plants with a power output above 11 kW, Tech. Rep. Doc. (2016), 14/1799739. [9] BDEW, Technische richtlinie erzeugungsanlagen am mittelspannungsnetz e richtlinie fur anschluss und parallelbetrieb von erzeugungsanlagen am mittelspannungsnetz, June 2008. [10] E. Troester, New German grid codes for connecting PV systems to the medium voltage power grid, in: Proc. of 2nd Int. Workshop Concentrating Photovoltaic Power Plants: Opt. Design, Prod., Grid Connection, 2009. [11] Y. Yang, K.A. Kim, F. Blaabjerg, A. Sangwongwanich, Advances in GridConnected Photovoltaic Power Conversion Systems, Publisher: Woodhead Publishing, 2018.
Advanced power control of photovoltaic systems Chapter  15 [12]
[13]
[14] [15]
[16]
[17]
[18]
[19]
[20]
[21] [22]
[23]
[24] [25] [26]
469
F. Blaabjerg, R. Teodorescu, M. Liserre, A.V. Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (October 2006) 1398e1409. M. Saitou, T. Shimizu, Generalized theory of instantaneous active and reactive powers in singlephase circuits based on Hilbert transform, in: Proc. 2002 IEEE 33rd Annu. IEEE Power Electron. Specialists Conf. Proceedings (Cat. No.02CH37289), vol. 3, 2002, pp. 1419e1424. T. Esram, P.L. Chapman, Comparison of photovoltaic array maximum power point tracking techniques, IEEE Trans. Energy Convers. 22 (2) (June 2007) 439e449. N. Femia, G. Petrone, G. Spagnuolo, M. Vitelli, Optimization of perturb and observe maximum power point tracking method, IEEE Trans. Power Electron. 20 (4) (July 2005) 963e973. A. Sangwongwanich, Y. Yang, D. Sera, H. Soltani, F. Blaabjerg, Analysis and modeling of interharmonics from gridconnected photovoltaic systems, IEEE Trans. Power Electron. 33 (10) (October 2018) 8353e8364. D. Sera, R. Teodorescu, J. Hantschel, M. Knoll, Optimized maximum power point tracker for fastchanging environmental conditions, IEEE Trans. Ind. Electron. 55 (7) (July 2008) 2629e2637. A. Sangwongwanich, Y. Yang, D. Sera, F. Blaabjerg, Interharmonics from gridconnected PV systems: mechanism and mitigation, in: Proc. 2017 IEEE 3rd Int. Future Energy Electron. Conf. (IFEECECCE Asia), June 2017, pp. 722e727. Z. Li, R. Zhao, Z. Xin, J.M. Guerrero, M. Savaghebi, P. Li, Inrush transient current analysis and suppression of photovoltaic gridconnected inverters during voltage sag, in: Proc. 2016 IEEE APEC, Long Beach, CA, 2016, pp. 3697e3703. P. Chao, et al., A unified modeling method of photovoltaic generation systems under balanced and unbalanced voltage dips, IEEE Trans. Sustain. Energy 10 (4) (October 2019) 1764e1774. Y. Bak, J. Lee, K. Lee, Lowvoltage ridethrough control strategy for a gridconnected energy storage system, Appl. Sci. 8 (2018) 57. A. Sangwongwanich, Y. Yang, F. Blaabjerg, H. Wang, Benchmarking of constant power generation strategies for singlephase gridconnected photovoltaic systems, IEEE Trans. Ind. Appl. 54 (1) (Febuary 2018) 447e457. H.D. Tafti, et al., Extended functionalities of photovoltaic systems with flexible power point tracking: recent advances, IEEE Trans. Power Electron. 35 (9) (September 2020) 9342e9356, https://doi.org/10.1109/TPEL.2020.2970447. N. Femia, G. Petrone, G. Spagnuolo, M. Vitelli, Power Electronics and Control Techniques for Maximum Energy Harvesting in Photovoltaic Systems, CRC press, 2012. A. Sangwongwanich, Y. Yang, F. Blaabjerg, Highperformance constant power generation in gridconnected PV systems, IEEE Trans. Power Electron. 31 (3) (March 2016) 1822e1825. H.D. Tafti, A. Sangwongwanich, Y. Yang, J. Pou, G. Konstantinou, F. Blaabjerg, An adaptive control scheme for flexible power point tracking in photovoltaic systems, IEEE Trans. Power Electron. 34 (6) (June 2019) 5451e5463.
This page intentionally left blank
Chapter 16
Low voltage ridethrough operation of singlephase PV systems Zhongting Tang1, Yongheng Yang2 1
Central South University, Changsha, China; 2Aalborg University, Aalborg, Denmark
16.1 Introduction As one of the ecofriendly renewable energy sources, the photovoltaic (PV) energy has experienced a very rapid growth rate of installation, which, however, is highly intermittent [1,2]. This inherent intermittent characteristic (i.e., the power generation is dependent on the environmental conditions) causes many challenging issues when the PV power is injected into the utility grid, such as power quality and network stability [3e6]. In order to overcome the above challenges, the gridconnected PV systems are required to have fully and flexibly supportive functions for the grid, when various disturbances (e.g., voltage and frequency faults) appear [7e9]. Especially, with a high penetration of PV power, the abnormal conditions of the utility grid may trigger wide area unintentional disconnection, threatening the equipment security and the entire grid stability (e.g., a largescale outage). There are various abnormal conditions in the grid, as exemplified in Fig. 16.1, including the voltage and frequency disturbances [10,11]. Generally, the events can be separated into longterm disturbances (e.g., harmonic distortion [13], unbalanced threephase voltages, flickering, and frequency variations) and shortterm disturbances (e.g., voltage sags, voltage swells, voltage spikes, and power outage). By comparison, the shortterm events are more likely to cause interruptions to the grid [10]. Therefore, advanced control strategies for gridconnected PV systems should be developed to provide a stable and gridfriendly operation under such abnormal grid conditions (e.g., voltage sags) [7,12]. A voltage sag is a short and fast transient event in gridconnected systems occurring due to various accidents (e.g., lightning strikes, power line short circuit, etc.). The temporary drop of the grid voltage amplitude may exceed the Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000068 Copyright © 2021 Elsevier Ltd. All rights reserved.
471
472 Control of Power Electronic Converters and Systems
FIGURE 16.1 Abnormal conditions in the utility grid voltage (i.e., including the disturbances/ events in amplitude and frequency).
normal voltage threshold, and, subsequently, trigger the islanding protection [14], causing the power supply interruption. When the amount of stoppage of power generation from PV systems reaches a certain level, the stability of the grid may be threatened. This chapter then focuses on the advanced control strategies under a typical transient disturbance (i.e., the voltage sag) in singlephase gridconnected PV systems. Thus, many grid codes and standards require gridconnected PV systems to have the capability to ridethrough the transient voltage sags [15]. Meanwhile, the voltage regulation capability should be provided by the PV generation systems to support the grid voltage recovery during shortterm events [16]. This is known as low voltage ridethrough (LVRT) operation. The response and voltage ridethrough requirements of the distributed energy resources (DERS) in the IEEE Std 15472018 (i.e., revision of IEEE Std 15472003) are shown in Fig. 16.2 [17,18]. As shown in Fig. 16.2, the operation zone is divided into different regions, containing the continuous operation region (i.e., normal operation mode), mandatory voltage ridethrough operation regions (i.e., mandatory connection to the utility grid), momentary cessation regions, trip regions, and undefined regions, where the PV systems may be tripped or may ridethrough. Additionally, the voltage regulation capability demanded in the IEEE Std 15472018 includes the voltagereactive power regulation and voltageactive
Low voltage ridethrough operation Chapter  16
473
FIGURE 16.2 Response and voltage ridethrough requirements for distributed energy resources to grid voltage faults adapted according to the IEEE Std 15472018. When outside the shaded areas, the system shall trip.
power regulation [7,18]. More specifically, in the case of LVRT operation, it is mandatory that the gridconnected PV systems should operate the following mutually exclusive regulation modes of reactive power control: l l l l
Constant power factor mode; Voltagereactive power mode; Active powerreactive power mode; and Constant reactive power mode.
For instance, Fig. 16.3 shows the voltagereactive power characteristic as an example, where the gridconnected PV systems should actively control the reactive power output according to the piecewise linear characteristics. The minimal reactive power injection (RPI) should arrive at 25% of the nameplate apparent power in normal DERs, while 44% is required in the area where the aggregated DERs penetration is high or the overall DERs power output has frequent large variations (e.g., due to the fluctuation of the PV energy resource). In addition to the requirements for the RPI, the voltage should also be regulated by changing the active power, following the voltageactive power characteristic. This requirement aims to limit the maximum active power during the voltage sag to avoid inverter tripping. Fig. 16.4 exemplifies two characteristics for different DERs (i.e., one can only generate active power, while the other can generate and absorb active power).
474 Control of Power Electronic Converters and Systems
FIGURE 16.3 Requirement of the voltagereactive power characteristic in the case of fault ridethrough operation (i.e., low voltage ridethrough (LVRT) operation) according to the IEEE Std 1547.
FIGURE 16.4 Requirement of the voltageactive power characteristic according to the IEEE Std 15472018.
The above operation modes are allowed to be flexibly adjusted by the utility grid operator, such as the power factor and the parameters in the adjustable range of the voltagereactive/active power characteristics [18]. Without special requirements from the utility grid operator, the PV systems should operate in the maximum power point tracking (MPPT) mode at the
Low voltage ridethrough operation Chapter  16
475
unity power factor and remain in the continuous operation region, as shown in Fig. 16.2 [19]. Once a grid voltage fault occurs (i.e., the mandatory ridethrough regions in Fig. 16.2), the gridconnected PV systems are demanded to stay connected to the utility grid [12,20]. In practice, the transient grid voltage faults occur more frequently, especially in remote areas with a weak grid. Thus, advanced control strategies should be developed for the gridconnected PV systems to provide a seamless operation transition between the normal operation (i.e., MPPT at the unity power factor) and LVRT operation (i.e., ridethrough voltage sags with voltage regulation capability). To provide dynamic grid support, the control method should enhance the RPI capability from the PV gridconnected converters [7,16,18]. Therefore, in addition to the voltageactive regulation capability, shown in Fig. 16.4, the active power should also be reduced to prevent overcurrent caused by the excessive apparent power, as discussed previously. In this chapter, solutions to LVRT operations for singlephase PV systems are firstly reviewed. To achieve flexible, gridfriendly, and gridsupporting PV systems, the LVRT operations for both the singlestage and doublestage PV systems are introduced. Since the voltagereactive regulation capability is demanded to provide dynamic supports for the grid voltage, RPI strategies considering the active power regulation for singlephase PV inverters are then presented in this chapter.
16.2 Low voltage ridethrough operations As mentioned previously, the LVRT operation means that the gridconnected PV systems must remain connected and ridethrough the grid voltage sags of a certain level (see Fig. 16.2) [21]. Simultaneously, the utility grid operator can require a certain amount of reactive power to be injected at the point of common coupling [14] from the gridconnected PV systems to support the grid voltage recovery, and by that, providing dynamic grid support capability [7,16,18]. The singlephase inverter is suitable to be applied in the lowpower or mediumpower PV systems, which means it has limited contribution to the utility grid voltage regulation. However, with the increasing penetration of PV systems, the total PV power is becoming much larger to affect the entire system stability [1]. Thus, the advanced and flexible control strategy should be developed for the individual gridconnected PV inverter to avoid the stoppage under the transient voltage sag (due to islanding protection) and to provide voltage regulation for the overcurrent protection and the grid voltage recovery. In general, the basic LVRT operation of the gridconnected PV systems (either singlephase or threephase) can be summarized in three steps: (1) MonitoringdSince the grid voltage amplitude is the basic judgment for the switchover of the normal operation and LVRT operation, the grid voltage online monitoring is very critical for fast detection. The priorart
476 Control of Power Electronic Converters and Systems
research efforts have verified that a phaselocked loop (PLL) system can provide a fast and accurate detection of the grid voltage [22]. Nevertheless, more advanced monitoring techniques, e.g., combined with artificial intelligence, are always of interest. (2) Continuous connection and accurate current reference generationdIt is mandatory to continuously connect to the utility grid in a required period. In addition, the controller should generate accurate current references to satisfy the demands in grid codes (i.e., reactive and active power regulation) and to prevent overcurrent or overvoltage tripping. The continuous connection can easily be implemented when a voltage sag happens. In addition, the major control is switched to the LVRT operation state to generate the current references as required [21]. As mentioned previously, various strategies (e.g., active power and reactive power control) can be implemented when generating the current references. (3) RecoverydWhen the grid voltage sag is cleared, the PV system should provide seamless transitions back to the normal operation (i.e., in the MPPT mode at unity power factor) [23]. The recovery further requires a robust and fast synchronization technique to avoid any potential large transient disturbances and to protect the equipment. Without physical inertia (i.e., the inherent characteristics in large wind power systems and conventional power generation systems with rotating machines) [10], PV systems have fast dynamics coming from the power electronic control. However, as increasing RPI under the LVRT operation is required, the active power should be reduced correspondingly to prevent the overcurrent tripping. When compared to the singlestage PV converter, the doublestage PV converter (i.e., including a DCeDC stage and an inverter stage) accepts a wider voltage range of the PV panels, enhanced by the frontend DCeDC converter [24]. Additionally, the doublestage structure can provide more flexible control due to the separated control for the gridconnected PV systems, where the DCeDC stage mainly aims to achieve the MPPT of the PV panels, and the inverter stage satisfies the requirements from the utility grid. Correspondingly, the solutions to reduce the active power for doublestage inverter are easier than those for the singlestage inverter. Possible methods to active power reduction for the singlephase PV systems can be summarized as follows: n
n
Modifying the MPPT control: The PV systems regulate the power tracking according to the powerevoltage (PV) characteristics of PV panels to reduce the active power (see also Chapter 14). Integrating storage systems: The PV systems with a storage device can store the excessive active power during the LVRT operation, and then provide the power to the local equipment in the islanding operation [7].
Low voltage ridethrough operation Chapter  16
n
477
Adopting a DC chopper: The excessive active power can be dissipated in the chopper resistor (i.e., the same as the braking resistor in wind turbine systems with rotating machines).
Compared to the above methods, modifying the MPPT is the most promising one due to its low system cost (i.e., no extra hardware). However, the conversion efficiency of the PV panel may be reduced. The second method can allow the MPPT control and ensure the conversion efficiency under the LVRT. Although being an advanced and promising technology in smart PV systems, the high price of the storage may lead to a relatively low costefficient of the PV systems. Nevertheless, along with the further development of storage technologies, the second solution will become much attractive and flexible. Regarding the conventional solution, i.e., employing a DC chopper, the MPPT control also can be continuous. Yet, the extra active power is only wasted as heat on the chopper resistor. Considering the above, the advanced control methods of the LVRT operation for singlephase PV systems will be implemented through modifying the MPPT control, which will be introduced in the following.
16.2.1 LVRT control using the singlephase PQ theory The RPI and active power regulation are critical tasks for the gridconnected PV systems during the LVRT operation. It seems that the droop control is a suitable choice to implement the reactive/active power regulation [25]. The relationship of the conventional droopcontrol method includes the voltagereactive power and frequencyactive power, where the system should be mainly inductive (i.e., a high X/R ratio) [21]. Seen from this aspect, the droop control is not a very feasible method for the LVRT operation of singlephase PV systems, which are generally connected to lowvoltage feeders (i.e., being resistive lines with a low X/R ratio). Consequently, to have a flexible control of the reactive/active power, one way is the control strategy based on the singlephase PQ theory to directly synthesize the power references under the LVRT operation for singlephase PV systems [26,27]. With additional adaptive filters, the LVRT control based on the singlephase PQ theory facilitates to regulate the RPI [21,28,29]. In addition, the control schemes can be implemented in either the stationary or the rotating reference frame [4]. Considering the entire control complexity and harmonic compensation capability, the control strategy implemented in the stationary reference frame (i.e., ab) is more advantageous. Fig. 16.5 shows the entire control system for the singlestage singlephase PV system, and the control is based on singlephase PQ theory. Since there is only a fullbridge (DCeAC) inverter to control, the PV input voltage (DClink voltage) vpv is regulated through the MPPT control (gives the active power reference, P ), and it must be higher than the peak value of the grid voltage. As
478 Control of Power Electronic Converters and Systems
FIGURE 16.5 Control structure of the singlestage singlephase gridconnected photovoltaic (PV) system: (A) hardware schematic and (B) multimode control structure based on the singlephase PQ theory, where F is the fault signal and q is the grid phase. Here, the control is implemented in the ab frame with vga ¼ vg being the grid voltage, vgb being the orthogonal inquadrature voltage, and vgm being the grid voltage amplitude, Gc(s) is the current controller, point of common coupling (PCC) is the point of commoncoupling, Zg is the grid impedance, and Cdc is the DClink capacitor.
shown in Fig. 16.5A, a synchronization and monitoring unit is adopted for the synchronization and the voltage information detection (e.g., the instantaneous grid voltage level). When there is a voltage sag fault, the signal F will then change the operation mode to the LVRT mode. Furthermore, for gridconnected system, the current controller should ensure the injection of highquality currents that are synchronized with the grid voltage phase [4,19]. Considering the control structure in Fig. 16.5B, the fundamentalfrequency current controller can be a periodic controller (e.g., a proportional resonant (PR) controller, a repetitive controller, or a deadbeat controller [30e33]) instead of the conventional proportional integral (PI) controller. Additionally, harmonic compensation based on parallel resonant controllers and repetitive controllers can be employed to further improve the current quality. In normal operation mode, the PV system should operate in the MPPT mode at unity power factor, thus ensuring the maximization of the power production from PV panels. According to Fig. 16.5A, the active power reference is determined by the MPPT control, i.e., P ¼ PMPP with PMPP representing the maximum power of the PV panels, while the reactive power reference is 0 var (Q ¼ 0 var) for the unity power factor. Once the grid voltage level is in the Mandatory Operation zone shown in Fig. 16.2, the power
Low voltage ridethrough operation Chapter  16
479
reference should be reconstructed as a function of the voltage amplitude following the required linear voltagereactive/active power capability. For instance, according to the requirements of the IEEE Std 15472018 in Fig. 16.3 [18], the reactive power that should be injected during grid faults is given as 8 >
: PN
0:9 p.u. vgm < 1:1 p.u. VL p.u. vgm < 0:9 p.u.
(16.1)
vgm < VL p.u.
where VL is the lower limit of the voltage for continuous operation, vgm is the grid voltage level, PN is the nominal grid power, Q is the required reactive power, and k is an adjustment factor that is given as k¼
Q=PN 2 p.u. 1 vgm
(16.2)
According to the singlephase PQ theory, the active power P and reactive power Q can be calculated in the abreference frame as 8 1 > > < P ¼ ðva ia þ vb ib Þ 2 (16.3) > > : Q ¼ 1 ðvb ia va ib Þ 2 in which the subscript a and b represents the corresponding a and b component of the grid voltage vg and grid current ig. Then, the currents in the abreference frame can be expressed as
ia ib
va 2 ¼ 2 va þ v2b vb
vb va
P Q
(16.4)
with v2a þ v2b ¼ v2gm. As the active and reactive power should be adjusted in the LVRT mode, it is intuitive to generate the current references according to Eq. (16.4) as GPIP ðsÞðP PÞ 2 ½ va vb ig ¼ ia ¼ 2 (16.5) va þ v2b GPIQ ðsÞðQ QÞ with GPIP(s) and GPIQ(s) being the PI controller for the active power and reactive power, respectively. Clearly, the current reference is an AC signal, and thus, the current controllers working in the abreference frame should be adopted, as discussed previously. It should be noted that the current reference
480 Control of Power Electronic Converters and Systems
can be obtained directly from the power references (i.e., the power is not directly controlled, and the PI controllers are set as unity gains), according to Eq. (16.5), as P 2 v v ig ¼ 2 ½ (16.6) a b va þ v2b Q which simplifies the control design. In order to validate the effectiveness of the abovediscussed control strategies, simulations have been carried out on a singlestage singlephase PV system and the control in Eq. (16.6) was adopted. The system parameters are listed in Table 16.1 and the MPPT sampling frequency is 200 Hz. In the simulations, the rated power was set as PN ¼ 1.03 kW. The current controller is a PR controller, where the proportional and resonant control gains (kp and kr) are set as 22 and 3000. In addition, the proportional and integral gains of GPIP(s) and GPIQ(s) are 6.2 and 1.5 for active power, respectively, and 1 and 50 for the reactive power. A secondorder generalized integrator PLL is adopted for synchronization and fault detection [21], i.e., the gridmonitoring system. A perturb and observe (P&O) MPPT algorithm (see Chapter 14) has been applied in this case study. The voltage lower limited VL is set to 0.5 p.u. in the simulations. The simulation results are shown in Figs. 16.6 and 16.7. Fig. 16.6 shows that the grid has a voltage sag of 0.35 p.u. In the beginning, the singlestage singlephase PV system runs in the normal operation (i.e., MPPT control in unity power). Then, at around 0.05 s, the grid voltage drops to 0.65 p.u., and then the system enters the LVRT operation mode. This voltage sag lasts for about 320 ms. It is noted that the entire generating unit should
TABLE 16.1 System and control parameters of a singlephase photovoltaic inverter with an LCL filter. Parameter
Symbol
Value
DClink voltage (at the maximum power point)
Vdc
400 V
Grid voltage amplitude
vgm
325 V
Grid nominal frequency
f0
50 Hz
LCL filter
L1
3.6 mH
Cf
2.35 mF
L2
4 mH
Sampling frequency
fs
10 kHz
Switching frequency
fsw
10 kHz
Low voltage ridethrough operation Chapter  16
481
FIGURE 16.6 Simulation results of a singlestage singlephase photovoltaic system under a grid voltage sag with the low voltage ridethrough (LVRT) control based on the singlephase PQ theory: (A) grid voltage vg and grid current ig, (B) injected active power P and reactive power Q.
FIGURE 16.7 Output power from the photovoltaic (PV) panels of the singlestage singlephase system under grid faults with the low voltage ridethrough (LVRT) control based on the singlephase PQ theory.
