Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
ON-LINE LOCAL LOAD MEASUREMENT BASED
VOLTAGE INSTABILITY PREDICTION
Momen Bahadornejad
B.Eng (Electrical Engineering)
M.Eng (Electrical Engineering)
A thesis submitted in partial fulfilment of the requirements for the degree of
Doctor of Philosophy
Centre for Built Environment and Engineering Research
School of Engineering Systems
Faculty of Built Environment and Engineering
Queensland University of Technology
2005
Statement of Original Authorship
The work contained in this thesis has not been previously submitted for a degree or diploma at any other higher education institution. To the best of my knowledge and belief, the thesis contains no material previously published or written by another except where due reference is made.
Signed: _____________________________
Date: ______________________________
Acknowledgement
The work on this thesis has been a challenging, inspiring and interesting
experience. I started my study facing some problems causing serious difficulties
in my concentration on the research. Therefore, I have had the opportunity to
become indebted to many people. However, there are some people I am specially
obliged.
First and foremost, I would like to express my sincerest appreciation to my
supervisor, Prof. Gerard Ledwich for his time, patience, guidance,
encouragement, and financial support. Without his support and supervision this
thesis would never be a success. I would specially acknowledge his invaluable
skills in supervising non-English background students.
Many thanks to my associate supervisor Dr Bouchra Senadji for her beneficial
advices at different stages of my PhD. My special gratitude is due to Prof John
Bell, Research Assistant Dean to the Faculty of Built Environment and
Engineering for supporting me when I was facing financial problems. I would
also like to thank Dr Chuanli Zhang for providing real data used in this thesis.
Thanks also goes to the School of Engineering staff and REE postgraduate
students for being friendly and having contribution to a good working
environment. My particular thanks goes to Mr Dan Moradian for his technical
support.
I would like to acknowledge the last year scholarship provided by the Ministry of
Science, Research and Technology of Iran and express my appreciation to the
people in charge in the Power and Water Institute of Technology (PWIT) for
providing financial support for my PhD. My special thanks goes to my fellow
colleagues in PWIT who helped me over the years of my study. I am also very
grateful to Prof Javad Farhoudi, the former Iranian Scientific Counsellor in
Australia for his continuing support. God bless them all.
I am very happy and proud of making wonderful friends in Australia who made
me and my family feel Australia as the second home. In particular I wish to
thank my best friend Mr Ghavam Nourbakhsh who was like a brother to me
during my staying in Australia. Ghavam helped me to come to Australia and
supported me to pursue my PhD. I would also like to express my sincere
gratitude to Mrs and Mr Famouri, Mrs and Mr Thomas, and Mrs and Mr
O’Connor for their openness, friendship, support and hospitality. Many thanks to
the Ashgrove State School staff and all the students and their parents who helped
my sons and gave them self-confidence and taught us how to adapt ourselves to
this new environment.
I have to express my deepest thanks to our families for their love, sincere wishes,
and continuing supports. Particularly, I would like to appreciate my brothers-in-
law, Fareid and Fardin, for their numerous and invaluable helps over the years of
my study. Their generosity and help have been an inspiration to me.
Last but above all, I would like to express my most heartfelt gratitude to my
wife, Tahmineh, and my sons, Aidin and Arvin, for understanding, supporting
and encouraging me to finish this thesis. It was in Australia that I found how
great they are and I am really blessed having this family. Tahmineh took the
whole family responsibility and did what I never could. Without her generosity,
endeavours, and devotions, I may never have been able to survive and complete
my studies. Far from home and missing their relatives, Aidin and Arvin were so
patient and their endeavours to do their best made me feel stronger. Thanks to
God for having such a wonderful family. This thesis actually belongs to them.
Dedication
This work is dedicated to:
o my principal supervisor,
o my wife and my two sons,
o my mother and my mother-in-law.
Abstract
Voltage instability is a major concern in operation of power systems and it is
well known that voltage instability and collapse have led to blackout or
abnormally low voltages in a significant part of the power system.
Consequently, tracking the proximity of the power system to an insecure voltage
condition has become an important element of any protection and control
scheme.
The expected time until instability is a critical aspect. There are a few energy
management systems including voltage stability analysis function in the real-
time environment of control centres, these are based on assumptions (such as off-
line models of the system loads) that may lead the system to an insecure
operation and/or poor utilization of the resources.