482 Control of Power Electronic Converters and Systems
cease energizing if the shortterm voltage sag lasts too long (i.e., requiring tripping if the grid voltage sags to 0.65 p.u. for more than 21 s, as shown in Fig. 16.2). Additionally, it will take a few milliseconds for the PLL (i.e., for the grid voltage detection) to respond to sudden changes of the grid voltage. During this period, the PV system stays in the MPPT operation at unity power factor. The grid current thus has a transient rise, as it is shown in Fig. 16.6A. Due to the same reason, the grid current has a temporary drop when the grid voltage recovers, which can also be seen in Fig. 16.6A. Once the grid voltage is detected, the control scheme switches to the LVRT operation from the MPPT, where the active power is reduced and reactive power increases according to the requirement in Eq. (16.1), as shown in Fig. 16.6B. It should be pointed that, on one hand, the active power regulation capability may be determined by the utility grid operator. On the other hand, the active power reduction should be enabled to prevent overcurrent tripping. This is mainly because the grid code demands the PV system provide RPI for dynamic grid support during the LVRT operation. However, the increasing of the reactive power may trip the overcurrent protection if the active power is the same as operating in MPPT mode. After the grid voltage sag is cleared (i.e., the voltage level back to 0.9 p.u.), the PV system recovers its normal operation to track the maximum output power of the PV panels again and only generating active power to the utility grid, as depicted in Fig. 16.7. As mentioned previously, the DClink voltage should be larger than the peak value of the grid voltage due to the simple hardware structure of the singlestage singlephase inverter. Therefore, when switching to the LVRT operation mode from the normal MPPT operation, the system can only operate in the right region of the maximum power point (MPP), i.e., a higher voltage, to reduce the active power output from the PV panels. There is no doubt that the switching losses will be increased due to the increased DClink voltage. Nevertheless, the above results have illustrated that the advanced control method based on the singlephase PQ theory can achieve the LVRT operation of singlestage singlephase gridconnected PV inverters, including the reactive/active power regulation.
16.2.2 LVRT control based on the powervoltage curve To have a wider input voltage range for PV panels, a conventional DCeDC converter is normally inserted between the PV panels and the inverter to form a doublestage PV converter, as shown in Fig. 16.8. Additionally, the doublestage structure can provide a flexible control strategy due to the separated control. More specifically, the DCeDC stage mainly aims to achieve the MPPT of the PV panels, and the inverter satisfies the requirements from the utility grid by controlling the DClink voltage and the grid current. In the
Low voltage ridethrough operation Chapter  16
483
FIGURE 16.8 Control structure of the doublestage singlephase gridconnected photovoltaic (PV) system with the low voltage ridethrough capability, where ginv is the gate signals for the DCeAC (fullbridge) inverter, and gb is the gate signal for the DCeDC converter. Here, the fault is continuously monitored, F is the monitored grid voltage amplitude that is fed to the control unit for the DCeDC converter, point of common coupling is the point of commoncoupling, Zg is the grid impedance, and Cdc is the DClink capacitor.
case of grid voltage faults, an advanced control strategy is developed by directly linking the active power from the PV panels with the voltage grid sag level [34]. According to Fig. 16.8, the advanced control strategy can be developed, as shown in Fig. 16.9, which should be able to provide seamless transitions between the normal operation and LVRT operation. In this control scheme, one loop is the conventional control system separated into the MPPT control of the DCeDC stage and a singlephase PQ control of the inverter, and the other is a proportional LVRT controller plugged into the DCeDC controller. Since the plugin controller is automatically enabled during the LVRT (the fault signal F is fed back), it becomes a promising method for the seamless transition control for the doublestage singlephase PV system. Moreover, as shown in Fig. 16.9, the inverter employs a dualloop control structure. In details, the outer loop is a voltage controller based on a PI controller to control the DClink voltage, while a PR controller with a parallel repetitive controller forms the internal current controller to compensate for harmonics of the injected current. As mentioned above, the novel control scheme links the active power with the voltage sag. To further explain this control strategy, the linear droop curve (i.e., the relationship between the PV voltage vpv and the PV power Ppv) of the PV panels can be used. Fig. 16.10 shows the powerevoltage characteristics of the PV panels, including the MPP and the LVRT points A and B (i.e., Point A, below the voltage at the MPP, is in the low voltage region, where dPpv/dvpv is small; while at B, dPpv/dvpv is large). As mentioned, the singlestage inverter can only operate at B, where, however, the increasing PV voltage vpv leads to higher switching losses. For the doublestage inverter, the PV system can be controlled to operate at both points in theory, but the steadystate performance
484 Control of Power Electronic Converters and Systems
FIGURE 16.9 Detailed control structure of the singlephase gridconnected photovoltaic system, where a simple controller is plugged in the control loop for the DCeDC converter to enable the low voltage ridethrough (LVRT) operation and PI is a proportional integral controller. Here, k is the control gain for the plugin controller, db is the duty cycle for the boost converter (to generate the gate signal gb), vinv is the reference voltage (to generate the gate signals ginv) for the inverter, and v0gm is the initial grid voltage amplitude (prefault value).
FIGURE 16.10 Powerevoltage linear characteristics of photovoltaic (PV) panels for doublestage gridconnected PV systems. The linear droop relationship can be utilized to implement the low voltage ridethrough (LVRT), i.e., automatic active power reduction during the LVRT operation. Here, vpv and Ppv are the PV voltage and power, respectively, with the superscript “m” denoting the corresponding variable at the maximum power point.
Low voltage ridethrough operation Chapter  16
485
is different, where the power variations of point A are smaller than point B. Additionally, the PV voltage of point B may be larger than the designed DClink voltage, where the boost converter becomes inactive, and in turn, the entire system acts as a singlestage inverter. With the above concerns, the doublestage singlephase PV system normally chooses to operate at point A during the LVRT. According to Fig. 16.10, the PV voltage at point A can be approximated as m vpv zvm (16.7) pv þ kpv Ppv Ppv where the superscript “m” represents the variable at the MPP, and kpv is the droop coefficient. Obviously, the droop coefficient is affected by the operating conditions. To estimate the information of kpv, only the voltage and power at the MPP are needed in the normal operation (i.e., the DCeDC stage operates at the MPPT under certain conditions, e.g., 25 C ambient temperature and 1kW/m2 solar irradiance level, and the inverter performs at unity power factor). Furthermore, referring to the previous discussions, for singlephase PV systems, the feeders are typically resistive with small X/R ratios. In such a case, the grid voltage frequency is affected by the reactive power, while the grid voltage level can be regulated through the active power control based on a droop relationship as vgm ¼ v0gm kd Pg P0g (16.8) with vgm being the amplitude of the grid voltage, Pg being the active power injected into the grid, kd is the active power derating factor (determined by the grid code or the inverter rating), and the superscript “0” indicating the corresponding initial value. Neglecting the power losses of the PV inverter, Pg is approximately equal to Ppv, which is expressed as Ppv zPg
(16.9)
Therefore, an approximate linear relationship can be found to enable an automatic active power reduction under the grid voltage sag. Combining Eqs. (16.7)e(16.9) gives the linear relationship between the grid voltage amplitude changes and the PV voltage changes as kd vgm v0gm z vpv vm (16.10) pv kpv which implies that the grid voltage amplitude has no changes during the normal operation of the PV system, i.e., vgm ¼ v0gm. Therefore, the DCeDC stage should operate at the MPP, i.e., the PV voltage is v0pv. When the grid voltage sag happens, the PV voltage will be regulated as the grid voltage changes, i.e., vgmev0gm, according to the linear relationship in Eq. (16.10).
486 Control of Power Electronic Converters and Systems
With the above, seamless transitions can be achieved by simply setting the reference voltage for the PV panels as 0 vpv ¼ vm (16.11) pv k vgm vgm where k ¼ kpv/kd is a linear coefficient. The proportional controller (i.e., the proportional gain is k) with the linear relationship in Eq. (16.11) can be directly plugged into the control of the DCeDC converter of the doublestage singlephase system. Since the PV system is controlled to operate at A, it can be derived that the control gain k is a positive value. According to Fig. 16.9, the grid fault will increase the PV voltage reference vpv , corresponding to an increase of the duty cycle db. As a result, the PV voltage vpv will be moved to the left side of the MPP in practice during the LVRT operation. Compared to the conventional control method for the doublestage singlephase PV system, only one parameter k is added. Since k ¼ kpv/kd, it is affected by the PV panel characteristics, the operation conditions (i.e., solar irradiance and ambient temperature), and also the grid impedance characteristics. Specifically, the powervoltage droop coefficient kpv of the PV panels can be expressed as kpv ¼
vm pv Pm pv
(16.12)
in which vmpv and Pmpv are the PV voltage and PV power when the PV system is operating at the MPPT mode under a uniform solar irradiance profile. It can be seen from Fig. 16.10 that the relationship of the droop curve is approximately linear. Thus, the gain of the droop curve kpv in Eq. (16.12) can be refined by using an accurate PV panel model or based on lookup tables. However, the design of the droop control coefficient kd at the grid side is related to the entire PV system (i.e., requiring a detailed analysis using smallsignal modeling). It should be designed from the aspect of stability analysis of the entire system. Nevertheless, substituting Eq. (16.12) into Eq. (16.11) leads to vm pv vpv ¼ vm vgm v0gm (16.13) pv m Ppv kd which is implemented in Fig. 16.9. Simulations are carried out on a doublestage singlephase gridconnected PV system (i.e., a DCeDC boost converter with a singlephase fullbridge inverter referring to Fig. 16.8) to validate the automatic control strategy. The parameters of the system are listed in Table 16.2 and the MPPT sampling frequency is 200 Hz. The control strategy shown in Fig. 16.9 has been implemented. There are three PV strings with 15 panels in series for each, and the panel parameters are also shown in Table 16.2. Accordingly, the maximum
Low voltage ridethrough operation Chapter  16
487
TABLE 16.2 System parameters of a doublestage singlephase photovoltaic (PV) inverter with an LCL filter. Parameter
Symbol
Value
DClink voltage
vdc
450 V
Boost converter inductor
L
2 mH
DClink capacitor
Cdc
2200 mF
Grid voltage amplitude
vgm
325 V
Grid nominal frequency
f
50 Hz
LCL filter
L1
4.76 mH
Cf
4.28 mF
L2
4 mH
Sampling frequency
fs
8 kHz
Fullbridge inverter switching frequency
fsw
8 kHz
Boost converter switching frequency
fb
16 kHz
Rated maximum power
Pmpp
65 W
Voltage at the maximum power
vmpp
17.6 V
Current at the maximum power
impp
3.69 A
Opencircuit voltage
vOC
21.7 V
Shortcircuit current
iSC
3.99 A
PV panels (3 strings and 15 panels per string)
power of the PV array under the standard test condition (i.e., 1kW/m2 solar irradiance level, 25 C ambient temperature) is 2.9 kW. As mentioned previously, the dualloop controller employs a PI controller for the DClink voltage control, a PR controller with a repetitive controller for the grid current control. Similar to the singlestage case, a secondorder generalized integratorbased PLL has been utilized to create the orthogonal voltage, and thus enabling the current control in the abreference frame. The P&O MPPT algorithm is employed in the DCeDC converter stage to track the PV power. All the control parameters are shown in Table 16.3. Fig. 16.11 shows the simulation results of the doublestage singlephase PV system in the same case of the LVRT operation. The performance of the active power regulation is presented in Fig. 16.11A, where the PV power is automatically reduced with the increase of the duty cycle db. In this simulation, the active power regulation aims to prevent inverters from exceeding the rated
488 Control of Power Electronic Converters and Systems
TABLE 16.3 Control parameters for simulations referring to Fig. 16.9. Parameter
Symbol
Value
DClink PI controller
kp
60
ki
250
kpr
20
kir
4500
Repetitive control gain
krc
6.5
Maximum power point tracking control gain
km
0.00167
Photovoltaic droop coefficient
kpv
0.09
Active power droop gain
kd
0.0317
Proportional resonant controller
power, which is done automatically according to the scheme in Fig. 16.9. The dynamics of the control system are also fast, as observed in Fig. 16.11. In addition, it can be illustrated in Fig. 16.11B that the DClink voltage can be maintained as a relatively stable value. During the LVRT operation, the reactive power is also injected from the PV system to the grid, as it is illustrated in Fig. 16.11A. As mentioned previously, the singlephase PV system has normally a low X/R ratio. Therefore, the injection of the reactive power has a minor contribution to the grid voltage recovery. Nevertheless, as it is demonstrated in Fig. 16.11, the plugin control strategy presented in Fig. 16.9 can effectively enable the RPI by directly setting the reactive power reference. In all, the above simulation results verify the effectiveness of the simple LVRT scheme for doublestage singlephase PV systems in terms of dynamics and control. To better understand the operation principle, Fig. 16.12 shows the trajectories of the PV panels under the grid fault (i.e., including the LVRT and recovery). It can be seen in Fig. 16.12 that with the LVRT scheme, the PV system can seamlessly transit between the MPPT mode and LVRT operation with fast dynamics. Moreover, the LVRT scheme regulates the active power automatically by monitoring the grid voltage amplitude instead of calculating the grid active power. However, when the PV inverter is controlled by a droop controller, the calculation may be inevitable. Notably, the active power droop coefficient employed in this case is not optimal, and it is related to the inverter system characteristics.
16.3 Reactive power injection strategies under LVRT As the penetration of PV energy is still increasing, many grid codes have updated the requirements on the RPI under grid faults [36]. As the critical
Low voltage ridethrough operation Chapter  16
489
FIGURE 16.11 Simulation results of a 3kW singlephase doublestage photovoltaic (PV) system under grid faults with the low voltage ridethrough (LVRT) control considering the PV panel inherent powervoltage characteristics: (A) grid voltage vg, grid current ig, and PV output power Ppv; (B) duty cycle db and DClink voltage vdc.
interface of the PV energy and the utility grid, the PV inverter can implement the RPI, which can contribute significantly for smart PV systems to improve system stability and reliability, and thus reduced cost. Considering the dynamic grid support requirement (i.e., the demands from the utility grid operator) and the maximum apparent power of PV inverters (i.e., as exemplified in Fig. 16.13), three RPI strategies [35,36] for singlephase systems are introduced in this section, i.e., the constant average active power control strategy (Const.P), constant active current control strategy (Const.Id), and constant peak current control (Const. Imax).
490 Control of Power Electronic Converters and Systems
FIGURE 16.12 Operation trajectories of photovoltaic (PV) panels of the 3kW singlephase system under the low voltage ridethrough operation based on the inherent powervoltage characteristics with the dynamic waveforms shown in Fig. 16.11.
FIGURE 16.13 PQ diagram of a singlephase photovoltaic (PV) inverter, where Qmax indicates the maximum reactive power exchange capability of the inverter with the maximum power point tracking control, Ppv is the PV inverter active power (can be below the MPP power), and Vgn and Imax are the grid voltage nominal amplitude and the rated current, respectively.
Low voltage ridethrough operation Chapter  16
491
16.3.1 Constant average active power control strategy (Const.P) The objective of the Const.P strategy is to maximize the power production of the PV panels even during the LVRT operation. In this case, the active power is maintained constant either by the MPPT control or an active power reduction scheme in the shortterm lowvoltage period. As a consequence, the amplitude of the injected grid current will inevitably increase. Considering the reactive power requirements (i.e., shown in Fig. 16.13) and the singlephase PQ theory, the currents in the dqsynchronous rotating reference frame can be expressed as 8 > < i d ¼ kd IN vgm (16.14) > : iq ¼ kð1 vgm ÞIN where kd ¼ Ppv/PN is the power derating factor (determined by the inverter rating) with Ppv being the injected active power and PN being the nominal power of the system and qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 1 1 (16.15) p.u. vgm 0:9 p.u. and Igm ¼ i2d þ i2q k with Igm being the amplitude of the grid current and k being an adjustable factor (k 2 p.u.). When the grid voltage amplitude is lower than (11/k) p.u., the system should inject a 100% current as demanded. During this period, the PV active power may still be injected to ensure the output power of the PV panels. However, if a large amount of active and reactive power is injected into the grid at the same time, the operation may trigger the overcurrent protection of the PV inverter, and subsequently, leading to a failure of the fault ridethrough operation. In order to tackle this issue (i.e., avoid the failure of the LVRT), the following constraints should be satisfied: 1 (1) When the voltage sag occurs at the condition as 1 k p.u. vgm 0:9 p.u., the constraint can be expressed as rﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2 I 1 max k2d þ k2 vgm v2gm (16.16) vgm IN (2) In addition, if the grid voltage drops to vgm 1 1k p.u., a full RPI may be mandatory (depending on the grid codes). Then, the limitation for the active power derating factor can be expressed in an inequality as 1 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ Imax (16.17) k2d þ v2gm vgm IN where Imax is the maximum inverter current limitation.
492 Control of Power Electronic Converters and Systems
Eqs. (16.16) and (16.17) illustrate the LVRT capability of the PV systems, which can be one of the design criteria for the component design of gridconnected PV inverters. Obviously, a larger design margin means a better LVRT capability (i.e., withstanding higher currents). On the other hand, this may lead to the selection of highrating devices (thus, increased cost). Seen from another perspective, the LVRT capability can be enhanced by adjusting the active power derating factor kd at the predesigned PV inverter. Fig. 16.14 shows the design considerations for a PV inverter with the LVRT capability and k ¼ 2 p.u. Clearly, the derating factor kd can affect the operation range of the PV inverter under LVRT operation. For instance, if the allowable maximum current of a PV inverter is Imax ¼ 1.5 IN in this case, the PV inverter has to reduce the active power when the voltage is below 0.72 p.u._ Otherwise, the inverter will be shut down due to overcurrent.
16.3.2 Constant active current control strategy (Const.Id) Another alternative RPI strategy is called Const.Id strategy, which aims to maintain a constant active current in the LVRT operation. The Const.Id strategy can automatically reduce active power output in response to voltage sags as id ¼
2P ¼ mIN ¼ Const: vgm
(16.18)
where m is a scaling factor and 0 m 1. Similarly, in the case of voltage sags, the injection of active currents according to Eq. (16.18) may also lead to
FIGURE 16.14 Photovoltaic inverter fault ridethrough capability with the Const.P strategy considering the reactive power injection requirement (k ¼ 2 p.u.).
Low voltage ridethrough operation Chapter  16
493
overcurrent. To avoid so, the following conditions should be kept when the control strategy is employed under the voltage sag. (1) If the monitored instantaneous grid voltage amplitude (e.g., through a 1 PLL) vgm is within 1 k p.u. vgm 0:9 p.u., the controller should satisfy
qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ Imax m2 þ k2 ð1 vgm Þ2 IN
(2) Otherwise, vgm sags severely to the range vgm PV inverter should meet the constraint as pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ Imax m2 þ 1 IN
(16.19) 1 1 k p.u., and the
(16.20)
Thus, the LVRT performance of the PV inverter is dependent on the scaling parameter m and the RPI demand, i.e., the gain k. The design consideration for the Const.Id strategy is presented in Fig. 16.15. Compared with the Const.P strategy, the PV inverter with the Const.Id scheme can be designed with a lower Imax/IN. That means when targeting the same LVRT capability, the PV inverter with the Const.Id strategy can select lower rated power devices, leading to a lower cost. Furthermore, in order to ensure the safety of the inverter in the LVRT operation, a lower m can be considered, i.e., the derating operation.
FIGURE 16.15 Photovoltaic inverter fault ridethrough capability with the Const.Id strategy considering the reactive power injection requirement (k ¼ 2 p.u.).
494 Control of Power Electronic Converters and Systems
16.3.3 Constant peak current control (Const.Igmax) It is known from the above discussions that the LVRT capability of the PV inverter with the two RPI strategies may be affected by the design of the control gains (kd and m) (i.e., the overcurrent protection may be easily enabled due to the improper gains). Therefore, to strictly avoid the inverter overcurrent tripping, the PV inverter can maintain a constant peak value of the injected current during LVRT operation. In this case, the peak of the grid current is given as Igmax ¼ nIN ¼ Const: Imax
(16.21)
where n is introduced to assist the design. Considering the RPI requirement as exemplified in Eq. (16.1), the grid current in the dqreference frame can be generated directly. Similarly, when the Const.Igmax strategy is adopted, two conditions should be considered: 1 (1) When the voltage sag is within 1 k p.u. vgm 0:9 p.u., the reference grid current can be derived as qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ ( id ¼ IN n2 k2 ð1 vgm Þ2 iq ¼ IN $kð1 vgm ÞIN (2) When the grid voltage level is lower and falls in vgm
(16.22) 1 1 k p.u., the
grid currents in the dqreference frame will become pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ id ¼ IN n2 1 iq ¼ IN
(16.23)
It should be pointed out that n has a maximum value of Imax/IN. With this constraint, the inverter has no risk of overcurrent tripping during the LVRT operation. However, in the design and operation phases of the PV inverters, the above constraints should be considered. A comparison of the above three RPI strategies for PV inverters is depicted in Fig. 16.16, where the inverter current limitation is Imax ¼ 1.5 IN and k ¼ 2 p.u._ Observations in Fig. 16.16 indicate that the parameters (kd, m, and n) affect the operation ranges of the PV inverters under the LVRT operation. For the Const.Igmax control strategy, the PV inverter will not be tripped off over a wide range of voltage sags. A similar capability for the PV inverters is also enabled when the Const.Id control strategy is adopted, as indicated in Figs. 16.15 and 16.16. However, the voltage range that the PV inverter can withstand during the LVRT is significantly
Low voltage ridethrough operation Chapter  16
495
FIGURE 16.16 Comparison of the three reactive power injection strategies for singlephase photovoltaic inverters (k ¼ 2 p.u.), where Imax ¼ 1.5IN.
limited when the Const.P control strategy is adopted if the derating operation is disabled. Nevertheless, with the above reactive control strategies, both active power and reactive power (following the demands) can be regulated under the LVRT operation.
16.4 Summary This chapter has reviewed the grid code (i.e., IEEE 1547) requirements under an abnormal grid conditiondgrid voltage sags that may affect the stability of the entire utility grid when the penetration level of the renewable energy system is high. Then, control strategies for the seamless transition under the voltage sag have been discussed, and the strategies are designed for singlephase gridconnected PV systems. Furthermore, for singlestage singlephase PV systems, an advanced control strategy which directly generates the reference reactive/active power has been implemented based on the singlephase PQ theory. Also, another advanced power control strategy has been developed by regulating reactive/active power automatically in accordance to the grid voltage sag level for doublestage singlephase PV systems. Both advanced control strategies can achieve seamless transitions between the LVRT operation and the normal operation of singlephase PV systems, but they have different dynamic performances. The advanced LVRT strategies have been validated through simulations. Furthermore, this chapter has also introduced various RPI strategies, as a dynamic grid support ability for singlephase gridconnected PV systems, which is required under grid voltage faults. The audience is recommended to validate the reactive power injection strategies as exercises to better understand them.
496 Control of Power Electronic Converters and Systems
References [1] F. Blaabjerg, Y. Yang, D. Yang, X. Wang, Distributed powergeneration systems and protection, Proc. IEEE 105 (7) (July, 2017) 1311e1331. [2] F. Blaabjerg, D.M. Ionel, Y. Yang, H. Wang, Renewable energy systems technology overview and perspectives, in: Renewable Energy Devices and Systems with Simulations in MATLABÒ and ANSYSÒ, CRC Press, 2017, pp. 1e16. [3] V. Knazkins, Stability of Power Systems with Large Amounts of Distributed Generation, KTH, Stockholm, Sweden, 2004. Ph.D. Dissertation. [4] F. Blaabjerg, R. Teodorescu, M. Liserre, A.V. Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (October, 2006) 1398e1409. [5] H. Wang, M. Liserre, F. Blaabjerg, Toward reliable power electronics: challenges, design tools, and opportunities, IEEE Ind. Electron. Mag. 7 (2) (June, 2013) 17e26. [6] E. Koutroulis, F. Blaabjerg, Design optimization of transformerless gridconnected PV inverters including reliability, IEEE Trans. Power Electron. 28 (1) (January, 2013) 325e335. [7] Y. Yang, P. Enjeti, F. Blaabjerg, H. Wang, Widescale adoption of photovoltaic energy: grid code modifications are explored in the distribution grid, IEEE Ind. Appl. Mag. 21 (5) (September, 2015) 21e31. [8] E. Serban, M. Ordonez, C. Pondiche, Voltage and frequency grid support strategies beyond standards, IEEE Trans. Power Electron. 32 (1) (January, 2017) 298e309. [9] Y. Xue, K.C. Divya, G. Griepentrog, M. Liviu, S. Suresh, M. Manjrekar, Towards next generation photovoltaic inverters, in: Proceedings of 2011 IEEE Energy Conversion Congress and Exposition (ECCE), 17e22 September, 2011, pp. 2467e2474. [10] K.C. Divya, P.S. Nagendra Rao, Effect of grid voltage and frequency variations on the output of wind generators, Elec. Power Compon. Syst. 36 (6) (2008) 602e614. [11] M.H.J. Bollen, I.Y.H. Gu, Signal Processing of Power Quality Disturbances, IEEE Press and John Wiley & Sons, Inc., 2006. [12] A. Luna, et al., Grid voltage synchronization for distributed generation systems under grid fault conditions, IEEE Trans. Ind. Appl. 51 (4) (JulyAugust, 2015) 3414e3425. [13] Y. Yang, K. Zhou, F. Blaabjerg, Current harmonics from singlephase gridconnected invertersexamination and suppression, IEEE J. Emerg. Sel. Topics Power Electron. 4 (1) (March, 2016) 221e233. [14] L.A.C. Lopes, S. Huili, Performance assessment of active frequency drifting islanding detection methods, IEEE Trans. Energy Convers. 21 (1) (March, 2006) 171e180. [15] Q. Zheng, J. Li, X. Ai, J. Wen, J. Fang, Overview of grid codes for photovoltaic integration, in: Proc. Of 2017 IEEE Conference on Energy Internet and Energy System Integration (EI2), Beijing, 2017, pp. 1e6. [16] R. Walling, A. Ellis, S. Gonzalez, Implementation of Voltage and Frequency RideThrough Requirements in Distributed Energy Resources e Interconnection Standards, Sandia National Laboratories, April, 2014. Tech. Rep., SAND 20143122. [17] IEEE Standard Committee, IEEE standard for interconnecting distributed resources with electric power systems, IEEE Std. 1547 (2003). [18] IEEE Standard Committee, IEEE Standard for Interconnection and Interoperability of Distributed Energy Resources with Associated Electric Power Systems Interfaces, April, 2018, pp. 1e138. IEEE Std 15472018 (Revision of IEEE Std 15472003).
Low voltage ridethrough operation Chapter  16 [19] [20]
[21]
[22]
[23] [24]
[25]
[26]
[27]
[28] [29] [30]
[31]
[32]
[33]
497
Y. Yang, F. Blaabjerg, Overview of singlephase gridconnected photovoltaic systems, Elec. Power Compon. Syst. 43 (12) (2015) 1352e1363. L. Meegahapola, M. Datta, I. Nutkani, J. Conroy, Role of fault ride through strategies for power grids with 100% power electronic  interfaced distributed renewable energy resources, in: Proc. Of WIREs Energy Environ, March, 2018, pp. 1e24. Y. Yang, F. Blaabjerg, Z. Zou, Benchmarking of Grid Fault Modes in SinglePhase GridConnected Photovoltaic Systems, IEEE Trans. Ind. Appl. 49 (5) (SeptembereOctober, 2013) 2167e2176. S. Kumar, A.K. Verma, I. Hussain, B. Singh, C. Jain, Better control for a solar energy system: using improved enhanced phaselocked loopbased control under variable solar intensity, IEEE Ind. Appl. Mag. 23 (2) (MarchApril, 2017) 24e36. A. Mojallal, S. Lotfifard, Enhancement of grid connected PV arrays fault ride through and post fault recovery performance, IEEE Trans. Smart Grid 10 (1) (January, 2019) 546e555. M. Mirhosseini, J. Pou, V.G. Agelidis, Single and twostage inverterbased gridconnected photovoltaic power plants with ridethrough capability under grid faults, IEEE Trans. Sustain. Energy 6 (3) (July, 2015) 1150e1159. Q.C. Zhong, T. Hornik, Robust droop control with improved voltage quality, in: Control of Power Inverters in Renewable Energy and Smart Grid Integration, John Wiley & Sons/ IEEE, 2012, pp. 335e346. M. Ebrahimi, S.A. Khajehoddin, M. KarimiGhartemani, Fast and robust singlephase DQ current controller for smart inverter Applications, IEEE Trans. Power Electron. 31 (5) (May 2016) 3968e3976. M. Saitou, T. Shimizu, Generalized theory of instantaneous active and reactive powers in singlephase circuits based on Hilbert transform, in: Proc. 2002 IEEE 33rd Annual IEEE Power Electronics Specialists Conference. Proceedings (Cat. No.02CH37289), vol. 3, 2002, pp. 1419e1424. Y. Yang, F. Blaabjerg, H. Wang, M.G. Simo˜es, Power control flexibilities for gridconnected multifunctional photovoltaic inverters, IET Renew. Power Gener. 10 (4) (4 2016) 504e513. Y. Yang, F. Blaabjerg, H. Wang, Lowvoltage ridethrough of singlephase transformerless photovoltaic inverters, IEEE Trans. Ind. Appl. 50 (3) (MayJune 2014) 1942e1952. M. Ciobotaru, R. Teodorescu, F. Blaabjerg, Control of singlestage singlephase PV inverter, in: Proceedings of 2005 European Conference on Power Electronics and Applications (EPE), Dresden, Germany, 2005, pp. 1e10. Y. Yang, K. Zhou, H. Wang, F. Blaabjerg, D. Wang, B. Zhang, Frequency adaptive selective harmonic control for gridconnected inverters, IEEE Trans. Power Electron. 30 (7) (July, 2015) 3912e3924. T. Jakub, P. Zdenek, B. Vojtech, Central difference model predictive current control of singlephase Hbridge inverter with LCL filter, in: Proceedings of 2013 European Conference on Power Electronics and Applications (EPE), Lille, France, 2e6 September, 2013, pp. 1e8. J.F. Stumper, V. Hagenmeyer, S. Kuehl, R. Kennel, Deadbeat control for electrical drives: a robust and performant design based on differential flatness, IEEE Trans. Power Electron. 30 (8) (August, 2015) 4585e4596.
498 Control of Power Electronic Converters and Systems [34] Y. Yang, A. Sangwongwanich, H. Liu, F. Blaabjerg, Low voltage ridethrough of twostage gridconnected photovoltaic systems through the inherent linear powervoltage characteristic, in: Proceedings of 2017 IEEE Applied Power Electronics Conference and Exposition (APEC), Tampa, FL, 2017, pp. 3582e3588. [35] Y. Yang, H. Wang, F. Blaabjerg, Reactive power injection strategies for singlephase photovoltaic systems considering grid requirements, IEEE Trans. Ind. Appl. 50 (6) (NovembrDecember, 2014) 4065e4076. [36] Y. Yang, K.A. Kim, F. Blaabjerg, A. Sangwongwanich, Advances in GridConnected Photovoltaic Power Conversion Systems, Woodhead Publishing, 2018, p. 213.
Chapter 17
Gridfollowing and gridforming PV and wind turbines P. Rodriguez, N.B. Lai Luxembourg Institute of Science & Technology, EschSurAlzette, Luxembourg
17.1 Introduction The share of renewable energy in global power capacity has grown rapidly in the last decade. As a consequence, more stringent grid interconnection requirements have been imposed by grid operators aiming to increase grid flexibility and maintain grid stability. Both photovoltaic (PV) power and wind power (WP) plants are connected to the grid through power converters which, besides transferring the generated DC power to the AC grid, should now be able to provide some services to the grid, such as dynamic control of active and reactive power, frequency and voltage ridethrough, reactive current injection during faults, and participation in a grid voltage and frequency control, etc. In this chapter, we introduce the typical structures of PV and WP systems with a particular focus on the power electronic parts, especially gridconnected power converters. These power converters can operate either in the gridfollowing or gridforming mode. The former has been widely used in PV and WP plants with maximum power injection as a primary objective, whereas the latter has emerged in the last few years due to growing concerns regarding the displacement of synchronous generation units by converterinterfaced resources and recent advances in energy storage systems. In addition to an overview of the typical control structures, this chapter analyzes different strategies for the control and synchronization of gridfollowing and gridforming power converters. In Section 17.2, the latest development of PV and WP systems is reviewed. In Section 17.3, the gridfollowing converter is discussed. Here, popular synchronization strategies and current controllers are explained. Building upon these control blocks, serval representative implementations of grid forming is considered in Section 17.4. Finally, this chapter summary is discussed in Section 17.5.
Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000226 Copyright © 2021 Elsevier Ltd. All rights reserved.
499
500 Control of Power Electronic Converters and Systems
17.2 PV and wind turbine systems The installed capacity of PV and WP has increased exponentially in the last decade with annual additions of just above 30% and 10%, respectively [1], as shown in Fig. 17.1. This amount of additions accounts for approximately 83% of renewable power and 62% of all net power capacity installed in 2018 [2]. It is forecasted that the growth for PV and WP will reach 2264 GW by 2024. The major players are China, the European Union, the United States, and India, which contribute to about 75% of PV and WP capacities. Depending on the power ratings and technologies used, there are different configurations for these systems. The typical structures of PV and WP systems are outlined in the following subsections.
17.2.1 PV structures There are four most common types of inverter used in PV plants, namely microinverter, string inverter, multistring inverter, and central inverter. The configuration for different types of inverters is as depicted in Fig. 17.1. The choice of PV inverters is often based on the required peak power and net efficiency [3]. For instance, l l
l
l
Microinverter: typically, in the 50e500 Wp range for only one panel. String inverters: typically, in the 0.4e2 kWp range for small rooftop plants with panels connected in one string. Multistring inverters: typically, in the 1.5e150 kWp range for mediumlarge rooftop plants with panels configured in several strings. Central inverters: typically, in the 80e4000 kW range with threephase topology and modular design for large power plants ranging to hundredths of a MWp.
1400
Wind
PV
1200 1000 800
600 400 200 0 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 ACTUALS
FORECAST
FIGURE 17.1 Global cumulative installations for photovoltaic (PV) and wind power in GW. Source: IEA.
Gridfollowing and gridforming PV and wind turbines Chapter  17
501
Microinverter is the most effective way to utilize PV modules as each inverter is attached to a module, as shown in Fig. 17.2A. This also makes this configuration easy to expand. However, for large PV plants, a high number of power converters are required resulting in high installation and maintenance costs. On the contrary, the central inverter structure depicted in Fig. 17.2D uses only one power converter for all PV modules, thus reducing the total capitalization and operation cost of the system. The most popular implementation is based on a string converter Fig. 17.2B. In this setup, each string is connected to a power converter which balances the cost and efficiency of the system. For larger PV plants, a multistring inverter can also be used as in Fig. 17.2C.
17.2.2 Grid converter for wind turbine systems WP is generated by using wind turbine generators (WTGs), particularly, induction generators, doublefed induction generators, or synchronous generators. Generally speaking, the WTG can be connected to the grid either directly or through a power converter. The arrangement of the WTG and power converter in WP systems makes up four major topologies of wind turbines, as shown in Fig. 17.3, which can be classified as follows: l l
l l
Type 1: Induction generator (fixed speedd1%e2% regulation) Type 2: Induction generator with variable rotor resistance (limited variable speedd10% regulation) Type 3: Doublefed induction generator (variable speedd30% regulation) Type 4: Synchronous or asynchronous generators with full converter (variable speedd100% regulation).
Fixedspeed wind turbines are the first generation of wind turbines. Even though they are directly connected to the grid, they require additional components, such as a soft starter to reduce current transients during the startup
AC grid
AC grid PV module
PV string
(A)
DC bus
DC bus
(B)
AC grid
AC grid
(C)
(D)
FIGURE 17.2 Configuration of photovoltaic (PV) systems: (A) Microinverter, (B) String inverter, (C) Multistring inverter, (D) Central inverter.
502 Control of Power Electronic Converters and Systems
AC Grid
AC Grid IG
IG
(A) Fixedspeed wind turbine (Type 1)
(B)
Variableslip wind turbine (Type 2)
AC Grid
AC Grid IG/ SG
IG
(C)
DFIG wind turbine (Type 3)
(D)
Full converter wind turbine (Type 4)
FIGURE 17.3 Configuration of wind turbine generators in wind power systems. (A) Fixedspeed wind turbine (Type 1); (B) Variableslip wind turbine (Type 2); (C) Doublefed induction generator wind turbine (Type 3); (D) Full converter wind turbine (Type 4).
and a capacitor bank to compensate for reactive power. They need to operate at a rather constant speed (1%e2% regulation range). The speed governing is generally achieved by changing the aerodynamic profile of the WT blades, which has a quite slow response. Thus, the Type 1 wind turbine is rather sensible to wind disturbances. Wind turbines based on induction generators with variable slip were often used to slightly extend the speed regulation range (10% regulation range). This design, often referred to as Type 2 wind turbines, incorporates controllable rotor resistance to control the rotor slip. To eliminate the power loss on rotor resistance and extend the speed regulation range (30% regulation range), doublefed induction generator turbines are employed. Type 4 wind turbines employ a full backtoback power converter, which extends the WT speed controllability (100% regulation range), increasing energy extraction under high and low wind speeds, and improves the interaction with the electrical grid under abnormal operating conditions, e.g., during grid faults.
17.3 Gridfollowing power converters 17.3.1 Definition Gridfollowing power converters are gridconnected converters operated as controllable current sources to deliver the desired value of active and reactive power to an energized grid [4]. The regulation of active and reactive power delivered to the grid is achieved by characterizing (following) the grid voltage and calculating corresponding references for the current controller. As gridfollowing power converters work as current sources, they are suitable for parallel operation in gridconnected mode. A simplified representation for a
Gridfollowing and gridforming PV and wind turbines Chapter  17
503
Grid
P*
Q
*
CP
i*
Z
FIGURE 17.4 Simplified diagram of gridfollowing power converter.
gridfollowing power converter is shown in Fig. 17.4, where P and Q are the active and the reactive power references to be delivered to the grid, respectively. In this operation mode, it is important to highlight that the power controller should be perfectly synchronized with the AC voltage at the connection point (amplitude, frequency, and phase angle), in order to regulate accurately the instantaneous active and reactive power exchanged with the electrical grid. Currently, most of the power converters employed in PV or WP systems operate in gridfollowing mode [5]. Even though these power converters use to operate at the maximum power point to maximize renewable energy yield, they can also contribute to control of the grid voltage amplitude and frequency by including some additional control loop to adjust the active and reactive power references, P and Q , respectively [6,7]. In principle, gridfollowing power converters cannot operate in islanded mode if there is not a local synchronous generator or a gridforming power converter, to set the grid voltage amplitude and frequency of the islanded part. Fig. 17.5 shows a typical control structure for a gridfollowing power converter. The control system of gridfollowing power converters is primarily composed of a grid synchronization unit, a current controller, and a power controller.
17.3.2 Synchronization strategies The accuracy in the estimation of the AC grid voltage parameters has a strong influence on the overall performance of gridfollowing power converters. A precise synchronization algorithm is needed to estimate the grid voltage parameters, i.e., voltage amplitude, frequency, and phase angle, as these values are needed for conducting a precise control of the instantaneous active and the reactive power delivered to the grid. Additionally, precise monitoring of the grid conditions is mandatory in order to determine the most suitable operation mode of the converters, as well as for properly supporting connection and disconnection sequences. In this regard, two widely used types of phaselocked loop (PLL) will be introduced in the following.
504 Control of Power Electronic Converters and Systems
VDC
PCC
LF
+ 
i abc
v abc
CF
RL
PWM
θ abc dq
ω ff
id
ud
abc
u dq
dq
uq
PI
vq vd
abc
v abc
dq
iq
vd
θ u abc
Grid synchronization
θ
PI
id*
vd P*
+ωL −ωL PI
iq*
vd Q*
vq Current control loop
FIGURE 17.5
Power control loop
Basic control structure of gridfollowing power converter.
17.3.2.1 Synchronous reference frame phaselocked loop The PLL technology has been extensively used to synchronize gridconnected power converters with the grid. In threephase systems, the synchronous reference frame phaselocked loop (SRFPLL) has been broadly used for this purpose. The structure of the SRFPLL is depicted in Fig. 17.6. The SRFPLL translates the threephase instantaneous voltage waveforms from the abc reference frame into the dq rotating reference frame through the Park transformation. The angular position of this dqreference frame is controlled through a feedback control loop which drives the vq component to zero [8,9]. As depicted in Fig. 17.6, the estimated grid frequency is u0 , being the value for rated frequency normally included as a feedforward term, u , to improve the dynamics of the phase estimation, q0 , which is obtained by integrating u0 [10,11]. Despite the good behavior of the SRFPLL under balanced grid conditions, its performance is deteriorated when the threephase input signal becomes unbalanced or distorted. To overcome such a drawback, some advanced grid synchronization techniques have been proposed in the literature. This is, for
Gridfollowing and gridforming PV and wind turbines Chapter  17
505
SRFPLL abc
v abc
dq
vq
T'
Z'
PI
T'
Z*
FIGURE 17.6 Block diagram of synchronous reference frame phaselocked loop (SRFPLL).
instance, the case of the decoupled doubleSRF PLL, which is an enhanced PLL that stems from the same operation principle as the SRFPLL with improved phase angle and magnitude estimation under unbalanced conditions [12,13].
17.3.2.2 Stationary reference frame frequencylocked loop The grid voltage vector can experience relevant phase angle jumps during grid faults, which can give rise to significant transient synchronization errors that might threaten the stability of gridfollowing power converters. As a consequence, other implementations based on a frequencylocked loop (FLL) can be used, since frequency presents a more stable evolution during grid faults. Among different proposals, the synchronization structures based on a secondorder generalized integrator (SOGI) and an FLL to detect the grid voltage components, v0ab and qv0ab , as well as the grid frequency, u0 [14], as it is presented in Fig. 17.7, have proven to be effective and accurate under arbitrary
DSOGI FLL
H vDE '
qvDE
FLL
' qvDE
Z' v abc
abc DE
vDE
H vDE
kSOGI
k H vDE
Z'
DSOGI
' vDE
' vDE
FIGURE 17.7 Block diagram of dual secondorder generalized integrator frequencylocked loop (DSOGIFLL).
506 Control of Power Electronic Converters and Systems
grid conditions. The SOGI is implemented in both a and baxes, giving rise to a dual SOGI (DSOGI) structure, which is an adaptive bandpass filter that provides the filtered version of the input voltage vector, v0ab , as well as its quadrature component, qv0ab . The resonance frequency of the DSOGI is the grid frequency estimated by the FLL. As previously mentioned, one of the main advantages of using an FLL lays on the fact that this structure is less sensitive than the PLL to phase angle jumps occurred in the grid voltage during transient grid faults, improving, thus, the power converter response under abnormal grid conditions. This interesting feature of the FLL provides a fast dynamic response with small overshoot, allowing thus a fast and smooth transition between the gridconnected and the islanded operating modes.
17.3.3 Current controllers The inner control loop of gridfollowing power converters is based on fast current controllers that regulate the current injected into the grid [15]. The references for current controllers are often generated by power controllers which determine the amount of power to be exchanged with the grid [16,17]. Such reference currents are usually calculated as a function of the reference powers, P and Q [18,19]. The most widely used solutions for implementing linear current controllers in threephase systems are those based on the wellknown PI controller working on a dqSRF or the ones based on the implementation of a resonant controller working on an abstationary reference frame [15]. In addition to these proposals, those controllers based on nonlinear control structures, such as hysteresis, sliding mode, or predictive controllers, can be also used for tracking sinusoidal reference currents in a fast and robust way [19].