Voltage instability is driven by the load dynamics, and investigations have
shown that load restoration due to the on-load tap changer (OLTC) action is the
main cause of the voltage instability. However, the aggregate loads seen from
bulk power delivery transformers are still the most uncertain power system
components, due to the uncertainty of the participation of individual loads and
shortcomings of the present approaches in the load modeling.
In order to develop and implement a true on-line voltage stability analysis
method, the on-line accurate modeling of the higher voltage (supply system) and
the lower voltage level (aggregate load) based on the local measurements is
required.
In this research, using the changes in the load bus measured voltage and current,
novel methods are developed to estimate the supply system equivalent and to
identify load parameters. Random changes in the load voltage and current are
processed to estimate the supply system Thevenin impedance and the composite
load components are identified in a peeling process using the load bus data
changes during a large disturbance in the system. The results are then used to
anticipate a possible long-term voltage instability caused by the on-load tap
changer operation following the disturbance. Work on the standard test system is
provided to validate the proposed methods.
II
The findings in this research are expected to provide a better understanding of
the load dynamics role in the voltage stability, and improve the reliability and
economy of the system operation by making it possible to decrease uncertainty
in security margins and determine accurately the transfer limits.
Key Words
Power system stability, Voltage stability, Long-term voltage instability, Voltage
collapse, Supply system modeling, Load restoration, Composite load, Induction
motor load, Constant impedance load, Constant power load, Load modeling,
Load peeling, Load parameters estimation, On-load tap changers, System
variations, Signal processing, On-line Identification
III
Table of Contents Abstract .................................................................................................................. I Key Words ............................................................................................................II Table of Contents ................................................................................................ III Table of Figures .................................................................................................. VI List of Tables....................................................................................................... XI Table of Symbols and Abbreviations............................................................... XIV Chapter 1 ............................................................................................................... 1 Introduction ........................................................................................................... 1
1.1 Motivation ................................................................................................. 3 1.2 Objectives and contributions..................................................................... 4 1.3 Thesis structure ......................................................................................... 6 1.4 Publications ............................................................................................... 8
Chapter 2 ............................................................................................................... 9 Literature Review of Local Data Based Voltage Stability Monitoring................. 9
2.1 Introduction ............................................................................................... 9 2.2 Definitions and Classification ................................................................. 11 2.3 Maximum Load Power: Nose curves...................................................... 13 2.4 Voltage Sensitivity of Loads................................................................... 17 2.5 Load Restoration and Voltage Stability .................................................. 19
2.5.1 Induction Motors................................................................................ 21 2.5.2 On-Load Tap Changers (OLTCs) ...................................................... 23 2.5.3 Thermostatic Loads............................................................................ 27
2.6 Literature Review on On-line Voltage Stability Analysis ....................... 27 2.7 Local Data Based Voltage Stability Monitoring...................................... 29 2.8 Literature Review on Aggregate Load Modeling .................................... 34
2.8.1 Component based and measurement based load modeling................ 36 2.9 Countermeasure to Long-term Voltage Instability .................................. 39 2.10 Summary ................................................................................................. 39
Chapter 3 ............................................................................................................. 41 Correlation Based System Thevenin Impedance Estimation .............................. 41
3.1 Introduction ............................................................................................. 41 3.2 Theory of the Correlation Based System Thevenin Impedance Estimation ………………………………………………………………………….42
3.2.1 Block diagram representation ........................................................... 43 3.2.2 Using system equivalent circuit ........................................................ 45 a) Voltage in phase and out-of-phase components with current ................ 46 b) Random changes in the load bus voltage and current............................ 47 Case 1: System with Constant Thevenin Voltage, 0EΔ = ...................... 48 Case 2: Random Changes in the Thevenin Source, 0EΔ ≠ .................... 49 Case 3: System with Dynamics................................................................... 49
3.3 Algorithm of the Correlation Based System Thevenin Impedance Estimation ....................................................................................................... 50 3.4 Simulation Results .................................................................................. 51