17.3.3.1 PI controller on the SRF The implementation of current controllers based on the dqSRF has been extensively used in the control of AC currents in threephase systems. Through the Park transformation, the sinusoidal currents under control can be represented as DC quantities on an orthogonal dq frame, rotating synchronously at the detected grid fundamental frequency. In this reference frame, two independent control loops are in charge of regulating the direct and quadrature current components. In the case of gridfollowing converters, the reference currents id and iq are usually provided by a power controller, which regulates the active and reactive power delivered to the grid. The instantaneous active and reactive power components are calculated by p ¼ vd id þ vq iq ; q ¼ vd iq vq id :
(17.1)
Gridfollowing and gridforming PV and wind turbines Chapter  17
507
Fig. 17.5 shows the structure of a dqbased synchronous current control, including the grid voltage feedforward and the decoupling network used to improve the performance of the controller [19]. However, in this standard structure, the PI controllers are unable to suppress the oscillations that appear in the dq signals under unbalanced grid conditions. To overcome this drawback, two dq synchronous controllers may be implemented in order to regulate independently both the positive and the negativesequence current components of the injected current. Likewise, multiple reference frames [20], rotating at multiples of the fundamental frequency, should be implemented to properly control the harmonic currents injected into the AC grid.
17.3.3.2 Resonant controller in a stationary reference frame This kind of controllers work with AC variables expressed on the abstationary reference frame [21]. In this case, the PI controllers are replaced by proportional resonant (PR) controllers, whose resonance frequency is tuned at the fundamental grid frequency estimated by the grid synchronization system [22,23]. The transfer function of a PR controller can be written as follows: Gab PR ðsÞ ¼ kP þ
n X kR s kih s þ ; s2 þ u20 h¼2 s2 þ ðhu0 Þ2
(17.2)
where kP is the proportional gain, kR is the resonant gain at the grid fundamental frequency, kih is the resonant gain at the hharmonic to be controlled, and u0 is the estimated fundamental frequency [15]. Similar to the case of dq synchronous controllers, the reference currents on the abstationary reference frame are calculated by the power controller, which is in charge of regulating the power exchanged with the grid. The instantaneous active and reactive power components in the abstationary reference frame are calculated by p ¼ va ia þ vb ib ; q ¼ vb ia va ib :
(17.3)
There is a significant advantage in the implementation of PR controllers in a stationary reference frame, instead of using PI controllers working in a dqSRF, when unbalanced sinusoidal currents control is intended. In such a case, those implementations based on PR controllers do not need to use any decoupling network, neither independent sequence controllers, since resonant controllers are able to regulate both positive and negativesequence components simultaneously. This feature makes PR controllers a very convenient candidate to be applied to regulate the current injected by gridfollowing power converters under generic grid conditions, even under grid faults. Moreover, additional harmonic compensators can be implemented straightforwardly, through tuning multiple PR controllers at the desired harmonic frequencies ðh $u0 Þ working in parallel. The control structure of a gridfollowing converter based on resonant controllers is shown in Fig. 17.8.
508 Control of Power Electronic Converters and Systems DG LF
VDC
i abc
CF
v abc
iD
i abc
iE
uD u abc
v abc
Pmeas
i abc
Qmeas
iD*
uDE
vD
iE*
vE uE
RL
i*pDE
6
Cp
i*qDE
Z' qu'DE
Cq u'DE
Z'
v abc
FIGURE 17.8 Basic structure of the gridfollowing converter implemented in Stationary Reference Frame Control, with proportional resonant controller (PR) and Harmonic Compensator (HC).
17.4 Gridforming power converters 17.4.1 Definition According to recent definitions, a gridforming power converter should be able to support the operation of the AC power system under normal, alerted, emergency, blackout, and restoration states without having to rely on services from synchronous generators [24]. Therefore, a gridforming power converter could be represented in a simplified manner as controlled AC voltage source with a given amplitude E and frequency u, as shown in Fig. 17.9. As a voltage source, such a power converter would present low output impedance, which might complicate its operation in parallel with other gridforming converters, e.g., when operating in interconnected grids. In such a case, the natural (unregulated) current sharing among paralleled gridforming converters would depend on the value of their physical output impedances, which are rather low. Therefore, in a realistic implementation of a gridforming power converter, it will be necessary to include an inner current controller to regulate and limit the maximum current injected by the power converter to the grid. Moreover, in order to naturally regulate the current sharing among multiple gridforming power converters connected in parallel, it would be interesting to incorporate a dedicated mechanism to emulate the output impedance shown by the power converter to the grid. In addition, some outer
Gridfollowing and gridforming PV and wind turbines Chapter  17
509
Grid
Z* E*
Cv
v*
Z
FIGURE 17.9 Basic representation of a gridforming power converter.
control loops would be enabled to contribute to grid frequency and voltage regulation by modifying the active and reactive power references of the power converter. These loops would play a similar role to the governor and the automatic voltage regulator in the case of synchronous generators. A simple example of a gridforming power converter could be a standalone UPS. This system remains connected from the main grid when the operating conditions are within certain limits. In the case of a grid failure, the power converter of the UPS forms the grid voltage. In the islanded part of the grid, the AC voltage generated by the gridforming power converter will be used as a reference for the rest of gridfollowing power converters connected to it.
17.4.2 Control schemes for gridforming power converter Fig. 17.10 shows an example of a controller for a simplified gridforming power converter, which is implemented by using two cascaded PI controllers working on the dqreference frame. The inputs to the control system are the grid voltage amplitude V and the grid frequency u to be formed by the power converter at the point of common coupling (PCC). The outer loop controls the grid voltage to match its reference value, while the inner control loop regulates the current supplied by the converter. Therefore, the controlled current flowing through the inductor LF charges the capacitor CF to keep the output voltage close to the reference provided to the voltage control loop. Usually, in industrial applications, these power converters are fed by stable DC voltage sources driven by PV panels, batteries, fuel cells, or other primary sources. It should be pointed out that, in the simplified implementation shown in Fig. 17.10, the voltage control loop of the gridforming power converter will be enabled only when it is disconnected from the main network and works in islanded mode. Otherwise, several gridforming converters would compete each other to set the grid voltage when working in parallel. As previously mentioned, the inner current loop of the cascade configuration in Fig. 17.10 regulates the current supplied by the power converter, tracking the reference
510 Control of Power Electronic Converters and Systems DG VDC
PCC
LF
+ 
CF
RL
PWM
T
T abc
i abc
u abc
dq
iq
abc
v abc
dq
u dq
vq
id*
ud PI
uq
iq*
Z*
vd
vd
T abc
dq
T
id
³
vd* PI
vq*
T dq
v*
abc
vq Current control loop
Volatge control loop
FIGURE 17.10 Cascaded implementation of gridforming power converter.
current provided by the outer voltage loop. It is worth noting that the gridforming power converters can be controlled in both the dqsynchronous [25] and the abstationary reference frames as in the case of gridfollowing power converters [26]. Some typical schemes for the control loops previously mentioned to regulated current and power sharing among paralleled gridforming converters are discussed in following sections.
17.4.3 Droop control in gridforming power converters The implementation of the gridforming power converter shown in Fig. 17.10 does not allow controlling power sharing control among paralleled power converters. To overcome such a shortcoming, several powersharing strategies have been proposed in the literature, such as centralized controllers, mastereslave, average load sharing, or circular chain controls [27]. However, such solutions use to assume that paralleled converters are physically connected close to each other and linked through highbandwidth communication networks. In real cases, distributed generator systems and loads may be spread over larger geographic areas, so communicationbased solutions can become impractical due to technical and economic reasons. To address this issue, droop control algorithms have been traditionally used to control the power sharing among gridconnected power converters without using any
Gridfollowing and gridforming PV and wind turbines Chapter  17
511
communication link, thereby eliminating the limits imposed by physical location and improving the grid performance [28]. The droop regulation techniques are implemented in gridforming power converters to regulate the exchange of active and reactive powers with the grid, thereby contributing to regulate the grid voltage frequency and amplitude, respectively.
17.4.3.1 Grid impedance influence on droop control Considering the power converter as an ideal controllable voltage source that is connected to an infinite bus through a given line impedance, as shown in Fig. 17.11A, the active and reactive powers that it will deliver to the grid can be written as follows: VA ½RðVA VB cos dÞ þ XVB sin d; R2 þ X 2
(17.4)
VA ½RVB sin d þ XðVA VB cos dÞ; R2 þ X 2
(17.5)
PA ¼ QA ¼
where PA and QA are the active and reactive powers, respectively, flowing from the source A (power converter) to the B (grid), VA and VB are the voltage magnitudes of these sources, d corresponds to the phase angle difference between the two voltage vectors, Z ¼ R þ jL is the interconnection line impedance, and q is the impedance angle. As R ¼ Z$cos q and X ¼ Z$sin q, the performance of this simplified electrical system can be depicted by its vector representation as shown in Fig. 17.11B [29]. 17.4.3.1.1 Inductive grid The inductive component of the line impedances in highvoltage (HV) and mediumvoltage (MV) transmission networks is typically much higher than the resistive one, as shown in Table 17.1 [30].
I I
VA0
jX
G R
VB G
I I
(A)
VA VB
jX ·I R·I
VQ
VD
(B)
FIGURE 17.11 Simplified modeling of power converter connection to a distribution network. (A) Equivalent circuit, (B) phasor diagram.
512 Control of Power Electronic Converters and Systems
TABLE 17.1 Typical line impedances values. Type of line
R (U/km)
X (U/km)
R/X (p.u.)
Lowvoltage line
0.642
0.083
7.7
Mediumvoltage line
0.161
0.190
0.85
Highvoltage line
0.06
0.191
0.31
Therefore, the resistive part can be neglected without losing generality. Additionally, the power angle, d, in such lines is rather small, so it can be assumed that sin dzd and cos dz1 [31]. Therefore, Eqs. (17.4) and (17.5) can be rewritten as follows: PA z QA z
VA ðVB sin dÞ X
VA ðVA VB cos dÞ X
0 dz
XPA ; VA V B
0 VA VB z
(17.6) XQA : VA
(17.7)
Expressions Eqs. (17.6) and (17.7) show a direct relationship between the power angle d and the active power P, as well as between the voltage difference VA VB and the reactive power Q. From Eqs. (17.6) and (17.7), a smallsignal analysis allows determining the relationship between the active power variation and grid frequency variation, as well as between reactive power variation and voltage variation. Such expressions are known as droop control expressions and can be described by Eqs. (17.8) and (17.9) for inductive lines as follows: f f0 ¼ kp ðP P0 Þ;
(17.8)
V V0 ¼ kq ðQ Q0 Þ;
(17.9)
where f f0 and V V0 represent the grid frequency and the voltage deviations, respectively, from their rated values, and P P0 and Q Q0 are the variations in the active and reactive powers delivered by the power converter to compensate such deviations. These relationships permit regulating the grid frequency and voltage at the point of connection of the power converter, by controlling the value of the active and reactive powers delivered to the grid and can be graphically represented by the droop characteristics shown in Fig. 17.12, where the gain of the control action in each case, i.e., the slope of the frequency and voltage droop characteristic, is set by the kp and kq parameters, as indicated in Eqs. (17.8) and (17.9). Therefore, as depicted in Fig. 17.12, each of the gridforming power converters will adjust its active and reactive power reference according to its P/f and Q/V droop characteristics to participate in the regulation of the grid frequency and voltage, respectively.
Gridfollowing and gridforming PV and wind turbines Chapter  17 Zline
Z
513
V
ZO
VO
PO
QO
P
Q
FIGURE 17.12 Frequency and voltage droop characteristics in grids with dominant inductive behavior.
17.4.3.1.2 Resistive grid On the contrary to the case of HV networks, the grid impedance in lowvoltage (LV) networks is mainly resistive, as it is shown in Table 17.1, and thus the inductive part can be neglected. As a consequence, maintaining the assumption that the power angle, d, is small, the expressions Eqs. (17.4) and (17.5) give rise to the following: PA z
VA ðVA VB cos dÞ R
QA ¼
VA $VB sin d R
0 VA V B z 0 dz
RPA ; VA
RQA : VA VB
(17.10) (17.11)
Therefore, the voltage amplitude in LV networks depends mainly on the active power flow, while their frequency is mainly affected by the reactive power injection. From Eqs. (17.10) and (17.11), the following droop control expressions can be written for resistive lines: V V0 ¼ kp ðP P0 Þ;
(17.12)
f f0 ¼ kq ðQ Q0 Þ;
(17.13)
being their droop characteristics represented in Fig. 17.13, which depicts the P=V and the Q=f droop control actions to be taken in resistive networks for regulating the grid voltage and frequency.
Zline
Z
V
ZO
VO
PO
P
QO
Q
FIGURE 17.13 Voltage and frequency droop characteristics in mainly resistive grids, generally in lowvoltage systems.
514 Control of Power Electronic Converters and Systems
17.4.3.1.3 General case In the general case, the combined effect of the resistive and inductive line impedance components should be taken into account in the droop control equations. To do that, a rotation matrix T is used to transform the active and reactive powers, P and Q, into the rotated power components, P0 and Q0 , as detailed in the following: 0 P cos f sin f P X=Z R=Z P P ¼ ½T$ ¼ $ ¼ $ ; Q sin f cos f Q R=Z X=Z Q Q0 (17.14) where f ¼ p=2 q ¼ arctan ðR =XÞ, f is the rotation angle of the matrix T and q is the angle of the line impedance, Z ¼ Z:q. Provided that d takes a small value, the application of the T rotation matrix to Eqs. (17.4) and (17.5) results in the following simplified equations: P0A z Q0A z
VA ðVB sin dÞ Z
0 dz
VA ðVA VB cos dÞ 0 X
ZP0A ; VA V B
VA VB z
(17.15) ZQ0A ; VA
(17.16)
where P0A and Q0A are the rotated components of PA and QA . From Eqs. (17.15) and (17.16), it can be concluded that the power angle, d, can be controlled by regulating the rotated active power, P0A , while the voltage difference VA VB can be changed by regulating the rotated reactive power, Q0A . Therefore, in a general case, the droop control equations can be written as follows: X R f f0 ¼ kp P0 P00 ¼ kp ðP P0 Þ þ kq ðQ Q0 Þ; Z Z
(17.17)
R X V V0 ¼ kq Q0 Q00 ¼ kp ðP P0 Þ kq ðQ Q0 Þ: Z Z
(17.18)
According to Eqs. (17.17) and (17.18), the contribution in the compensation of the frequency and the voltage amplitude variations by each gridforming power converter can be adjusted by changing the values kp and kq .
17.4.3.2 Virtual impedance control Conventional P=f and Q=V droop controls have been proven to be an effective solution for regulating the voltage magnitude and frequency in transmission networks, where the lines have a predominant inductive behavior. However, as shown previously, the performance of this kind of control is highly dependent on the R=X ratio of the line [32]. Due to this feature, this method cannot be directly applied in all kinds of MV networks, unless some grid impedance
Gridfollowing and gridforming PV and wind turbines Chapter  17
515
estimation algorithms are implemented to calculate the rotated powers as indicated in Eq. (17.14). This issue is even more important when the droop control is applied to LV distribution networks, as stated in Refs. [28,33]. In such a case, a small mismatch in the grid impedance estimation results in an inefficient power sharing among the droopcontrolled generations. As an intuitive solution to address these drawbacks, resulting from the strong dependence of the conventional droop controller performance on the line impedance value, large inductors could be used to link the power converter to the AC bus, and thereby the line impedance would be predominantly inductive. However, this is not an efficient solution since, in addition to the increase in the size and the costs, the DCbus voltage level should be significantly increased to compensate for the high voltage drop across these inductors, reducing thus the overall efficiency. A more effective solution is to virtually introduce the effect of this link impedance into the control system of the gridforming power converter. This concept was successfully implemented in Refs. [32,34], where an adjustable virtual output impedance was used for regulating the power sharing among parallelized inverters and for limiting overcurrent under grid disturbances. It is worth to remark that the value of the virtual impedance should be higher than the actual line impedance, otherwise it will not have a predominant effect in the power flow equations. The virtual impedance modifies the power converter output voltage reference as indicated in Eq. (17.19), where the modified voltage reference, v , is obtained by subtracting the virtual voltage drop across the virtual impedance, ZV $igrid , from the reference value originally provided by the droop equations, v . v ¼ v ZV $igrid :
(17.19)
The value of ZV sets the controller response, hence it must be considered as a control variable and should be selected according to the nominal power of the converter. An example of the implementation of the virtual impedance concept in a gridforming power converter is shown in Fig. 17.14. Fig. 17.14. also represents the implementation of the droop equations shown in Eqs. (17.8) and (17.9) to regulate the frequency amplitude of the virtual electromotive force (emf), e ¼ E$sinðutÞ, assuming the virtual output impedance of the power converter is predominantly inductive. Conceptually, these droop control loops enable the gridforming power converter to conduct grid synchronization based on power balance, instead on direct grid voltage parameters estimation through a PLL or FLL. For instance, in case the grid frequency was lower than the frequency of the virtual emf of the power converter, the power angle d would trend to increase. As a consequence, according to Eq. (17.6), the active power delivered by the power converter
516 Control of Power Electronic Converters and Systems
FIGURE 17.14 Block diagram of the virtual output impedance loop working with P and Q droop method in the grid power converter.
would trend to increase as well. Such an increment in the output active power would be detected by the P/u droop control loop of Fig. 17.14, and according to Eq. (17.8), the internal frequency of the gridforming converter would be reduced according to the parameter kP. A similar process, but driven by reactive power given by Eq. (17.9), would be followed in case of deviations in the value of the voltage at the PCC.
17.4.4 The synchronous power controller Although droop controllers somehow allow implementing gridforming functionalities in power converters, they do not physically describe the dynamic response of electromechanical synchronous generators. This is particularly interesting when the gridforming converters are required to provide inertial response in case of transient grid frequency deviations. Actually, the inertialess behavior of most of conventional gridfollowing power converters has raised significant concerns about power system stability, particularly when their share in power systems becomes substantial. This has led to a new implementation of gridforming power converters which is based on the concept of virtual synchronous machine (VSM). In this implementation, the control system of the power converter is altered to mimic the dynamic response of an electromechanical synchronous machine. There have been various proposals for VSM including the VISMA [34], the synchronverter [35], the synchronous power controller (SPC) [36], to name a few of them. Among these approaches, the SPC has stood out as one of most the simple and effective practical implementations. The block diagram of the SPC is shown in Fig. 17.15, where the power converter is controlled as a current source. This enables the SPC to be implemented in most of the
Gridfollowing and gridforming PV and wind turbines Chapter  17
517
FIGURE 17.15 Block diagram the synchronous power controller.
gridfollowing power converters currently available in the market, which generally use current controllers similar to the ones described in Section 17.2. Since girdforming converter should provide support services during grid transients, such as inertia emulation, the main requirements for current controller are often low settling time and robustness. The fact of using a currentcontrolled voltage source converter to connect to the electrical grid requires to substitute the virtual impedance block shown in Fig. 17.14 by a virtual admittance block. In this manner, the references for the current controller are generated by such a virtual admittance block, whose input is the difference between measured grid voltage, v, and the virtual emf, e. Actually, such a current reference results from solving the igrid current in Eq. (17.19). Therefore, the transfer function of the virtual admittance block can be written as follows: Ga ðsÞ ¼
1 ; R þ Ls
(17.20)
where R and L represent the value for the virtual resistance and the virtual inductance, respectively. It is worth noting that Eq. (17.20) has the form of a lowpass filter rather than a derivative term that would result from Eq. (17.19) [37], which results in a more stable implementation of the virtual impedance (admittance) effect. Parameters of the virtual admittance block are usually chosen to meet the reactive power support requirements [38]. Combining multiple virtual admittance blocks to control different voltage sequences and frequencies is also used [39].
518 Control of Power Electronic Converters and Systems
The virtual emf, e, of the SPC is generated by using a voltagecontrolled oscillator, which combines the outputs of the power loop controller (PLC) and the PIbased reactive controller. Similar to a synchronous generator, the grid synchronization through input/output power balancing is achieved by such PLC. As shown in Ref. [40], several transfer functions can be used to implement the PLC. For instance, it can be implemented through a simple firstorder lowpass filter transfer function as follows: uc PLCðsÞ ¼ k ; (17.21) s þ uc where k and uc are the proportional gain and the cutoff frequency, respectively. This feature allows the SPC to provide synthetic inertia to support the grid. For designing the PLC, a smallsignal model of the SPC is often adopted, as the one shown in Fig. 17.16, where Pmax ¼ EV=X results from the synchronous power expression shown in Eq. (17.6). From Fig. 17.16, the closed loop of the closedloop transfer function of the resulting powerlocked loop can be given as follows: Ps kPmax uc u2n ¼ 2 ¼ 2 ; Pin s þ uc s þ kPmax uc s þ 2xun s þ u2n
(17.22)
where x and un are the damping factor and natural frequency. From Eq. (17.22), the gain and cutoff frequency can be calculated as follows: un uc ¼ 2xun and k ¼ ; (17.23) 2xPmax giving rise to the following expression to calculate the virtual moment of inertia emulated by the SPC controller: J¼
Pmax ; u2n us
(17.24)
where us is fundamental grid frequency in rad/s. It is worth to notice that multiple PLCs can be paralleled and tuned with different subsynchronous frequencies to damp multiple modal oscillations in power systems [41].
Zn Pin
Zs
[ PLC
'Z
Tg
Zr
³
Tr
G
Pmax
Ps
Ps FIGURE 17.16 Small signal of active power control loop of the synchronous power controller.
Gridfollowing and gridforming PV and wind turbines Chapter  17
519
17.5 Conclusions In this chapter, various control schemes for the gridconnected power converters used in PV and WP systems have been discussed. The most relevant concepts presented in this chapter are the following: l
l
l
l
l
l
l
l
l
There are four main configurations for power converter in PV systems, namely microinverter, string inverter, multistring inverter, and central inverter. Depending on the combination of the induction machine and the power converter, WP systems are classified into four types: fixedspeed, variableslip, doublefed induction generator, and full converter wind turbines. A gridconnected power converter in PV and WP can be operated either in gridfollowing or gridforming mode. The most frequently used synchronization mechanisms for grid connected are the SRFPLL and the FLL in the rotating and the stationary reference frames, respectively. Current controllers can be implemented by using PI controllers in SRF or PR controller in the stationary reference frame. Gridforming power converters need some current/power sharing mechanism to operate in parallel or gridconnected mode. Gridforming power converters can be implemented by using both conventional droop strategies or VSMs approaches. Droopbased strategies are sensitive to line impedance. Virtual impedance/ admittance can be used to address such a limitation. The SPC is one of the most effective strategies to implement a gridforming power converter, which can provide not only steadystate voltage and frequency regulation but also dynamic support to the grid such as inertial response and power oscillation damping.
References [1] [2] [3] [4] [5] [6] [7]
IEA, Renewables 2019, 2019. Paris. REN21, Renewables 2019 Global Status Report, REN21 Secretariat, Paris, 2019. Fraunhofer Institute for Solar Energy, Photovoltaics Report, 2020. J. Rocabert, A. Luna, F. Blaabjerg, P. Rodrı´guez, Control of power converters in AC microgrids, IEEE Trans. Power Electron. 27 (11) (2012) 4734e4749. J.T. Bialasiewicz, Renewable energy systems with photovoltaic power generators: operation and modeling, IEEE Trans. Ind. Electron. 55 (7) (July 2008) 2752e2758. H.H. Zeineldin, A Qef droop curve for facilitating islanding detection of inverterbased distributed generation, IEEE Trans. Power Electron. 24 (3) (March 2009) 665e673. H.H. Zeineldin, E.F. Elsaadany, M.M.A. Salama, Distributed generation microgrid operation: control and protection, in: In 2006 Power Systems Conference: Advanced Metering, Protection, Control, Communication, and Distributed Resources, 2006, pp. 105e111.
520 Control of Power Electronic Converters and Systems [8] P. Rodriguez, A. Lunar, R. Teodorescu, F. Iov, F. Blaabjerg, Fault ridethrough capability implementation in wind turbine converters using a decoupled double synchronous reference frame PLL, in: In 2007 European Conference on Power Electronics and Applications, 2007, pp. 1e10. [9] V. Kaura, V. Blasko, Operation of a phase locked loop system under distorted utility conditions, IEEE Trans. Ind. Appl. 33 (1) (January 1997) 58e63. [10] S.K. Chung, A phase tracking system for three phase utility interface inverters, IEEE Trans. Power Electron. 15 (3) (May 2000) 431e438. [11] A. Timbus, R. Teodorescu, F. Blaabjerg, M. Liserre, Synchronization methods for three phase distributed power generation systems. An overview and evaluation, in: In PESC Record  IEEE Annual Power Electronics Specialists Conference, 2005, 2005, pp. 2474e2481. [12] P. Rodrı´guez, J. Pou, J. Bergas, J.I. Candela, R.P. Burgos, D. Boroyevich, Decoupled double synchronous reference frame PLL for power converters control, IEEE Trans. Power Electron. 22 (2) (March 2007) 584e592. [13] S. Cobreces, E.J. Bueno, D. Pizarro, F.J. Rodriguez, F. Huerta, Grid impedance monitoring system for distributed power generation electronic interfaces, IEEE Trans. Instrum. Meas. 58 (9) (2009) 3112e3121. [14] P. Rodrı´guez, A. Luna, M. Ciobotaru, R. Teodorescu, F. Blaabjerg, Advanced grid synchronization system for power converters under unbalanced and distorted operating conditions, in: In IECON Proceedings (Industrial Electronics Conference), 2006, pp. 5173e5178. [15] F. Blaabjerg, R. Teodorescu, M. Liserre, A.V. Timbus, Overview of control and grid synchronization for distributed power generation systems, IEEE Trans. Ind. Electron. 53 (5) (October 2006) 1398e1409. [16] N. Pogaku, M. Prodanovic, T.C. Green, Modeling, analysis and testing of autonomous operation of an inverterbased microgrid, IEEE Trans. Power Electron. 22 (2) (March 2007) 613e625. [17] M.C. Chandorkar, D.M. Divan, R. Adapa, Control of parallel connected inverters in standalone AC supply systems, IEEE Trans. Ind. Appl. 29 (1) (January 1993) 136e143. [18] P. RodrI´guez, A. Timbus, R. Teodorescu, M. Liserre, F. Blaabjerg, Reactive power control for improving wind turbine system behavior under grid faults, IEEE Trans. Power Electron. 24 (7) (July 2009) 1798e1801. [19] A. Timbus, M. Liserre, R. Teodorescu, P. Rodriguez, F. Blaabjerg, Evaluation of current controllers for distributed power generation systems, IEEE Trans. Power Electron. 24 (3) (March 2009) 654e664. [20] M. Liserre, R. Teodorescu, F. Blaabjerg, Multiple harmonics control for threephase grid converter systems with the use of PIRES current controller in a rotating frame, IEEE Trans. Power Electron. 21 (3) (May 2006) 836e841. [21] E. Clarke, Circuit Analysis of AC Power Systems, vol. 1, Wiley, New York, NY, 1943. [22] R. Ca´rdenas, C. Juri, R. Pen˜a, P. Wheeler, J. Clare, The application of resonant controllers to fourleg matrix converters feeding unbalanced or nonlinear loads, IEEE Trans. Power Electron. 27 (3) (March 2012) 1120e1129. [23] A.M. Roslan, K.H. Ahmed, S.J. Finney, B.W. Williams, Improved instantaneous average currentsharing control scheme for parallelconnected inverter considering line impedance impact in microgrid networks, IEEE Trans. Power Electron. 26 (3) (March 2011) 702e716.
Gridfollowing and gridforming PV and wind turbines Chapter  17 [24]
[25] [26]
[27] [28]
[29] [30] [31]
[32]
[33]
[34] [35] [36]
[37]
[38] [39]
[40]
[41]
521
ENTSOE Technical Group on High Penetration of Power Electronic Interfaced Power Sources, High Penetration of Power Electronic Interfaced Power Sources and the Potential Contribution of Grid Forming Converters, January 2020. F. Katiraei, R. Iravani, N. Hatziargyriou, A. Dimeas, Microgrids management, IEEE Power Energy Mag. 6 (3) (May 2008) 54e65. R. Teodorescu, F. Blaabjerg, Flexible control of small wind turbines with grid failure detection operating in standalone and gridconnected mode, IEEE Trans. Power Electron. 19 (5) (2004) 1323e1332. J.M. Guerrero, L. Hang, J. Uceda, Control of distributed uninterruptible power supply systems, IEEE Trans. Ind. Electron. 55 (8) (August 2008) 2845e2859. J.M. Guerrero, J. Matas, L.G. De Vicun˜a, N. Berbel, J. Sosa, Wirelesscontrol strategy for parallel operation of distributed generation inverters, in: In IEEE International Symposium on Industrial Electronics, vol. II, 2005, pp. 845e850. K. De Brabandere, Voltage and Frequency Droop Control in Low Voltage Grids by Distributed Generators with Inverter FrontEnd, October 2006, p. 227. A. Engler, Applicability of droops in low voltage grids, Int. J. Distrib. Energy Resour. Smart Grids 1 (1) (2005) 1e5. H. Laaksonen, P. Saari, R. Komulainen, Voltage and frequency control of inverter based weak LV network microgrid, in: 2005 International Conference on Future Power Systems, 2005, 2005, p. 6. W. Yao, M. Chen, J. Matas, J.M. Guerrero, Z. Qian, Design and analysis of the droop control method for parallel inverters considering the impact of the complex impedance on the power sharing, IEEE Trans. Ind. Electron. 58 (2) (February 2011) 576e588. J.M. Guerrero, L. Garcia de Vicuna, J. Matas, M. Castilla, J. Miret, Output impedance design of parallelconnected UPS inverters with wireless loadsharing control, IEEE Trans. Ind. Electron. 52 (4) (August 2005) 1126e1135. H. Beck, R. Hesse, Virtual synchronous machine, in: In 2007 9th International Conference on Electrical Power Quality and Utilisation, 2007, pp. 1e6. Q. Zhong, G. Weiss, Synchronverters: inverters that mimic synchronous generators, IEEE Trans. Ind. Electron. 58 (4) (April 2011) 1259e1267. P. Rodrı´guez, C. Citro, J.I. Candela, J. Rocabert, A. Luna, Flexible grid connection and islanding of SPCbased PV power converters, IEEE Trans. Ind. Appl. 54 (3) (May 2018) 2690e2702. S. D’Arco, J.A. Suul, O.B. Fosso, Control system tuning and stability analysis of virtual synchronous machines, in: In 2013 IEEE Energy Conversion Congress and Exposition, 2013, pp. 2664e2671. W. Zhang, J. Rocabert, J.I. Candela, P. Rodriguez, Synchronous power control of gridconnected power converters under asymmetrical grid fault, Energies 10 (7) (July 2017) 950. A. Tarraso´, J.I. Candela, J. Rocabert, P. Rodriguez, Grid voltage harmonic damping method for SPC based power converters with multiple virtual admittance control, in: 2017 IEEE Energy Convers. Congr. Expo. ECCE, January 2017, pp. 64e68. W. Zhang, et al., Comparison of different power loop controllers for synchronous power controlled gridinteractive converters, in: 2015 IEEE Energy Conversion Congress and Exposition, Montreal, QC, 2015, pp. 3780e3787. P. Rodriguez, J.I. Candela, J. Rocabert, R. Teodorescu, Virtual Controller of Electromechanical Characteristics for Static Power Converters, US20140067138A1, 2012.
This page intentionally left blank
Chapter 18
Virtual inertia operation of renewables Rasool Heydari1, Mehdi Savaghebi1, Frede Blaabjerg2 1 University of Southern Denmark, Odense, Denmark; 2Department of Energy Technology, Aalborg University, Aalborg, Denmark
18.1 Introduction Currently, around 28% of the overall greenhouse gas emissions are produced by fossil fuelebased electricity generations [1]. “Replacing fossil fuels as an energy source with green renewable energy is the most important action we can take to address the impacts of climate change,” Rame Hemstreet, Chief Energy Officer [2]. Therefore, the European Union (EU) has set energy targets to reduce greenhouse gas emissions to 90% by 2050 [3]. For instance, Germany has already decided to shut down all the coalbased generation by 2038 [4]. An increasing number of countries, e.g., Ireland and Denmark, some states in the United States, e.g., California and Hawaii, as well as utilities, e.g., American Electric Power and Xcel Energy, set targets of 100% renewables or 100% carbonfree generation [5]. Distributed renewable energy sources (RESs) connected by power converters are becoming prominent components in the modern power system. Based on the international renewable energy agency reports, the proportion of the installed microsource distributed generation units capacity has been increased from 77.5 GW in 2000 to 1225 GW in 2018, as shown in Fig. 18.1 [6]. Furthermore, the global wind energy council has reported that more than 341,000 wind turbines generated electricity in 2017 [7], and the EU has set a binding target of 20% electrical generation from renewable sources by 2020 [6]. However, high penetration of RESs leads to some technical challenges such as instability problems, frequency deviations, and synchronization problems in the power grid due to the lowinertia characteristics of nonsynchronous generators (NSGs). Power converters, which serve as interfaces between an RES and the common AC bus, have no mechanical kinetic energy. Therefore, high penetration of the converterbased generations and NSGs, with zero inertia, Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000111 Copyright © 2021 Elsevier Ltd. All rights reserved.
523
524 Control of Power Electronic Converters and Systems
FIGURE 18.1 Global cumulative installed DGU capacity trend from 2000 to 2018 [4].
introduces new challenges in terms of frequency and voltage control as well as stability of the grid. Lower inertia in the grid yields a more considerable impact of disturbances on the system frequency and voltage control. The rotating parts of the synchronous generators and turbines interchange inertial energy with the grid so that the primary controllers have time to react to the grid disturbance or load/generation imbalance. However, the converterinterfaced RES does not provide any inertial response toward the power grid. Consequently, a large generation/load imbalance or a perturbation on the system may lead to the significant frequency and voltage deviations, which in turn may lead to load shedding or system instabilities in the power system and in worst case power outage. This socalled lowinertia challenge is stated as the main bottleneck to integrate a high penetration of NSG in the power network in both literature and industry. For instance, due to the increasing penetration of NSG, frequency violations have increased in the Nordic grid [8]. In Ireland and Northern Ireland, the inadequate grid inertia significantly affected the frequency and voltage regulations [9]. Many aspects should be considered in the design phase of RES integration regarding inertia challenges and providing virtual inertia as a solution. Efficiency, reliability, and stability are the significant concerns of operation and control of NSGbased systems. This chapter first overviews the lowinertia grid challenges and main resultant difficulties in the world power systems. Then, the focus is on virtual inertia concept, theory, and control solutions in a grid with a high penetration of renewables. Design examples and some PowerFactory simulation results are also provided.
Virtual inertia operation of renewables Chapter  18
525
18.2 Evolution of green energy transition Wind and solar photovoltaics (PV) generation are currently the fastest growing sources of electricity due to the drop in the price of power electronic components and renewable energy technologies [6]. There are two main challenges with high penetration of NSGs from the power system operation point of view. First, PV and wind are variable energy resources, and their outputs are uncertain and change over minutes, hours, or days. This variable characteristic leads to challenges with power system balancing and resource adequacy during peak demand. Secondly, these NSGs, e.g., wind and PV, are connected to the main grid through power converters, which means they are nonsynchronously coupled to the main grid. Changing fault behavior from power converters, lower short circuit capability, and decreasing system inertia are sources of concern as SGs are displaced by NSGs massively. This lowinertia challenge degrades the grid reliability, stability, and performance, if not addressed appropriately. The main focus of this chapter is on the second challenge, i.e., lowinertia grids with high penetration of RESs.
18.2.1 Lowinertia grid challenges The green transition from conventional rotational synchronous generators toward power converterebased NSG, as in the case of RESs, battery energy storages, or highvoltage DC (HVDC) links, will pose major challenges to the control, operation, and stability of power grid due to [10e13] l
l
l
The loss of kinetic energy stored in the rotating mass of the synchronous generator. The loss of a robust and stable synchronization mechanism, which is physically inherent in a synchronous machine. The loss of a global rotating speed and frequency signal, and consequently, loss of a robust frequency and voltage control.
In the modern power grid, these functionalities have to be addressed by proper control of the power converters. For instance, Fig. 18.2 shows a frequency response comparison between a lowinertia system and a synchronous generatorebased highinertia grid. In the lowinertia grid, in the event of a generation loss, the rate of change of frequency (RoCoF) is higher, and consequently, the frequency nadir is lower due to the lack of system inertia [14,15]. This problem may lead to major instability and cascading failures in the power system. Experiences from Ireland indicate that at the high level of wind power penetration (above approximately 50%), the frequency deviations and dynamic stability deteriorate significantly (see Fig. 18.3 [16]).
526 Control of Power Electronic Converters and Systems
FIGURE 18.2 A comparison between the system frequency response in a lowinertia grid (solid black line) and an SGbased highinertia network (dashed).
FIGURE 18.3 The frequency deviation versus wind penetration in Ireland [16].
The grid of South Australia state is located at the end of the Eastern Australian interconnection. On September 28, 2016, cascading failure of the electricity transmission network led to a huge blackout in South Australia, where 40% of generation in this region is provided by the wind turbines. When a fault had occurred, 9 out of 15 wind farms did not respond to the system accurately and just shut off, which resulted in almost the entire state losing its electricity supply [17].
Virtual inertia operation of renewables Chapter  18
527
Therefore, Australian Electricity Market Operator studied a large electromagnetic transient and electromechanics model of the South Australian interconnected grid, which includes all synchronous generators, gridconnected RESs, generation protection systems, and the transmission network. The results of the studies have identified the following: (1) The technical requirement for a minimum of four or five, 150e200 MVA, synchronous generators to keep online in South Australia at all times. (2) The limited maximum wind output under lowinertia conditions. These constraints resulted in the curtailment of 4% of output wind generation in winter, 2018 [17]. Eastern and Western Denmark are connected to the large interconnected synchronous areas in Scandinavia and Continental Europe, respectively [18]. In 2003, technical errors on the Swedish power plants resulted in substantial power outages in large parts of Sweden and Eastern Denmark. The first contingency has been occurred with the outage of a 1200MW nuclear unit in Southern Sweden, due to some problems with a steam valve. Consequently, the system experienced voltage collapse leading to the separation of a region of the Southern Swedish and Eastern Denmark systems. Therefore, this lowinertia islanded system collapsed in both frequency and voltage, and thus led to a huge blackout in Copenhagen and Eastern Denmark [19]. The plans for decommissioning nuclear plants, e.g., in Germany and Sweden [20], will adversely affect the interconnected grid inertia so that it would be a major challenge for the integration of NSG in Denmark and Europe in the near future. Considering the aforementioned problems, Ireland currently limits its instantaneous penetration of RES to 55% [9]. Ireland’s grid is islanded with limited DC interconnection to Great Britain. During November and December 2018, the wind power supplied around 43% of the total electrical energy consumed in Ireland (over the year, wind provides approximately 30% of entire demand). In 2008, the transmission system operator of Ireland and Northern Ireland (EirGrid) analyzed the impact of high penetrations of NSG on the grid transient stability. The results indicate that the operator faces several daunting challenges to reaching 75% system nonsynchronous penetration and 1Hz/s RoCoF [5,16]. Today, EirGrid operates the interconnected grid up to 55% NSG with a 0.5Hz/s RoCoF limit [9]. With high NSG penetration levels, the grid instability issues significantly arise due to the low kinetic energy and synchronizing torque [21].
18.2.2 Control of power convertereinterfaced renewables In the conventional power system, synchronous generators provide the inertial response and ensure system stability. However, the power convertere interfaced RES does not contribute to such ancillary services [22].
528 Control of Power Electronic Converters and Systems
All the rotating mass of synchronous generators in power systems are electromechanically coupled to each other; hence, during stable operation, they rotate with synchronous speed. It equivalently can be represented as a huge shaft rotating at the nominal synchronous frequency (60 or 50 Hz). Fig. 18.4 shows the strong coupling of synchronous generators in the conventional power system and equivalent rotating shaft. This huge shaft equivalently represents the inertia of the system, which consists of all the rotating mass of synchronous generators in the grid. By operating renewables, they just follow the reference angular speed of the synchronous generators in the system, since the power convertereinterfaced renewables have no rotating mass to contribute in power system kinetic energy. Power convertereinterfaced renewables should be able to provide inertia for the system in order to allow their high penetration in the modern grids. In a general structure, the power converter controllers can be divided into two main structures: grid following and grid forming [23,24]. Gridfollowing controllers represent the most common type of control structure for gridconnected wind or PV converters. At the core of its operation, a gridfollowing control strategy employs a phaselocked loop in order to determine the instantaneous sinusoidal voltage angle at the converter output. Therefore, the power electronics are operated to inject a controlled current
FIGURE 18.4 The strong coupling of synchronous generator in the electric power grid, where also renewables are interconnected.
Virtual inertia operation of renewables Chapter  18
529
into the main power grid that “follows” the sinusoidal reference voltage. Thus, it is called a gridfollowing unit. As the main limitation, gridfollowing converters work under the presumption that a stiff AC voltage with minimal amplitude and frequency deviations is maintained at its terminals so that it can only follow its local voltage and inject a controlled current. In practice, this translates to the assumption that the collective behavior of the synchronous machines, the generator and system controllers, as well as voltage regulating equipment on the system, provides a sufficiently stiff frequency and voltage at any point on the grid. Conventionally, this assumption has held up relatively well because the cumulative amount of converterinterfaced RES with gridfollowing inverters has been relatively small compared to conventional synchronous generators that regulate the system frequency and voltages [24]. Moreover, the gridfollowing control structure allows all the harvested energy to be exported to the main grid and provides accurate control of the current flow [25,26]. However, with higher penetration of gridfollowing converterinterfaced RESs, various significant concerns are raised: l l l
l l
Increasing the RoCoF, and consequently, frequency instability. Loss of synchronizing power/torque. Voltage instability challenges during postfault, e.g., voltage collapse, or postfault overvoltage. Challenges with modeling the electricity power systems dynamic behavior. Subsynchronous oscillations and possible interaction with conventional synchronous generator in the system.
To overcome this shortcoming of a conventional gridfollowing control structure, it is necessary to develop the nextgeneration gridforming control strategies that enable the transition to a converterbased generation and which are capable of regulating the system voltages and frequency, directly, through local decentralized control structure.
18.3 Virtual inertiabased control A solution toward stabilizing the lowinertia grid and increasing the penetration of RESs and NSGs while the system maintained stable is to virtually provide additional inertia. The virtual inertia can be provided by RESs utilizing shortterm energy storages together with a power converter and a proper control structure. Inertia can be also emulated directly from RESs, with the cost of operating the wind or PV below the maximum power point [27]. Notice that to employ the controller providing virtual inertia for an NSG converter, a certain amount of energy storage or a highspeed responding energy source is needed to interchange synthetic inertial energy with the grid [28].
530 Control of Power Electronic Converters and Systems
18.3.1 Virtual synchronous machine The concept of virtual synchronous machine (VSM) or virtual synchronous generator control strategy is presented in the literature with different original variants: (1) VISMA: VIrtual Synchronous MAchine (VISMA) is proposed in Ref. [29], in which a standalone power converter mimics a synchronous machine. Virtual torque and virtual excitation are embedded into the VISMA to regulate the output power. Furthermore, a cascaded structure is implemented where voltage and frequency controllers feed references to the current controller. (2) Synchronverter: Operating a power converter to mimic a synchronous generator is the socalled Synchronverter and initially presented in Ref. [30]. A synchronverter is equivalent to a synchronous machine with a capacitor bank connected in parallel to the stator terminal. (3) PSC: Power synchronization control (PSC) is presented in Ref. [31], which emulates the power synchronization mechanism between synchronous machines in the power system. The proposed PSC is a gridforming control structure for the gridconnected power converters, but may be of most importance for HVDC connected to a weak AC grid system. The power converter in all various aforementioned approaches will then operate like a synchronous machine, providing inertia and damping properties of a conventional synchronous generator. In this chapter, the notation of “VSM” is employed.
FIGURE 18.5 The general concept of virtual synchronous machine (VSM).
Virtual inertia operation of renewables Chapter  18
531
The aim of VSM control is the emulation of a synchronous machine dynamics by a power converter, so that, from the power system point of view, the converter should be able to have very similar dynamics as a synchronous machine (Fig. 18.5). The major objectives of VSM control of power converters are as follows [32]: l l
l l
Providing virtual inertia to contribute to system frequency control. Enhancing the stability properties for connection to lowinertia grids and weak grids. Enabling stable operation with higher penetration of NSGs. Improving the frequency nadir (maximum frequency deviation) and the RoCoF in the system.
Mostly, the various approaches for VSM control are based on emulating the swing equation of a synchronous machine, which is shown in Fig. 18.6.