3.4.1 Case 1 System with Constant Thevenin, no dynamics ...................... 52 3.4.2 Case 2: Random Changes in the supply system, no dynamics .......... 55 3.4.3 Effect of the electrical distance between two loads in estimation ..... 56
IV
3.4.4 Case 3a: System with dynamics, no random changes in supply system 57 3.4.5 Case 3b: System with dynamics and random changes in supply system 62 3.4.6 Case 4: Simulation with Swing Reference Bus ................................. 63
3.5 Application of the Proposed Method to Real Data................................... 66 3.6 Summary................................................................................................... 69
Chapter 4.............................................................................................................. 71 On-line Load Characterization by Sequential Peeling ........................................ 71
4.1 Introduction .............................................................................................. 71 4.2 Theory of the On-line Load Characterization Using Load Bus Data ....... 73
4.2.1 Load active power peeling.................................................................. 74 4.2.2 Induction Motor Reactive Power Estimation ..................................... 75 4.2.3 fpk , fqk and ff pq kk Ratio Evaluation Using Load Bus Data .. 80
4.2.3 Removal of the Undesired Components from the Load Bus Data ..... 81 4.2.4 Estimation other Components of the Load Reactive Power............... 83
4.3 Algorithm of the Load Characterization by Sequential Peeling............... 84 4.4 Simulation................................................................................................. 87 4.5 Summary................................................................................................... 93
Chapter 5.............................................................................................................. 95 On-line Estimation of the Remaining Time to a Long-term Voltage Instability. 95
5.1 Introduction .............................................................................................. 95 5.2 Long-term Voltage Instability Prediction Considering Constant Impedance Load and OLTC ............................................................................ 96
5.2.1 System Description............................................................................. 96 5.2.2 Time to collapse estimation using impedance matching criteria........ 99 5.2.3 Time to collapse estimation using load power changes ................... 101 5.2.4 The Algorithms................................................................................. 102 A: Algorithm Based on the Impedance Matching Criteria........................ 102 B: Algorithm Based on the Load Power Changes..................................... 103 5.2.5 Simulation........................................................................................ 103
5.3 On-Line Voltage Collapse Prediction Considering Composite Load and ON-Load Tap Changer .................................................................................. 110
5.3.1 System Description.......................................................................... 110 5.3.2 Tap initial ratio estimation............................................................... 112 5.3.3 Taps to collapse estimation.............................................................. 114 5.3.4 The Algorithm ................................................................................. 115 5.3.5 Simulation........................................................................................ 116 5.3.6 Taps to collapse estimation with different disturbance sizes .......... 117 5.3.7 Effects of the load and measurement uncertainties on the estimation 119 5.3.8 Effect of the load composition on the voltage collapse................... 121
5.4 Summary................................................................................................. 122 Chapter 6............................................................................................................ 123 Case Study: BPA Test System........................................................................... 123
6.1 Introduction ............................................................................................ 123 6.2 Test System Description......................................................................... 124 6.3 Validation of the Proposed “System Thevenin Impedance Estimation” Method........................................................................................................... 128
V
6.3.1 Supply System without Random Changes ....................................... 133 6.3.2 Supply System with Random Changes ............................................ 134 6.3.3 Statistical Evaluation of the Estimated Thevenin Impedance.......... 137
6.4 Validation of the Proposed “Load Characterization by Sequential Peeling” Method .......................................................................................................... 138
6.4.1 Statistical Evaluation of the Estimated Power Components ........... 145 6.5 Validation of the Proposed “On-line Estimation of the Time to a Long-term Voltage Instability” Methods................................................................ 148
6.5.1 Maximizing OLTC Secondary Side Voltage Criterion................... 149 6.5.2 Impedance Matching Criterion ....................................................... 152 6.5.3 Simulation Results .......................................................................... 153
6.6 Summary ................................................................................................. 160 Chapter 7 ........................................................................................................... 161 Conclusions ....................................................................................................... 161
7.1 Summary of the Results ......................................................................... 162 7.2 Future Research...................................................................................... 166
References ......................................................................................................... 169 Appendix A: One-step-ahead Prediction for Removing Dynamic Component 179 Appendix B: Frequency Relation between Buses............................................. 180 Appendix C: BPA Test System Data ................................................................ 182