18.3.2 Concepts and fundamentals The kinetic energy (Er) of the rotating mass in a synchronous machine with the moment of inertia J kg m2 and angular rotor speed ur ½rad =s is as follows: 1 Er ¼ Ju2r 2
(18.1)
FIGURE 18.6 Virtual synchronous machine (VSM) control structure emulated from the dynamic model of a synchronous machine.
532 Control of Power Electronic Converters and Systems
The power, which accelerates the rotating mass, can be obtained by differentiating (18.1) as follows: dEr dur ¼ Jur : dt dt
(18.2)
If the variables are expressed in the perunit system, and the angular speed varies slightly around the nominal angular synchronous frequency ur0 ½rad =s, the firstorder swing equation can be expressed as follows: Pm Pe ¼ 2H H¼
dur þ PD dt
Ju2r0 2SBase
(18.3) (18.4)
dd ¼ ur ug dt
(18.5)
PD ¼ Kd ður ug Þ
(18.6)
where Pm is the prime mover mechanical power, Pe is the electrical power. The damping term PD ¼ Kd ður ug Þ is used to emulate the effect of damper windings in the synchronous machine to suppress the grid frequency oscillations. Furthermore, ur , ug , and d are the rotor angular speed, grid reference machine rotor angular speed, and the rotor angle, respectively. Finally, H ½s is the inertia constant and SBase ½VA is the base rating of the machine. Note that in a practical case, ug (the speed of the reference machine) is not available. Thus, the damping term loop presented here is hard to realize without remote measurements, and a wide area control system [28]. By employing the control law in (18.3), the VSM control should enable the power converter to appear as a VSM, with a similar inertial response, in the grid. Fig. 18.6 shows the control schematic of VSM emulating the proposed swing Eq. (18.3).
18.4 VSM implementation In order to evaluate the performance of the proposed control structure, a 50Hz, fourmachine test system inspired by that of Ref. [33] is implemented in PowerFactory DIgSILENT. This is a 230 kV power grid containing four synchronous generators and two main loads in its original form. All the generators are equipped with the automatic voltage regulator, and G1 and G3 are equipped with hydrofrequency governors. Fig. 18.7 shows the electrical network of the test system.
Virtual inertia operation of renewables Chapter  18
533
FIGURE 18.7 A fourmachine test bench implemented in PowerFactory DIgSILENT, to illustrate inertia influence.
18.4.1 NSG penetration level In the first study, the penetration level of NSGs is increased in order to analyze the system behavior. Generator 2 (G2) and generator 4 (G4) of the original model are replaced with wind turbine generators with the same power rating in three scenarios, without inertia emulation: l
l
l
Scenario 1: All the generators in the system are synchronous hydrogenerators with hydroturbines. Scenario 2: 25% of the generators are replaced with NSG. (G2 is replaced with a wind turbine generator with the same power rating.) Scenario 3: 50% of the generators are replaced with NSG. (G2 and G4 are replaced with wind turbine generators with the same power rating.)
As it is expected and shown in Fig. 18.8, when a generation trip (G1.1) is applied, by increasing the penetration of NSGs, the RoCoF is increased, and also the maximum frequency deviation is increased.
FIGURE 18.8 Effect of increasing the penetration of nonsynchronous generator.
534 Control of Power Electronic Converters and Systems
18.4.2 VSM performance As it can be seen from Fig. 18.7, a 140 MVA synchronous condenser (SC) and a converterinterfaced energy storage system (ESS) with the same power rating are included at node 6 to be alternatively connected in different scenarios. SC is a synchronous machine where its rotating shaft spins freely. The SC is connected to the same node as the converterinterfaced ESS to serve as a benchmark for analyzing the behavior of a converter controlled by the VSM. As shown in Table 18.1, the SC and converterinterfaced ESS, controlled by the VSM, have same parameters. The studies are implemented by means of RMS, phasorbased simulation in DIgSILENT PowerFactory. The converter at node 6 is controlled by the proposed VSM approach with the inertia constant H ¼ 5 like the SC. Therefore, it should have a similar dynamic behavior as the SC, which is connected to the same node. A generation trip at node 1 is applied for two cases, i.e., the SC is connected or the converter controlled by the VSM is connected. As depicted in Fig. 18.9, the converter controlled by the VSM has a very similar dynamic behavior as a synchronous machine. Therefore, from the power system point of view, the VSM control performance is validated.
18.4.3 Fault right through capability It is important to note that in a controller based on the swing equation, the current is not controlled in the short time frame directly. However, the converter current needs to be limited, especially during a fault, in order to avoid overcurrent blocking of the converter [28,34]. Therefore, a current limitation controller should be implemented inside the VSM control. Fig. 18.10 shows
TABLE 18.1 Parameters of the virtual synchronous machine (VSM) controller and synchronous condenser benchmark. Parameters
Synchronous condenser
Converter controlled by the VSM
Sn (MVA)
140
140
Inertia constant H (s)
2.5
2.5
Inductance Lc (p.u.)
0.25
0.25
Damping factor Kd
e
25
Current control loop bandwidth
e
20
Current limit (p.u.)
e
1.0
Virtual inertia operation of renewables Chapter  18
535
FIGURE 18.9 System frequency and the active output power of a synchronous condenser (PSC) and a power converter (PVSM) as the virtual synchronous machine (VSM).
FIGURE 18.10 Current limitation control of voltage source converters proposed in Ref. [28].
the current limitation strategy proposed in Ref. [28]. In case that the converter current is above its limit (Imax), the desired voltage control law is defined as follows: (18.7) vcref ¼ ac Lc icref ic þ ju1 Lc ic þ vp Here, ac is the desired current control loop bandwidth, and icref stands for the converter inner current reference. ic , Lc , and u1 stand for the converter current, the converter inductance, and the network angular frequency, respectively. The term ju1 Lc ic represents the feedback decoupling of the
536 Control of Power Electronic Converters and Systems
voltage across the converter inductor and vp is a feedforward of the point of common coupling voltage filtered through a lowpass filter. In normal operation, when the converter current is less than the current limit, icref0 ¼ icref and vcref0 ¼ vcref . In this case, the reference current can be defined as follows: 1 c icref0 ¼ (18.8) vref0 ju1 Lc ic vp þ ic ac Lc The current reference icref in (18.8) is an indication of the converter real current flow. During a fault, the current limitation is achieved automatically by limiting the magnitude of icref to the predefined maximum current (Imax). In the next study, a threephase to ground fault cleared without line disconnection at t ¼ 6 s is applied in the middle of the line 2, between nodes 6 and 7, near the converter controlled by the VSM and SC, as shown in Fig. 18.7. Fig. 18.11 shows the output voltage and current of the SC during the fault. As can be seen, the output current is larger than 2 p.u. The same fault is applied for the ESS/VSM without the current limitation controller. Fig. 18.12 shows the voltage and current of the ESS/VSM. As it is expected, a similar dynamic behavior as for the SC can be observed. However, for the converter, it is essential to respect the current limitation. Thus, the current limit control is activated for the same fault case, see Fig. 18.13. As it can be seen, during the fault, the current is limited by the VSM controller to 1 p.u.
18.5 Summary and future trend The synchronous rotating inertia is the key parameter of power system stability. Increasing the penetration of NSGs, e.g., wind turbine, PV, and ESS, in
FIGURE 18.11 Voltage and current of the synchronous condenser during a 100m threephase to ground fault at line 2 in Fig. 18.7.
Virtual inertia operation of renewables Chapter  18
537
FIGURE 18.12 Voltage and current of converter controlled by virtual synchronous machine without current limitation during a 100m threephase to ground fault at the line 2, in Fig. 18.7.
FIGURE 18.13 Voltage and current of converter controlled by virtual synchronous machine with the current limitation controller, during a 100ms threephase to ground fault at the line 2, in Fig. 18.7.
the power grids leads to some stability challenges due to the lack of rotating mass and kinetic inertia in the system. In this chapter, the effect of high penetration of renewables on power system stability is shown. Moreover, the recent lowinertia grid challenges in Europe and South Australia are discussed. Technically, in order to achieve 100% renewables, providing virtual inertia is a demand. The VSM is a promising solution to emulate the behavior of a synchronous machine and provide inertia virtually. In this chapter, the theoretical concept of VSM, as well as the control structure, is presented. Furthermore, the VSM implementation in the power system is investigated, and some RMS simulation results are provided in a fault situation. Since in a
538 Control of Power Electronic Converters and Systems
controller based on the swing equation, the current is not controlled in the short time frame directly, a current limitation controller has also been implemented in VSM and scrutinized in this chapter. Although several aspects of the virtual operation of renewables are investigated in this chapter, there are still some open issues: l
l
l
l
Inertia estimation: Determining the system inertia is going to be crucial for the power system operators in the modern power grid with the high RES penetrations. Accurate inertia estimation will help the grid operators to set up a framework for planning and procuring the inertial services. Improvement of modeling and analysis: Although there are many of the challenges related to the control structure of converterinterfaced RESs, and their inertial responses have been highlighted in the recent publications, a lack of VSM modeling and implementation in power systemelevel studies is still observed. Storage system for virtual inertia: Typically, batteries, capacitors, and supercapacitors have been proposed as ESS for dynamic control of power convertereinterfaced RESs. Coordination, price, lifetime, and reliability of the designed ESS should be examined. Coordination between VSM and synchronous machines: Standards related to overall power system performance, with higher penetration of RESs controlled by VSM, and their coordination, should comprehensively be analyzed.
References [1] EPA. United Stated Environmental Protection Agency, Sources of Greenhouse Gas Emissions, Climate Change, 2015 [Online]. Available, http://www.epa.gov/climatechange/ ghgemissions/sources/transportation.html. (Accessed 8 April 2020). [2] Importance of Renewable Energy in the Fight against Climate Change, World Wildlife Magazine (WWF), 2015 [Online]. Available: https://www.worldwildlife.org/magazine/ issues/summer2015/articles/importanceofrenewableenergyinthefightagainstclimatechange–3. (Accessed 22 April 2020). [3] European Commission, Energy Roadmap 2050, 2011. [4] German commission proposes coal exit by 2038  Clean Energy Wire. [Online]. Available: https://www.cleanenergywire.org/factsheets/germancommissionproposescoalexit2038. (Accessed 08 April 2020). [5] D. Lew, et al., Secrets of successful integration: operating experience with high levels of variable, inverterbased generation, IEEE Power Energy Mag. 17 (6) (2019) 24e34. [6] International renewable energy agency (IRENA), Trends in Renewable Energy. [Online]. Available: https://www.irena.org/Statistics/ViewDatabyTopic/CapacityandGeneration/ StatisticsTimeSeries. (Accessed 08 April 2020). [7] Global Wind Energy Council, Global Wind Report 2019. [Online]. Available: https://gwec. net/globalwindreport2019/. (Accessed 08 April 2020). [8] ENTSOE, High Penetration of Power Electronic Interfaced Power Sources, HPoPEIPS), 2017.
Virtual inertia operation of renewables Chapter  18 [9]
[10]
[11]
[12] [13]
[14] [15]
[16]
[17]
[18] [19]
[20]
[21] [22]
[23]
[24] [25]
539
B. Kroposki, et al., Achieving a 100% renewable grid: operating electric power systems with extremely high levels of variable renewable energy, IEEE Power Energy Mag. 15 (2) (2017) 61e73. F. Dorfler, S. Bolognani, J.W. SimpsonPorco, S. Grammatico, Distributed control and optimization for autonomous power grids, in: 2019 18th European Control Conference, ECC 2019, 2019, pp. 2436e2453. T. Ackermann, T. Prevost, V. Vittal, A.J. Roscoe, J. Matevosyan, N. Miller, Paving the way: a future without inertia is closer than you think, IEEE Power Energy Mag. 15 (6) (2017) 61e69. L. Zhang, H.P. Nee, L. Harnefors, Analysis of stability limitations of a VSCHVDC link using powersynchronization control, IEEE Trans. Power Syst. 26 (3) (2011) 1326e1337. C. Li, I. Cvetkovic, R. Burgos, D. Boroyevich, C. Engineering, V. Tech, Assessment of virtual synchronous machine based control in gridtied power converters, in: 2018 Int. Power Electron. Conf. (IPECNiigata 2018 ECCE Asia), 2018, pp. 790e794. Y. Khayat, R. Heydari, et al., On the secondary control architectures of AC microgrids: an overview, IEEE Trans. Power Electron. 35 (6) (2020) 6482e6500. R. Heydari, T. Dragicevic, F. Blaabjerg, Highbandwidth secondary voltage and frequency control of VSCBased AC microgrid, IEEE Trans. Power Electron. 34 (11) (2019) 11320e11331. L. M. Robbie Aherne, Operating High Variable, Renewable Generation Power Systems Lessons Learned from Ireland and Northern Ireland. [Online]. Available: https:// cleanenergysolutions.org/training/highvariablerenewablegenerationpowersystemsireland. (Accessed 08 April 2020). AEMO, Update Report e Black System Event in South Australia on an Update to the Preliminary Operating Incident Report for the National Electricity Market, October 2016. dk Energinet, New Paths to the Energy of the Future Annual Report 2017, no. 28980671, 2017. G. Andersson, et al., Causes of the 2003 major grid blackouts in North America Europe, and recommended means to improve system dynamic performance, IEEE Trans. Power Syst. 20 (4) (2005) 1922e1928. Decommissioning of Nuclear Facilities: Germany’s Experience, International Atomic Energy Agency, 2016 [Online]. Available: https://www.iaea.org/newscenter/news/ decommissioningofnuclearfacilitiesgermanysexperience. J. Chen, T. O’Donnell, Parameter constraints for virtual synchronous generator considering stability, IEEE Trans. Power Syst. 34 (3) (2019) 2479e2481. F. Milano, F. Dorfler, G. Hug, D.J. Hill, G. Verbic, Foundations and challenges of lowinertia systems (Invited Paper), in: 20th Power Syst. Comput. Conf. PSCC 2018, 2018, pp. 1e25. B.J. Matevosyan, H. Urdal, S. Achilles, J. Macdowell, J.O. Sullivan, R. Quint, Gridforming inverters: are they the key for high renewable penetration? IEEE Power Energy Mag. 17 (3) (2019) 89e98. J. Rocabert, A. Luna, F. Blaabjerg, Control of power converters in AC microgrids, IEEE Trans. Power Electron. 27 (11) (2012) 4734e4749. B.K. Poolla, D. Groß, F. Do¨rfler, Placement and implementation of gridforming and gridfollowing virtual inertia and fast frequency response, IEEE Trans. Power Syst. 34 (4) (2019) 3035e3046.
540 Control of Power Electronic Converters and Systems [26] Z. Wang, F. Zhuo, H. Yi, J. Wu, F. Wang, Z. Zeng, Analysis of dynamic frequency performance among voltagecontrolled inverters considering virtual inertia interaction in microgrid, IEEE Trans. Ind. Appl. 55 (4) (2019) 4135e4144. [27] J. Fang, H. Li, Y. Tang, F. Blaabjerg, On the inertia of future moreelectronics power systems, IEEE J. Emerg. Sel. Top. Power Electron. 7 (4) (2019) 2130e2146. [28] R. Heydari, N. Johansson, L. Harnefors, F. Blaabjerg, A VSM converter controller implemented for RMS simulation studies of electrical power systems, in: 18th Wind Integration Workshop, 2019, pp. 1e6. [29] H.P. Beck, R. Hesse, Virtual synchronous machine, in: 2007 9th International Conference on Electrical Power Quality and Utilisation, EPQU, 2007. [30] Q.C. Zhong, G. Weiss, Synchronverters: inverters that mimic synchronous generators, IEEE Trans. Ind. Electron. 58 (4) (2011) 1259e1267. [31] L. Zhang, L. Harnefors, H. Nee, Powersynchronization control of gridconnected voltagesource converters, IEEE Trans. Power Syst. 25 (2) (2010) 809e820. [32] R. Heydari, M. Savaghebi, F. Blaabjerg, Fast frequency control of lowinertia hybrid grid utilizing extended virtual synchronous machine, in: Power Electron. Drive Syst. Technol. Conf. (PEDSTC 2020), 2020, pp. 1e5. [33] P. Kundur, Power System Stability and Control, McGraw Hill, 1998. [34] P. Control, L. Zhang, L. Harnefors, S. Member, H. Nee, S. Member, Interconnection of two very weak AC systems by VSCHVDC links using, IEEE Trans. Power Syst. 26 (1) (2011) 344e355.
Chapter 19
Virtual inertia emulating in power electronicebased power systems Yousef Khayat1, Mobin Naderi1, Hassan Bevrani1, Frede Blaabjerg2 1 University of Kurdistan, Sanandaj, Kurdistan, Iran; 2Department of Energy Technology, Aalborg University, Aalborg, Denmark
19.1 Introduction The concept of virtual inertia is adopted from the moment of inertia of synchronous generator (SG) rotating masses operating in a power system. The conventional power systems profit from the inertial function of numerous existing SGs to solve or improve the frequency challenges, e.g., low damping and high nadir due to load/generation disturbances. However, the modern power systems are becoming penetrated by a large number of the inertialess power electroniceinterfaced distributed energy resources (DERs) while typically are voltage source converters with a DC link. Hence, the rotational DER portion decreases impressively and the power system frequency experiences more intensive changes due to disturbances than the conventional SGdominated power systems. In order to solve the lowinertia challenges of the power electronicebased power systems, the concept of virtual inertia is introduced, which is realized by applying a control function on the power electroniceinterfaced DERs to mimic the SG inertial dynamics and to provide the inertia, virtually. The main source of the virtual inertia is the shortterm stored energy in the DC link of the DER power converters, which should be injected to the AC side according to the virtual inertia control objective. In fact, the DER power converters are controlled to surmount the lowinertia challenges including highfrequency nadir and high rate of change of frequency (RoCoF), low frequency/power oscillation damping, frequency instability, severe changes triggering protection devices mistakenly, and all key functions for power system operation, as mentioned before in Chapter 17. The idea of emulating inertia dynamics for power electroniceinterfaced DERs has been presented in different control structures using many control methods. Firstly, the virtual synchronous machine is introduced to allow a grid Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000196 Copyright © 2021 Elsevier Ltd. All rights reserved.
541
542 Control of Power Electronic Converters and Systems
compatible integration of DERs to be electromechanical SGs [11]. The idea is presented as a solution for injecting the virtual inertia by the inverterbased DERs in the form of virtual synchronous generator (VSG) [1], and it has developed extensively until now [2,3]. In Ref. [4], an internal synchronization mechanism has been proposed for the gridconnecting voltage source converters, where the synchronous machine dynamics is mimicked. Finally, the gridconnected operation of an inverter with the ideal DC link is realized by emulating the SG dynamics entitled Synchronverter [5], which is developed employing the nonlinear stabilizer [6], mimicking a synchronous condenser [7], and improving the dynamic response [8]. Among the virtual inertia control structures, the VSG is more taken into account and developed. The VSGbased gridsupporting DERs are allowed to be some active components to support the frequency dynamics of the microgrids [9]. By adjusting the VSG inertia and/or damping coefficient, the output power oscillations and the overall VSG behavior can be improved significantly, which is demonstrated by smallsignal modeling and sensitivity analysis [10,11]. A virtual capacitor control to enhance the reactive power sharing in VSGbased MGs in Ref. [12] and an adaptive linear quadratic regulatore based VSG in Ref. [13] to improve the inertial response of the system, have also been presented. In order to minimize the frequency deviation and output power oscillation of the VSG, a selfadaptive VSG control method based on RoCoF [14] and a selftuningebased algorithm [15] have been employed to continuously optimize the virtual inertia among predefined amounts of inertia moment and damping coefficient. Similar methods are addressed in Refs. [16,17], where different values of the parameters have been employed to improve the frequency response. The virtual inertia response is improved by tuning droop gain in Ref. [18,19]. In Ref. [20], a fuzzy control method is used to improve the frequency response of a wind turbine system, which is developed for islanded and gridconnected microgrids in Ref. [3,21], and at the same time emulate virtual inertia. The rest of this chapter is organized as follows: Section 19.2 basically deals with the inertia concept. Inertia challenges of the power electronicebased power systems are introduced in Section 19.3. In Section 19.4, the virtual inertia is presented as a fundamental solution. Finally, a case study is addressed in Section 19.5 to show the benefits of the virtual inertia using an adaptive control method.
19.2 Inertia concept Inertia is the property of an object to continue its existing rest state or uniform motion into a straight line, unless its state is changed by an external force/ disturbance. In fact, inertia is the object resistance to its velocity changes. It can be simply said that the larger inertia of the object leads to a slower velocity changes. The inertia of an object corresponds to its mass. It is easy to know
Virtual inertia emulating in power electronic Chapter  19
543
that objects of higher mass resist against changes in motion/velocity more than objects of lower mass. In other words, heavier objects have larger inertia. The moment of inertia or rotational inertia can be considered as the required torque for a desired angular acceleration similar to the object mass, which determines the force required for a desired linear acceleration. Considering T as the torque applied to a rotatable mass and a as the angular acceleration, the moment of inertia J is obtained as follows: T ¼ Ja:
(19.1)
For a given a, a larger moment of inertia causes a higher torque. The moment of inertia is also known as the dual of mass in the angular motion. It is realized for a differential mass dm rotating at a perpendicular distance r from the rotation axis as follows: Z J ¼ r 2 dm: (19.2) Generally, it is obvious that a larger mass results in a higher inertia. The SGs as the backbone of the traditional power systems benefit from the concept of inertia. The SG rotor with other coupling masses, especially turbines, provides an inertia to resist the frequency (electrical velocity) changes, i.e., acceleration/deceleration caused by power system disturbances, e.g., faults and load changes. Eq. (19.1) shows such a resistance, where T is the accelerating torque as follows: T ¼ Tm T e ;
(19.3)
and a is the angular acceleration as follows: a¼
du : dt
(19.4)
Tm is the mechanical torque rotating the SG rotor to produce electromotive force, Te is the electrical restraining torque against Tm, and u is the angular frequency or mechanical rotor speed. Substituting Eqs. (19.3) and (19.4) into Eq. (19.1), the wellknown swing equation can be found as follows: J
du ¼ Tm Te : dt
(19.5)
Without any disturbance and in the steadystate condition, Te equals to Tm. Thus, du/dt equals to zero and the frequency is remained constant. During power disturbances, Te, which equals to Pe/u, starts to change immediately. However, Tm related to the mechanical and control instruments cannot be influenced immediately, and it varies with a delay around a few seconds according to the rotating mass amount and controller delays. Therefore, in the first moments of the disturbance, the difference between Te and Tm causes frequency variations. According to the disturbance size, a damped oscillation,
544 Control of Power Electronic Converters and Systems
constant oscillation, or instability may occur. Obviously, for a given TmeTe, a larger amount of inertia J leads to a lower RoCoF (du/dt). Moreover, considering u0 as the predisturbance frequency, and integrating Eq. (19.5), one can find the frequency u as follows: Z 1 t u ¼ u0 þ ðTm Te Þdt; (19.6) J t0 where the right term can be considered as the overshoot/undershoot of the transient response. Therefore, a larger amount of inertia results in a lower frequency overshoot/undershoot. The swing equation is generalized to the power system loadgeneration balance [22,23]. For a power system including a number of SGs and loads interconnected through a power network, the electrical torque can be considered as follows: Te ¼ TeS þ TeD ;
(19.7)
where TeS is the synchronizing torque and TeD is the damping torque, generally provided by the load damping and SG rotor damping windings. TeD can be considered as a coefficient of the frequency changes given as follows: TeD ¼ KTD Du;
(19.8)
where KTD is the damping torque coefficient. The synchronizing torque can also be considered as a coefficient of the rotor angle changes Dd. Substituting Eq. (19.8) into Eq. (19.5), the swing equation is rewritten as follows: J
du ¼ Tm TeS KTD Du: dt
(19.9)
Since the active power is a more common variable in the power system studies, Eq. (19.9) is desired to be rewritten as follows considering P ¼ Tu: Ju
du ¼ Pm PeS KD Du; dt
(19.10)
where Pm and PeS are the mechanical power and synchronizing electrical power of the power system, and KD is known as the damping coefficient. Eq. (19.10) is a nonlinear model of the power system swing equation due to the term u(du/dt). Nevertheless, in the literature [24], the frequency variation is considered as being low enough as for small disturbances such that the u multiplied by du/dt is considered constant as u0. Hence, a linear perturbed form of the swing equation is achieved as follows: Ju0
dDu ¼ Pm PeS KD Du; dt
(19.11)
where PmPeS can be considered as the network power changes DP independent of the frequency changes. Hence, the linear frequency model of
Virtual inertia emulating in power electronic Chapter  19
545
the power system loadgeneration balance in the Laplace domain is calculated as follows: Du ¼
1 DP: Ju0 s þ KD
(19.12)
In this linear model, the inverse relationship between the frequency changes and the moment of inertia is also obvious. Furthermore, the s coefficient of J means its different influence on the frequency changes and RoCoF proportion to the frequency value of the disturbances. In fact, the moment of inertia J has a larger impact for larger frequencies, i.e., the term Ju0s is larger. Since the highfrequency disturbances occur in transients, the moment of inertia impacts on the transients more than the steady state. RoCoF and overshoot/undershoot (frequency nadir) are the most common indices to recognize the inertia influence on the transients, which were already explained. The following example shows these facts for a simple typical swing equation. Example 1. Consider a single SG power system with KD ¼ 300 W s/rad affected by the main damping sources such as RL loads. (a) Compare nonlinear and linear models of the swing equation for J ¼ 2 W s2/rad and DP ¼ 2000uðtÞ 2000e2ðt10Þ uðt 10Þ W. (b) Investigate the impact of inertia amount on the transients of the swing equation for J1 ¼ 0, J2 ¼ 0.5, J3 ¼ 1, J4 ¼ 2, and J5 ¼ 4 W s2/rad. Solution. Fig. 19.1A shows the power changes, which is a constant step before t ¼ 10 s and is suddenly reduced at t ¼ 10 s, then it is exponentially returned back to zero. Fig. 19.1B shows the frequency of both nonlinear and linear models of the swing equation, i.e., Eqs. (19.10) and (19.12). The frequency error of the linear model calculated as Eu ¼ 100 ðuNL uL Þ= uNL is shown in Fig. 19.1C, which can be absolutely negligible for both power changes at t ¼ 0 s and t ¼ 10 s. The frequency error is investigated for larger power changes as 10 kW. The largest frequency error is obtained as 0.1%. Therefore, the linear model is validated comparing the nonlinear model, and it is creditable for smallsignal studies. Fig. 19.2 shows the angular frequency for different moments of inertia when the damping coefficient and the input are permanent in the all situations. J1 ¼ 0 is equivalent to a power system without rotating generation units, e.g., a power electronicebased microgrid. The transients of the power changes completely appear in the frequency dynamics. Such an inertialess system suffers from the very large RoCoF and frequency overshoot/undershoot, which may trigger protection relays by mistake for small disturbances and noncritical situations, and subsequently cause frequency instability. As shown in Fig. 19.2, the RoCoF is as 13.33, 6.15, 2.88, and 1.78 rad/s2 for J1 ¼ 0, J2 ¼ 0.5, J3 ¼ 1, J4 ¼ 2, and J5 ¼ 4, respectively. Therefore, in an inertialess power electronicebased power system, one can decrease the RoCoF by designing the controller of power electronic devices to provide a virtual
546 Control of Power Electronic Converters and Systems
2000
ΔP (W)
1000 0 1000 2000 0
2
4
6
8
10
12
14
16
18
20
Time (s)
ω (rad/s)
(a) 318
Z NLM
317
Z LM
316 315 314 0
2
4
6
8
10
12
14
16
18
20
12
14
16
18
20
Time (s)
Eω (0.001 %)
10
(b)
x 10
5 0
5 0
2
4
6
8
10
Time (s)
(c) FIGURE 19.1 Comparison between linear and nonlinear models of the swing equation. (A) Active power changes, (B) angular frequency changes, (C) relative frequency error.
320
13.33 rad/s2
ω (rad/s)
318 316 314
6.15 rad/s2
J1=0 J =0.5 2
2.88 rad/s2 1.78 rad/s2
J3=1 J =2 4
0.38 rad/s 1.15 rad/s
312
5
4 rad/s
310 0
J =4
2
4
6
8
10
12
14
16
18
20
Time (s) FIGURE 19.2 The impact of inertia amount on the transients of the nonlinear swing equation for J1 ¼ 0, J2 ¼ 0.5, J3 ¼ 1, J4 ¼ 2, and J5 ¼ 4.
Virtual inertia emulating in power electronic Chapter  19
547
inertia. Obviously, a larger (virtual) moment of inertia leads to a lower RoCoF. The second span of the frequency response from t ¼ 10 s is against to the transient power change. For this span, the frequency undershoot/nadir is shown in Fig. 19.2, which decreases when the moment of inertia is reduced. It was already discussed based on Eq. (19.6). Both low RoCoF and low frequency nadir are vital during transients for power systems. However, they cannot be achieved for inertialess power electronicebased power systems without applying virtual inertia control algorithms and schemes.
19.3 Inertia challenges of power electronicebased power systems Installing renewable energy sources like wind and solar power units to power systems is becoming bulky, such that the total installed capacity of wind energy was reached 592 GW by 2018, and it is expected to be added more than 300 GW until 2023 [41]. The same growth for solar power through PV technologies are expected, since by the end of 2018 the total installed PV capacity was reached 509 GW [42]. These show a transition and an evolution for the energy paradigms, which needs a comprehensive remodeling and dynamic analysis of the modern power electronicedominated power systems in a multitimescale manner. Transition from rotating massebased SGs toward power electronic interfaces, which are employed to converse renewable energies in a more controllable way, is becoming more and more important and challengeable. This transition leads to a power electronicedominant power system with no or few rotating massebased SGs. As mentioned before, it leads to a lowinertia power electronicebased power system, which presents a stochastic and timedependent inertial feature as they are weather dependent. The inertial features for these systems are expected to be a function of the expected wind power, solar power, and ESS’s state of charge among all. It means that the system’s inertia is going to be a stochastic variable and mainly dependent on weather conditions. In addition, sometimes, situations may be emerged in which a required inertial support or an acceptable inertial response cannot be provided properly, as discussed in technical reports [25e27]. Possible solutions to provide a sufficient level to satisfy the inertial requirements are needed not only remodeling through a timedependent stochastic manner but also considering the prime mover’s type and capacity regarding to the employed device, its energy form, and its possible costs/limitations. Selfsynchronization of SGs through the grid, their kinetic energy as safeguard against gridside disturbance, and their wellknown dynamics and control is going to be replaced with no inherent synchronization through the grid and no (a sufficient) energy storage as inertial support of power electronic interfaces. In confront, inertial emulation in power electronicebased power systems has not been matured enough. Some possible forms and techniques to emulate the inertial response are summed up in Table 19.1.
548 Control of Power Electronic Converters and Systems
TABLE 19.1 Inertia emulating methods in power electronicebased power systems. Emerging technology
Inertial type 0
Being a function of 0
Energy form
Main challenges
Energy storages
ESS capacity and charge and discharge rate
ESS capacity and SoC
Electrical
Costly and need appropriate planning and control studies.
Wind generation
Rotating mass (depending to the type of wind turbine)
The mass, speed, and shape of the turbine’s rotor, and the MPP
Kinetic
Based on the wind turbine types the emulation complication changes. For example, for a Type IV wind turbine, inertial emulation needs more complicated control approach regarding a Type I wind turbine and an MPP algorithm.
Solar generation
Adjusting the operating point on the voltage epower curve
MPP
Electrical
Coordinated control for ESS and DClink capacitor for the inertial provision, balancing the PV output power (by an MPP).
In addition, Table 19.2 sums up technical aspects to be considered for inertial proposes in both the planning and operating steps. These issues can be employed to find a better solution for concerns like which storage solution is better for (inductive/resistive) weak and very weak grids, how much capacity is needed for inertial support? What is the dynamic effect of ESSs on the inertial response during contingencies? How can designers find the best position of ESSs for inertial purposes? Which required control and coordination of ESSs can be done for inertial support? A number of control solutions for different types of gridconnected systems are given in Table 19.3. Possible ways to emulate the virtual inertia for gridfeeding (in electromagnetic timescale) and gridsupporting and gridforming (in electromechanical timescale) converters are further summarized.
Virtual inertia emulating in power electronic Chapter  19
549
TABLE 19.2 Challenges for inertia emulating in power electronicebased power systems. Planning challenges
Control challenges
Cost evaluation
Capital costs, repair and maintenance costs, power losses [28]
ESS type
Flywheels, supercapacitors, LiIon batteries, etc. [29]
Sizing issues
Arbitrary, simulation based, probabilistic based [30,31]
Market design
Considering RoCoF and nadir constraints for frequency control in market designs [32,33]
Placement considerations
Power limit of converters, SoC, and line capacity constraints [34]
Adaptive and robust manners [14e16] Timedelay considerations [35] Saturation drawbacks for power injection in inner loops [36]
TABLE 19.3 Control solutions for inertia emulation. Gridtied type
Operation mode
Inertia realization
Grid feeding
Current source converters
Viable by mapping the PLL such as swing equation [37] and [44]
Grid supporting
Synchronous condensers
Secondorder nonlinear swing [38]
Grid forming
First orders: Power synchronization control [4], droop control based [39], etc.
Viable by augmenting a lowpass filter
Second orders: VSG, synchronous power control [40], synchronverters [5], etc.
Secondorder linear swing equation
19.4 Adaptive inertia for gridconnected VSGs The circuit topology of a gridconnected VSG with associated setup equipment is shown in Fig. 19.3. As it can be seen, a threephase inverter converses the DCside power to the AC side, like a Type IV wind turbine system. An ideal voltage source in series with the impedance Zg is considered as the grid’s model. An LCL filter is incorporated to reduce the switching ripples of the voltage and currents of the converter. Current control loop, as the first control
550 Control of Power Electronic Converters and Systems
FIGURE 19.3 The overall system of a gridforming gridtied virtual synchronous generator connected to an external Thevenin modeled grid (circuit diagram and setup equipment).
loop, tracks the reference currents generated by a virtual impedance loop. Moreover, it provides also fault ridethrough capability by limiting the converter current during transients. The virtual impedance loop, which is an emulation of the SG’s stator dynamics, is also employed to generate the reference currents, and limit the inrush fault currents as well [43].
19.4.1 VSG principle Mechanical features of the SG rotor is emulated through the rotor inertia and damping behavior by forming the active power balance and the virtual rotor angle correlation in the active power loop. In the same way, reactive power loop tunes the voltage amplitude of the VSG. A control block diagram for a typical VSG is shown in Fig. 19.4. According to the active and reactive power setpoints Pset and Qset , the governor and droop controllers adjust the active and reactive power references for the active and reactive power loops, as like as the governor and automatic voltage regulator (AVR) units are doing, respectively. For instance, the active power loop adjusts the frequency, based on the active power changes from the droop controller with a defined moment of inertia and a damping coefficient. The Q Vpcc AVR and P uvsg governor droops can be expressed as follows: P ¼ Pset þ ðu0 uvsg Þkgov ; Q ¼ Qset þ ðE0 Vpcc Þkavr ;
(19.13)
where u0 and E0 are the nominal frequency and voltage amplitude, respectively. uvsg is the generated virtual frequency by the active power loop, Vpcc is
Virtual inertia emulating in power electronic Chapter  19
551
abc/DE
Esi n(Z v sgt+ T 0 )
Vi rtual Impedance
C urre nt C ont rol le r
PW M
DE /abc
Rea ct iv e Po we r Lo op
( E*
Tv sg
1 s
ꞷv sg
Q
1 s
Qv dro op
Ps et
*
P
1 JZ 0
P
D
Z0
Qs et E0
Q*
1 Kvs
Z0 PZ d roop
Act iv e Po w er Lo op
FIGURE 19.4 Control block diagram of the studied virtual synchronous generator implemented in the system in Fig. 19.3.
the root mean square value of the PCC voltage, and kgov , kavr are the constant droop gains for the governor and the AVR loops, respectively. The governor and AVR droop gains are calculated by considering the operational constraints as kgov ¼
Pmax ; P droop% u0 100
kavr ¼
Qmax ; Q droop% E0 100
(19.14)
where Pdroop% and Qdroop% are constant droop rates. In the active power control loop, the swing equation, which describes the electromechanical dynamics of an SG, can be realized for a gridconnected VSG as J
d 2 qvsg ¼ ðT Tem Þ Dðwvsg w0 Þ; dt2
(19.15)
where J stands for the virtual inertia, D denotes the damping constant, T ¼ P =u0 means the governor torque, and Tem ¼ Pe =u0 is the electromechanical torque. In the same way, the reactive power control loop dynamics is implemented as Kv
dE ¼ ðQs Qe Þ kavr ðVpcc E0 Þ; dt
(19.16)
where Kv is the integrator gain for the voltage control loop. Pe and Qe stand for the active and reactive powers transferred from the VSG toward the grid side, which can be expressed as Pe ¼ E2 G þ EVg B sin d EVg G cos d;
(19.17) Qe ¼ E2 B EVg B cos d EVg G sin d; where G ¼ rt rt2 þx2t , and B ¼ xt rt2 þx2t . The xt and rt are the total inductance and resistance of the connection impedance between the VSG and the grid side, respectively.
19.4.2 Adaptive inertia The value of the virtual inertia J together with the damping constant D determines the time constant of the VSG unit. How to select the proper values for
552 Control of Power Electronic Converters and Systems
them is a challenging issue without a straightforward routine for the weak grids. Mimicking a synchronous machine, J can be described as J¼
2HSbase ; u20
(19.18)
where H is the machine inertia constant based on second, Sbase is the machine base power, and u0 is the nominal frequency of the system. The parameter H states that for which period of time the machine is able to inject its nominal power toward the grid based merely on the stored energy in the rotating mass. The higher H results in a slower response but smaller frequency deviation after a change or disturbance. However, it depends on the machine size and power, for a typical SG, H varies between 2 and 10 s. In addition to gridforming VSGs with a constant value for the moment of inertia, alternating inertia based on a simple bangbang approach can be proposed. To this end, considering the power angle curve shown in Fig. 19.5, with its initial operating point a0. After a disturbance in the system, the operating point moves along the curve between a1 and a3. The machine mode during each segment of the curve has been summarized in Table 19.4, which shows the acceleration or deceleration mode. From Eq. (19.15), it can be concluded that the rate of acceleration or deceleration du dt has a reverse relation 1. According to this, selecting a large value for J during the with J, i.e., du f dt J acceleration modes, and a small value during deceleration modes, is recommended [16]. However, in this chapter, inspired from the alternative based bangbang inertia, we have employed a weightingbased alternative inertia as du J ¼ J0 þ J1 $sgnðDuÞ$sgn (19.19) $k$ðu u0 Þ2 ; dt where multiplication of sgnðDuÞ$sgn du shows the acceleration or deceldt
Pelec (kW)
eration mode, and the weighting factor k shows the effective weighting gain for correcting the moment of the inertia J for smaller deviation.
Accelerating Area
Pmech
Decelerating Area D
D
D
D D
0
0
G
G
Grad
G
Gcrit
FIGURE 19.5 Powerangle curve of a typical synchronous generator to detect acceleration and deceleration modes: blue (black in print version) curve for before disturbance, and orange (gray in print version) curve for during disturbance.
Virtual inertia emulating in power electronic Chapter  19
553
TABLE 19.4 Alternative J determining for different machine modes illustrated in Fig. 19.5. Segment
sgn(Du)
sgn(du dt )
Mode
Alternative J
a1a2
D
D
Accelerating
Big J
a2a3
D
e
Decelerating
Small J
a3a2
e
e
Accelerating
Big J
a2a1
e
D
Decelerating
Small J
19.5 Simulation and experimental results In this section, both simulation and experimental results are presented to demonstrate the proposed adaptive inertia control method of the VSG. A comparison with the existing methods is also given.
19.5.1 Simulation results Simulations based on Simulink models in Matlab software are performed to evaluate the proposed strategy for the case study which its single line diagram is shown in Fig. 19.3. A Type IV wind turbine configuration consisting of an SG using a back to back converter interfaced with the grid through an LCL filter has been considered. Parameters used for the simulation and experimental verification of the system are shown in Table 19.5. As the turbineside converter is tightly regulated with its independent control objectives and has been separated by a DC link, the gridside converter can be considered as a decoupled system. Therefore, only a gridside converter with a constant TABLE 19.5 Test system parameters for the control technique shown in Fig. 19.4. Parameter
Value
p.u.
Parameter
Value
p.u.
Pset
1.3 kW
1.0
Qset
0
0.0
E0
110 V
1.0
L1, L2
3 mH
0.105
Vg
110 V
1.0
Cf
10 mF
35.44
u0
314 rad/s
1.0
Zg
jug 0:01 U
0.37
kgov
Pset =0:06u0 W.s/ rad
0.06
kavr
Qmax =0:04Vg Var/v
0.04
J
6 kg m2
D
10 rad/s.W
554 Control of Power Electronic Converters and Systems
DCvoltage source is considered for the analysis and some comparisons are performed throughout this section. The proposed adaptive inertia is compared to the constant inertia, which represents a comprehensive and implementable inertia response for a gridforming VSG converter during fault conditions. To assess the effect of the inertia control on the stability of the studied systems, a symmetrical threephase voltage sag which reduces 10% nominal voltage magnitude and the duration of 0.15 s was applied from the grid side, and the performance of the system is analyzed. The reference power and damping coefficient of the VSG were 1 p.u. and 10 rad/s.W, respectively, and a fixed inertia factor of 6 kg m2 is applied. Fig. 19.6 shows the PCC voltages, threephase injected currents, active and reactive powers, the fixed moment of the inertia J ¼ 6 kg m2, and the VSG’s frequency. The threephase currents are limited to 20% more than their rated value 1 p.u. during grid faults where the converter should be saturated in order to be in a safe operation. During the voltage sag, the gridforming converters show of course their grid supporting feature by injecting the reactive power to
V (p.u.)
PCC voltages Fault starts
1
Fault ends
0 1
I (p.u.)
Threephase Currents 1 0
P & Q (p.u.)
1 Active and Reactive Powers
1.5 1.0 0.5 0.0
Inertia Constant
J (kg.m2)
7 6 5
f (Hz)
Frequency 50.1 50.0 49.9 2.0
2.2
2.4
FIGURE 19.6 PCC voltages, threephase currents, active and reactive powers, constant moment of inertia of VSG with fixed J ¼ 6 kg m2, and frequency waveforms during a gridside fault to the system shown in Fig. 19.3.
Virtual inertia emulating in power electronic Chapter  19
Fault starts
V (p.u.)
1
Fault ends
555
PCC voltages
0 1
I (p.u.)
Threephase Currents 1 0
P & Q (p.u.)
1.5
J (kg.m2)
1
7.0 6.0 5.0
Active and Reactive Powers
1.0 0.5 0.0 Inertia Constant
Frequency
f (Hz)
50.1 50.0 49.9 2.0
2.2
2.4
FIGURE 19.7 PCC voltages, threephase currents, active and reactive powers, where the adaptive moment of inertia of virtual synchronous generator is a weighting function of frequency error, and frequency waveforms during a gridside fault for the system shown in Fig. 19.3.
support the voltage as it can be observed. Consequently, active power injection will be reduced due to Vg reduction in Eq. (19.19), and the VSG frequency will have a jump, and for preventing the RoCoF relays in the system from malfunctioning an inertial support is needed. The same test is simulated using the proposed structure Eq. (19.19) as it can be perceived in Fig. 19.7. For the proposed case, a better inertial response is obtained, as can be observed. In the proposed method, the moment of the inertia equipped with a weighting function of the frequency error which regarding to the machine mode leads to a better response. In this regard, a comparison of frequency responses has, realized by different methods, has been shown in Fig. 19.8.
19.5.2 Experimental results To further verify the proposed adaptive inertia control, simulated cases are tested experimentally in the laboratory setup as shown in Fig. 19.3. A Regatron DC power source is used as the DC power supply with desired DC value of
556 Control of Power Electronic Converters and Systems J = 0.5 (kg.m2)
J = 6 (kg.m2)
BangBang inertia
Adaptive inertia
50.15 50.1
f (Hz)
50.05 50 49.95 49.9 49.85 1.95
2
2.05
2.1
2.15
2.2
2.25
Time (seconds)
FIGURE 19.8 A comparison of frequency responses for different virtual inertia realizations.