VI
Table of Figures Figure 1.1 The voltage collapse causing grid separation in the blackout in
southern Sweden and eastern Denmark, September 23, 2003 [5]. ................ 3
Figure 2.1 Power transfer limits with uncertainty margins [6]............................ 10
Figure 2.2 Classification of power system stability [8]....................................... 13
Figure 2.3 Figure 2.3 Two-bus system............................................................... 13
Figure 2.4 The so-called onion surface [6] as given by equation (2.4) . ............. 14
Figure 2.5 The onion surface projected onto the PV-plane. The vertex of each
curve is the maximum loadability point of the system................................ 16
Figure 2.6 Influence of the static load characteristic on P-V curves. The load
characteristic is shown for α equal to 2, 1, 0, -0.1, -0.4 and -0.5 [61] ........ 18
Figure 2.7 Per phase equivalent circuit of induction motor ................................ 19
Figure 2.8 Function of an OLTC control system [31]......................................... 23
Figure 2.9 Generator-line –LTC system [6] ........................................................ 24
Figure 2.10 P-V curves of generator-line-LTC system [6] ................................ 25
Figure 2.11 Two-bus system ............................................................................... 29
Figure 2.12 Maximal power transfer is reached (voltage instability) when the
apparent impedance of the load bus reaches the Thevenin circle [14] ........ 32
Figure 2.13 Load power response to a voltage drop............................................ 35
Figure 2.14 Block diagram of input-output transfer function model ................. 36
Figure 3.1. Time measurement of load voltage and current magnitudes ........... 42
Figure 3.2. Block diagram representation of proposed method for system
identification................................................................................................ 43
Figure 3.3 load connected to the equivalent circuit of the system ...................... 46
Figure 3.5 Four bus test system with two different variable loads...................... 51
Figure 3.6 Simulation of load admittance changes in case 1............................... 53
Figure 3.7 Magnitudes and phases of the load voltage and current in case 1 ..... 54
Figure 3.8 Difference of the real component of the load bus voltage and its auto-
correlation in case 1..................................................................................... 54
Figure 3.9 Simulation of load admittances changes in case 2 ............................. 56
Figure 3.10. Four bus test system with dynamic load ......................................... 58
Figure 3.11 Simulation of load #1 admittance changes in case 3a...................... 58
VII
Figure 3.12 Variation of generator angle in case 3a ........................................... 59
Figure 3.13 Variation of induction motor speed in case 3a ................................ 59
Figure 3.14 Magnitudes and angles of the load bus voltage and current ............ 60
Figure 3.15. Difference of the voltage imaginary component and its auto-
correlation in case 3a................................................................................... 61
Figure 3.16. Difference of the voltage imaginary component and its auto-
correlation in case 3a after removing dynamic component ........................ 61
Figure 3.17. Simulation of load #1 and load #2 admittances in case 3b............ 62
Figure 3.18. Reference bus angle changes and its reflection in other buses ....... 63
Figure 3.19. Load bus voltage and angle after removing the effect of reference
bus rotation.................................................................................................. 64
Figure 3.20. Time measurement of load voltage and current magnitudes ......... 66
Figure 3.21. Time measurement of load voltage and current angles, angle of the
first data point in current is chosen as reference ......................................... 67
Figure 3.22. Real data load admittance magnitude and angle............................. 67
Figure 3.23. Real data voltage real component changes and its auto-correlation68
Figure 3.24. Residuals of the real data voltage real component changes and its
auto-correlation ........................................................................................... 69
Figure 4.1. One-line diagram a simple power system with composite load ....... 73
Figure 4.2 Equivalent circuit of a power system with composite load ............... 73
Figure 4.3 Active powers of different load components following a disturbance
in supply system.......................................................................................... 74
Figure 4.4 Per-phase equivalent circuit of induction motor............................... 76
Figure 4.5 Block diagram of a system………………………………………….81 Figure 4.6 Reactive power of different load components following a disturbance in supply system………………………………………………………………...83 Figure 4.7 Four bus test system………………………………………………...87 Figure 4.8 Simulation of load #2 admittance and system Thevenin impedance. 88
Figure 4.9 Simulation of load bus voltage and induction motor slip .................. 88
Figure 4.10 Simulation of load active and reactive powers ................................ 89
Figure 4.11 (a) and (b): changes in the load bus voltage and its autocorrelation
(c) and (d): load active power changes and its autocorrelation.................. 91
Figure 4.12 (a) and (b): voltage dependent load active power changes and its autocorrelation ……………………………………………………………… 91 Figure 4.13 (a) and (b): frequency dependent load active power changes and its
autocorrelation............................................................................................. 92
VIII
Figure 5.1. Simple system with OLTC and constant impedance load ................ 97
Figure 5.2. Changes in impedances, tap ratio, primary and secondary side
voltages and power ...................................................................................... 97
Figure 5.3. Changes in impedances, Load voltage and power ............................ 98