650 V. The converter under test is the gridtied converter, which is a Danfoss FC302 2.2 kVA frequency inverter. The PCC voltages, converter currents, active and reactive powers, moment of inertia, and frequency are being measured. Between the LCL filter and Chroma 61845 (grid simulator), a line impedance is inserted. The grid simulator is specifically programmed to emulate the grid fault, i.e., voltage sag, by directly controlling the threephase voltages at its output. By this, at the fault instant of interest, the amplitudes of the threephase voltages are reduced to 0.9 pu for 0.16 ms as shown in Fig. 19.9. From the measurements, the actual programming, control, and data acquisition are processed through a dSPACE expansion box consisting of a DS1007 PPC processor board for code compiling, a DS5101 digital waveform output board for PWM pulse signal generation, and a DS2004 highspeed A/D board for measuring of the currents and voltages. The parameter values for the setup and control can be seen in Table 19.5 respectively. The response for constant J and adaptive J are depicted in Figs. 19.10 and 19.11. The experimental results are in good agreement with the simulation results where it is clear that appropriate inertia solutions can provide properly sufficient inertial support. As it can be seen from Figs. 19.10 and 19.11, a significant role of inertial control in RoCoF and nadir values can be observed, where for a constant j ¼ 6 kg.m2 these values are 10 Hz/s and 50.16 Hz, respectively. For the
FIGURE 19.9 Experimental Results: the gridside threephase PCC voltages during a voltage sag of 10%.
Virtual inertia emulating in power electronic Chapter  19
557
FIGURE 19.10 Fixed value moment of inertia of the virtual synchronous generator and the frequency response during a low voltage ridethrough.
FIGURE 19.11 The proposed adaptive moment of inertia and the frequency response control during a low voltage ridethrough.
proposed adaptive control using a weighting function of frequency changes, the results as shown provide a RoCoF equals to 2 Hz/s and a nadir equals to 49.95 Hz.
19.6 Summary In this chapter, after discussing the inertia concept, the key role of inertial control in future modern power electronicebased power systems have been mentioned. Ways to provide inertial requirements, device and current technologies, gridtied topologies, design considerations, planning concerns,
558 Control of Power Electronic Converters and Systems
market and operation challenges, and control solutions in power electronice based power systems have been reviewed and further summarized. A promising role of advanced virtual inertia control approaches, which may present better inertial response, have been investigated, not only by simulations studies but also by experimental test. Experiments show good agreements with the simulation studies to demonstrate how an improved virtual inertia can be obtained.
References [1] J. Driesen, K. Visscher, Virtual synchronous generators, IEEE Power Energy Soc. General MeetingConv. and Del. of Electr. Energy the 21st Century (2008) 1e3. [2] H. Bevrani, T. Ise, Y. Miura, Virtual synchronous generators: a survey and new perspectives, Int. J. Electr. Power Energy Syst. 54 (2014) 244e254. [3] A. Karimi, Y. Khayat, M. Naderi, T. Dragicevic, R. Mirzaei, F. Blaabjerg, H. Bevrani, Inertia response improvement in microgrids: a fuzzybased virtual synchronous generator control, IEEE Trans. Power Electron. (2019), https://doi.org/10.1109/TPEL.2019.2937397. [4] L. Zhang, L. Harnefors, H.P. Nee, Powersynchronization control of gridconnected voltagesource converters, IEEE Trans. Power Syst. 25 (2) (2009) 809e820. [5] Q.C. Zhong, G. Weiss, Synchronverters: inverters that mimic synchronous generators, IEEE Trans. Ind. Electron. 58 (4) (2010) 1259e1267. [6] M. Ashabani, M.Y. AbdelRady, A flexible control strategy for gridconnected and islanded microgrids with enhanced stability using nonlinear microgrid stabilizer, IEEE Trans. Smart Grid 3 (3) (2012) 1291e1301. [7] P.L. Nguyen, Q.C. Zhong, F. Blaabjerg, J.M. Guerrero, “Synchronverterbased operation of STATCOM to mimic synchronous condensers, 7th IEEE Conf. Ind. Electron. and Appl. (ICIEA) (2012) 942e947. [8] S. Dong, Y.C. Chen, Adjusting synchronverter dynamic response speed via damping correction loop, IEEE Trans. Energy Convers. 32 (2) (2016) 608e619. [9] S.A. Khajehoddin, M. KarimiGhartemani, M. Ebrahimi, Grid supporting inverters with improved dynamics, IEEE Trans. Ind. Electron. 66 (5) (May 2019) 3655e3667. [10] J. Liu, Y. Miura, H. Bevrani, T. Ise, Enhanced virtual synchronous generator control for parallel inverters in microgrids, IEEE Trans. Smart Grid 8 (5) (September, 2017) 2268e2277. [11] H. Wu, et al., Smallsignal modeling and parameters design for virtual synchronous generators, IEEE Trans. Ind. Electron. 63 (7) (July, 2016) 4292e4303. [12] H. Xu, X. Zhang, F. Liu, R. Shi, C. Yu, R. Cao, A reactive power sharing strategy of VSG based on virtual capacitor algorithm, IEEE Trans. Ind. Electron. 64 (9) (September, 2017) 7520e7531. [13] U. Markovic, Z. Chu, P. Aristidou, G. HugGlanzmann, LQRbased adaptive virtual synchronous machine for power systems with high inverter penetration, IEEE Trans. Sustain. Energy 10 (3) (July, 2019) 1501e1512. [14] D. Li, Q. Zhu, S. Lin, X. Bian, A selfadaptive inertia and damping combination control of VSG to support frequency stability, IEEE Trans. Energy Convers. 32 (1) (March, 2017) 397e398.
Virtual inertia emulating in power electronic Chapter  19 [15]
[16]
[17] [18] [19] [20]
[21]
[22] [23] [24] [25] [26] [27] [28] [29] [30] [31]
[32]
[33]
[34]
[35]
559
M.A. Torres L, L.A. Lopes, L.A. Moran T, J.R. Espinoza C, Selftuning virtual synchronous machine: a control strategy for energy storage systems to support dynamic frequency control, IEEE Trans. Energy Convers. 29 (4) (December, 2014) 833e840. J. Alipoor, Y. Miura, T. Ise, Power system stabilization using virtual synchronous generator with alternating moment of inertia, IEEE J. Emerg. Sel. Topics Power Electron. 3 (2) (June, 2015) 451e458. J. Alipoor, T. Ise, Y. Miura, Stability assessment and optimization methods for microgrid with multiple VSG units, IEEE Trans. Smart Grid 9 (2) (March, 2018) 1462e1471. N. Soni, S. Doolla, M.C. Chandorkar, Improvement of transient response in microgrids using virtual inertia, IEEE Trans. Power Deliv. 28 (3) (July, 2013) 1830e1838. D.K. Dheer, N. Soni, S. Doolla, Improvement of small signal stability margin and transient response in inverterdominated microgrids, Sustain. Energy, Grids Netw. 5 (2016) 135e147. S. Zhang, Y. Mishra, M. Shahidehpour, Fuzzylogic based frequency controller for wind farms augmented with energy storage systems, IEEE Trans. Power Syst. 31 (2) (March, 2015) 1595e1603. A. Karimi, Y. Khayat, M. Naderi, R. Mirzaee, H. Bevrani, Improving transient performance of VSG based microgrids by virtual facts’functions, in: Proc. IEEE Smart Grid Conf., March, 2017, pp. 1e6. H. Bevrani, Robust Power System Frequency Control, second ed., Springer, New York, NY, USA, 2014. H. Bevrani, B. Franc¸ois, T. Ise, Microgrid Dynamics and Control, Wiley, Hoboken, NJ, USA, 2017. J.J. Grainger, W.D. Stevenson, Power System Analysis, McGrawHill, 1994. Global Wind Energy Council, Global Wind Report 2018,” GEWC, Tech. Rep, Brussels, Belgium, April, 2019. Solar Power Europe, Global Market Outlook for Solar Power 2019 2023, Tech. Rep., Brussels, Belgium, May, 2019. N. Miller, D. Lew, R. Piwko, Technology capabilities for fast frequency response, GE Energy Consulting, Tech. Rep. (2017). F. Statnett, Challenges and opportunities for the nordic power system, Tech. Rep. (2016). ERCOT, Future ancillary services in ERCOT, Tech. Rep. (2013). H. Thiesen, C. Jauch, A. Gloe, Design of a system substituting today’s inherent inertia in the European continental synchronous area, Energies 9 (8) (July, 2016) 582. G. Delille, B. Francois, G. Malarange, Dynamic frequency control support by energy storage to reduce the impact of wind and solar generation on isolated power system’s inertia, IEEE Trans. Sustain. Energy 3 (4) (October, 2012) 931e939. V. Knap, S.K. Chaudhary, D.I. Stroe, M. Swierczynski, B.I. Craciun, R. Teodorescu, Sizing of an energy storage system for grid inertial response and primary frequency reserve, IEEE Trans. Power Syst. 31 (5) (September, 2016) 3447e3456. S. Wogrin, D.F. Gayme, Optimizing storage siting, sizing, and technology portfolios in transmissionconstrained networks, IEEE Trans. Power Syst. 30 (6) (November, 2015) 3304e3313. E. Ela, V. Gevorgian, A. Tuohy, B. Kirby, M. Milligan, M. O’Malley, Market designs for the primary frequency response ancillary servicedpart I: motivation and design, IEEE Trans. Power Syst. 29 (1) (January, 2014) 421e431. E. Ela, V. Gevorgian, A. Tuohy, B. Kirby, M. Milligan, M. O’Malley, Market designs for the primary frequency response ancillary servicedpart II: case studies, IEEE Trans. Power Syst. 29 (1) (January, 2014) 432e440.
560 Control of Power Electronic Converters and Systems [36] B.K. Poolla, S. Bolognani, F. Dorfler, Optimal placement of virtual inertia in power grids, IEEE Trans. Automat. Contr. 62 (12) (December, 2017) 6209e6220. [37] A.F. Hoke, M. Shirazi, S. Chakraborty, E. Muljadi, D. Maksimovic, Rapid active power control of photovoltaic systems for grid frequency support, IEEE J. Emerg. Sel. Topics Power Electron. 5 (3) (September, 2017) 1154e1163. [38] M.G. Taul, X. Wang, P. Davari, F. Blaabjerg, Current limiting control with enhanced dynamics of gridforming converters during fault conditions, IEEE Trans. Emerg. Sel. Topics Power Electron. 8 (2) (2019) 1062e1073. [39] Q. Hu, J. Hu, H. Yuan, H. Tang, Y. Li, Synchronizing stability of dfigbased wind turbines attached to weak ac grid, in: Proc. IEEE ICEMS, October, 2014, pp. 2618e2624. ¨ ngquist, and H. P. Nee. Frequency response [40] H. Zhang, J. P. Hasler, N. Johansson, L. A improvement with synchronous condenser and power electronics converters. In 2017 IEEE 3rd International Future Energy Electronics Conference and ECCE Asia (IFEEC 2017ECCE Asia) (pp. 10021007). IEEE. [41] Y. Khayat, R. Heydari, et al., On the secondary control architectures of AC microgrids: an overview, IEEE Trans. Power Electron. 35 (6) (2020) 6482e6500. [42] P. Rodriguez, I. Candela, A. Luna, Control of pv generation systems using the synchronous power controller, in: Proc. IEEE ECCE, September, 2013, pp. 993e998. [43] X. Lu, J. Wang, J.M. Guerrero, D. Zhao, Virtualimpedancebased fault current limiters for inverter dominated ac microgrids, IEEE Trans. Smart Grid 9 (3) (2016) 1599e1612. [44] Y. Khayat, S. Golestan, J. Guerrero, J.C. Vasquez, H. Bevrani, DClink voltage control aided for the inertial support during severe faults in weak grids, IEEE J. Emerg. Sel. Topics Power Electron. (2020), https://doi.org/10.1109/JESTPE.2020.3033657.
Further reading [1] H. Beck, R. Hesse, Virtual synchronous machine, 9th IEEE Int. Conf. Electr. Power Qual. Utilisat. (2007) 1e6.
Chapter 20
Abnormal operation of wind turbine systems Dao Zhou, Mads Graungaard Taul Department of Energy Technology, Aalborg University, Aalborg, Denmark
20.1 Introduction This chapter starts with the various grid fault types in the power system and the evolution progress of wind turbine systems. Aligned with modern grid codes on lowvoltage ridethrough, Type III and Type IV wind turbine configurations are preferred nowadays. In the case of the doubly fed induction generator (DFIG)based wind turbine system (Type III), due to the direct link between the generator stator and power grid, the power faults introduce the transient stator flux, which may cause the rotor overvoltage. Consequently, the control scheme of the rotorside converter (RSC) is comprehensively addressed subject to symmetrical and asymmetrical faults. In respect to the permanent magnet synchronous generatorebased wind turbine system (Type IV), as the power grid is fully decoupled from the generator, the control scheme of the gridside converter (GSC) is in focus. Different controller strategies for both symmetrical and asymmetrical faults are thoroughly studied and tested. To that end, a control method that injects asymmetrical converter current in compliance with recent grid code requirements is demonstrated.
20.1.1 Classification of grid faults Grid faults in power systems are classified into symmetrical faults (threephase fault 3Fg) and asymmetrical faults (singlephase fault Fg, phasetophase fault 2F, and twophase fault 2Fg) [1,2]. Compared to symmetrical grid faults, an asymmetrical grid fault contains a negativesequence component and possibly a zerosequence component. A typical configuration between the power grid and a wind farm is shown in Fig. 20.1. T1 represents the stepup transformer in order to increase the low voltage of wind turbine terminals to a medium voltage (Bus1) used in the wind farm collector grid, which is typically located in the nacelle of the wind turbine. Then, a large collector stepup transformer T2, sized at the nominal Control of Power Electronic Converters and Systems. https://doi.org/10.1016/B9780128194324.000044 Copyright © 2021 Elsevier Ltd. All rights reserved.
561
562 Control of Power Electronic Converters and Systems
FIGURE 20.1 A typical configuration for a wind farm which is linked to the power transmission grid.
capacity of the wind farm, is used to increase the collection grid voltage to the transmissionlevel voltage (PCC). Due to the existence of a Y/d transformer in the power transmission system, the zerosequence component of the voltage circulates within the delta transformer and cannot be propagated, so the type of the asymmetrical grid fault may be changed remarkably. A short circuit fault happens typically at the terminals of the transmission line. Depending on the source impedance Zs and the impedance between the point of common coupling (PCC) and fault location Zf, the dip level pdip can be defined as follows: pdip ¼
Zs Zf þ Zs
(20.1)
For simplicity, only the parasitic resistance is considered in the transmission line, which avoids the phase angle jump between the grid and PCC voltage. It can be seen that the dip level can be varied from 0 to 1 because of the fault distance to the PCC. Assuming that various asymmetrical grid faults may occur at the same location, the calculation of the voltage at the turbine terminal will be different [3]. With various dip levels, the positive, negative, and zero components of the voltages at PCC and Bus1 are listed and summarized in Table 20.1. As shown in Fig. 20.2, it can be seen that fault type can be changed through the Y/d transformer. Moreover, it is noted that the positivesequence component remains the same, while the negative component becomes inverse. It is evident that the 2F fault introduces the highest negative component.
20.1.2 Grid code requirements on lowvoltage ridethrough Wind turbines and large wind power plants are subject to grid code requirements listing needed functionalities in order to be connected to the grid. Most countries have their own set of requirements, designed for the specific power system needs, and may differ among different countries and transmission system operators [4]. Despite this, most grid codes include requirements on the following: extended operation range of grid frequency and
Abnormal operation of wind turbine systems Chapter  20
563
TABLE 20.1 Positive (Pos.), negative (Neg.), and zero components of grid faults through Y/d transformer. Fault types PCC
Bus1
Fg
3Fg
2F
2Fg
Dip level
pdip
pdip
pdip
pdip
Classification
A
B
C
E
Pos., neg., and zero components
1pdip; 0; 0
1pdip/3; pdip/3; pdip/3
1pdip/2; pdip/2; 0
12 pdip/3; pdip/3; pdip/3
Classification
A
C
D
F
Pos., neg., and zero components
1pdip; 0; 0
1pdip/3; pdip/3; 0
1pdip/2; pdip/2; 0
12 pdip/3; pdip/3; 0
FIGURE 20.2 Classification of various fault types: Phasors of threephase voltage before (dotted) and during fault (solid) are displayed.
voltage, frequency stabilization through active power control, voltage stabilization through reactive power control, and lowvoltage ridethrough (LVRT) requirements [5]. This section will specifically address the LVRT requirements, which are to define the proper control actions of wind turbines during different grid faults in the remainder of this chapter.
564 Control of Power Electronic Converters and Systems
With wind power generators providing a significant part of the total electrical power generation, direct disconnection following a grid fault cannot be tolerated since this will imply an increased risk of a large loss of generation and decreased security of supply. To alleviate such undesired events, wind turbines must be able to ridethrough lowvoltage conditions and provide dynamic voltage support via injection of reactive current. Discussion and analysis of recurring faults and highvoltage ridethrough requirements due to capacitor energization or large load drops is beyond the scope of this chapter. LVRT capability is defined by a voltage sag tolerance curve and the necessary injection of reactive current during voltage sags. For most grid codes, one single voltage sag tolerance curve is defined to both address symmetrical as well as asymmetrical faults. Other requirements as the ones specified in VDEARN4110/20 have individual requirements for symmetrical and asymmetrical faults. This is shown in Fig. 20.3A where it can be noticed that more stringent requirements on the voltage sag are present for twophase faults (2F, 2Fg) compared to threephase symmetrical faults (3Fg). Also, for highvoltage connections, ridethrough at the point of connection should be accomplished during zerovoltage conditions for a specified amount of time. Besides this, the gridconnected converter should provide dynamic voltage support through reactive current injection when the voltage deviates from its nominal value, as shown in Fig. 20.3B. Here, a proportionality constant of 2 is used between the reactive current component and the change in the voltage sequences. Mathematically, the required injection of reactive current is þ; Diþ; Q ¼ 2$Dv
(20.2)
where vþ ¼
V þ Vpf ; VN
v ¼ V =VN .
(20.3)
Vpf is the mean value of the prefault network voltage measured over 50 fundamental cycles, and VN is the nominal voltage. In this regard, the required response time of reactive current injection is usually limited to be below 20e30 ms. The defined voltage drop is usually measured at the point of connection of the wind turbine or wind farm and often at the point of generator connection, i.e., directly at the highvoltage transformer terminal of each individual turbine. The reactive current support curve is typically defined from the positivesequence voltage where during asymmetrical faults, a limit on the voltage swell in the nonfaulty phases is often included. However, to address that most grid faults are asymmetrical and to avert overvoltages in nonfaulty phases, the German VDEARN4110/20 grid codes require an injection of both positive and negativesequence reactive
Abnormal operation of wind turbine systems Chapter  20
FIGURE 20.3 Lowvoltage ridethrough requirements of VDEARN4110/20: (A) voltage sag tolerance curve for threephase and twophase faults, (B) reactive current provision as a function of change in voltage sequence components.
565
566 Control of Power Electronic Converters and Systems
current during asymmetrical faults [6,7]. This is evident from Fig. 20.3B where the injected current in either the positive or negative sequence is proportional to the voltage deviation in the respective sequence frame.
20.1.3 Fundamental wind turbine configurations Up until now, wind turbine configurations can be generally categorized into four concepts (Type IeIV) [8,9]. The main differences between these concepts lie in the types of generator, the controllability of generator speed, and the usage of power electronics. A Type I wind turbine configuration is shown in Fig. 20.4, which was very popular in the 1980s. The wind turbine is equipped with a squirrelcage induction generator, and a smooth grid connection can be achieved using a soft starter that consists of bidirectional thyristors. As a capacitor bank is required to compensate for the reactive power to excite the asynchronous generator, it becomes the main drawback of this concept. Besides, since the rotational speed is fixed without any controllability, the wind fluctuations are directly transferred into electrical power pulsations, which affect stability issues, especially in the case of a weak grid and need to be dealt with in the design. As presented in Fig. 20.5, the Type II wind turbine configuration emerged in the mid1990s. It introduced a variable rotor resistor and thus limited speed
FIGURE 20.4 Type Idfixed speed wind turbine with a direct grid connection.
FIGURE 20.5 Type IIdpartial variable speed wind turbine with a rotor resistor.
Abnormal operation of wind turbine systems Chapter  20
567
controllability of the generator rotor speed. A woundrotor induction generator and corresponding capacitor compensator are typically used, and the generator is directly connected to the grid through a soft starter. The improvement of this concept is that the rotational speed of the generator can be partially adjusted by changing the rotor resistance. This feature contributes to mechanical stress relief. However, the constant power dissipation in the rotor resistor is the main disadvantage. The Type III topology is shown in Fig. 20.6, which is usually based on a DFIG. It employs a power converter rated at approximately 30% of the nominal generator power in order to handle the slip power from the rotor of the generator. The power converter is connected to the rotor through slip rings and makes the rotor current as well as rotor speed under control, while the stator is directly linked to the grid. The fraction of slip power through the converter makes this concept attractive from an economic point of view. However, the main drawbacks lie in the use of slip rings, and also an additional crowbar is needed to protect the generatorside converter under grid faults. As shown in Fig. 20.7, the Type IV configuration equipped with an asynchronous or a synchronous generator is considered as a promising technology for multiMW wind turbine systems. The stator windings of the generator are connected to the grid through a fullscale power converter, which
FIGURE 20.6 Type IIIdvariable speed wind turbine with a partialscale power converter and a doubly fed induction generator.
FIGURE 20.7 Type IVdvariable speed wind turbine with a fullscale power converter.
568 Control of Power Electronic Converters and Systems
performs the reactive power compensation and also a smooth grid connection for the entire operating speed. Some variable speed wind turbine systems may be gearless by the introduction of a multipole generator. The elimination of slip rings, more straightforward gearbox, and full power controllability during the grid faults are the main advantages. However, to satisfy the power rating, the widely used approach nowadays is to implement several converter modules or power devices in parallel, which are challenging the complexity and reliability of the whole wind turbine system.
20.2 Control of type III wind t