Figure 5.4. Simulation of load admittance ........................................................ 104
Figure 5.5 (a): System and load impedances, (b): Load real power, k=2.......... 105
Figure 5.6. (a): System and load impedances, (b): Load real power, k=2.5..... 105
Figure 5.7. (a): Changes in the system and load impedances,........................... 106
(b): changes in load real power, k=3.................................................................. 106
Figure 5.8. Comparison of estimated and actual values of power changes due to
tapping, k=2 ............................................................................................... 108
Figure 5.9. Comparison of estimated and actual values of power changes due to
tapping, k=2.5 ............................................................................................ 108
Figure 5.10. Comparison of estimated and actual values of power changes due to
tapping, k=3 ............................................................................................... 109
Figure 5.11 Simple power system with composite load and on load tap changer
................................................................................................................... 110
Figure 5.12 (a): tap position, primary and secondary voltages, (b): induction
motor slip, (c): load active powers, (d): load reactive powers .................. 111
Figure 5.13. Changes in the transformer secondary voltage and load power due
to taping ..................................................................................................... 112
Figure 5.14. Four bus test system with composite load..................................... 117
Figure 5.15. Simulation of the OLTC primary and secondary voltage changes,
(a): k=1.4, (b): k=2.5, (c): k=3 ............................................................... 118
Figure 5.16. Estimated OLTC secondary voltage for 10 successive tapings, k=3
................................................................................................................... 119
Figure 5.17. Time measurements of load voltage and current magnitudes in the
Brisbane load bus....................................................................................... 120
Figure 5.18. Simulation of load admittance with random changes ................... 120
Figure 5.19. Simulation of the load reactive powers, Induction motor: 60%,
Constant impedance: 35%, Constant power: 5%, k=2.9............................ 122
Figure 6.1. The BPA test system ....................................................................... 124
Figure 6.2. Overexcitation limiter characteristic ............................................... 125
Figure 6.3. Simulation of load admittance changes in bus 10........................... 127
IX
Figure 6.4. Tap ratio, magnitudes of bus 9 and bus 10 voltages....................... 127
Figure 6.5 Simulation of the magnitudes and angles of the bus 9 voltage and
current ....................................................................................................... 129
Figure 6.6 Magnitudes and angles in bus 9 voltage and current from disturbance
until start of OLTC operation................................................................... 129
Figure 6.7 Changes in the components of the bus 9 voltage and current from the
line trip disturbance until start of OLTC operation................................... 130
Figure 6.8 Auto correlations of the post-disturbance changes in the components
of the bus 9 voltage and current phasors................................................... 130
Figure 6.9 Post-disturbance changes in the load admittance magnitude and its
autocorrelation until start of OLTC operation .......................................... 131
Figure 6.10 Residuals of the post-disturbance changes in the components of the
load voltage and current until OLTC operation ........................................ 131
Figure 6.11 Autocorrelations of the residuals of the post-disturbance changes in
the load admittance magnitude until OLTC operation.............................. 132
Figure 6.12 Simulation of the magnitudes of the system Thevenin and load
impedances from the bus 9 view point...................................................... 133
Figure 6.13 Simulation of local and remote load admittances.......................... 134
Figure 6.14 Histograms of the components of the estimated Thevenin impedance
................................................................................................................... 138
Figure 6.15 Simulation of the load voltage and current magnitudes ................ 140
Figure 6.16 Simulation of the load total active and reactive powers ................ 140
Figure 6.17 Simulation of induction motor load power.................................... 141
Figure 6.18 Simulation of the constant impedance load power ........................ 141
Figure 6.19 Simulation of the constant power load .......................................... 142
Figure 6.20 Changes in the load post-disturbance active power....................... 143
Figure 6.21 Changes in the load post-disturbance reactive power.................... 143
Figure 6.22 Histograms of the estimated induction motor load active and reactive
power......................................................................................................... 146
Figure 6.23 Histograms of the estimated constant impedance load active and
reactive power ........................................................................................... 147
Figure 6.24 Histograms of the estimated constant power load active and reactive
power......................................................................................................... 147
X
Figure 6.25 Equivalent of the BPA test system with OLTC transformer and
composite load in bus 10. .......................................................................... 149
Figure 6.26 Simulation of the Thevenin voltage magnitude ............................. 151
Figure 6.27 Comparison of the estimated and actual values of the system
Thevenin impedance.................................................................................. 152
Figure 6.28. Simulation of the OLTC primary and secondary side voltages and
tap ratio in case1 ........................................................................................ 154
Figure 6.29. Simulation of the OLTC primary and secondary side voltages and
tap ratio in case2 ........................................................................................ 154
Figure 6.30. Simulation of the OLTC primary and secondary side voltages and
tap ratio in case3 ........................................................................................ 155
Figure 6.31. Estimated OLTC secondary side voltages, system Thevenin
impedance, and load impedance in case1.................................................. 157
Figure 6.32 Estimated OLTC secondary side voltages, system Thevenin
impedance, and load impedance in case2.................................................. 158
Figure 6.33 Comparison of the estimated OLTC secondary side voltages to the
actual values in case 3 ............................................................................... 159
Figure 6.34. Estimated system Thevenin impedance and load impedance and
their comparison to the actual values in case 3 ......................................... 159
Figure B.1 Three bus system ............................................................................. 180
XI
List of Tables Table 1-1 Voltage collapse incidents[4] ............................................................... 2
Table 3.1 Comparison of the estimated and actual values of the system Thevenin
impedance in case 1 .................................................................................... 55
Table 3.2 Comparison of the estimated and actual values of the system Thevenin
impedance for a simulation time, T=100 sec and a time step ΔT = 0.02 sec
and different ratios of random changes in load #2 with respect to load #1. 56
Table 3.3 Comparison of the estimated and actual values of the system Thevenin
impedance for a simulation time T=100 sec and a time step TΔ = 0.04 for
different values of 2Z ................................................................................. 56
Table 3.4 Comparison of the estimated and actual values of the system Thevenin
impedance for a T=100 sec simulation time and a time step TΔ = 0.04 in
case 3, no changes in load #2 .................................................................... 62
Table 3.5 Comparison of the estimated and actual values of the system Thevenin
impedance for a T=100 sec simulation time and a time step TΔ = 0.04 in
case 3b (random changes in load #2) .......................................................... 63
Table 3.6 Comparison of the estimated and actual values of the system Thevenin
impedance for a 100 sec simulation time and 0.04 sec time steps for
different values of rotations in reference bus (bus 1) angle, without
random changes in supply system ........................................................... 65
Table 3.7 Comparison of the estimated and actual values of the system Thevenin
impedance for a 100 sec simulation time and 0.04 sec time steps for
different values of rotations in reference bus (bus 1) angle, with random
changes in supply system .......................................................................... 65
Table 3.8. Estimated values of the system Thevenin impedance using 100 sec
successive time frames of Brisbane load centre measured voltage and
current phasors, starting at 9 am on 2002/06/06 ......................................... 69
Table 4.1 Comparison of the load estimated and actual powers for different
compositions of loads and a 10% change in the system impedance ........... 90
Table 4.2 Comparison of the load estimated and actual powers for different sizes
of the system disturbance (K) ..................................................................... 93
XII
Table 5.1. OLTC tap initial ratio, step size, lower limit, time delay and dead
band. .......................................................................................................... 104
Table 5.2. Actual and estimated system impedance, initial tap ratio and taps to
collapse for different values of k in (prefault)sk*Zt)(post faulsZ = ................... 106
Table 5.3. OLTC tap initial ratio, step size, lower limit, time delay, voltage
reference, and dead-band. .......................................................................... 117
Table 5.4. Estimated initial tap ratio and taps to collapse for different values of k
in %e)disturbanc(preZ/et)disturbanc(post Zk thth −−= ....................... 118
Table 5.5. Estimated initial tap ratio and taps to collapse for with different
disturbance sizes and random changes in the load .................................... 121
Table 5.6. Comparison of the taps to collapse for different load compositions 121
Table 6.1 Comparison of the estimated and actual values of the system Thevenin
impedance, supply system without random changes, time step=0.04 sec. 134
Table 6.2 Comparison of the estimated and actual values of the system Thevenin
impedance, supply system with random changes, Time step=0.04 sec..... 135
Table 6.3 Comparison of the estimated and actual values of the system Thevenin
impedance, supply system with random changes, Time step=0.02 sec..... 135
Table 6.4 Comparison of the estimated and actual values of the system Thevenin
impedance for different random change sizes for remote load (K), supply
system with random changes, Time step=0.02 sec.................................... 136
Table 6.5 Statistical parameters of the histograms of the components of the
estimated thevenin impedances ................................................................. 137
Table 6.6 Comparison of the load estimated and actual powers for different
compositions of loads, disturbance is the loss of one of the branches
between bus 5 and bus 6, P1 & Q1: Induction motor load, P2 & Q2:
Constant impedance load, P3&Q3: Constant power load ......................... 144
Table 6.7 Comparison of the load estimated and actual powers for different sizes
of the system disturbance. ......................................................................... 145
Table 6.8 Statistical parameters of the histograms of the components of the
estimated thevenin impedances ................................................................. 148
Table 6.9 mean values of the estimation error in the different components of the
load active and reactive power .................................................................. 148
XIII
Table 6.10 Comparison of the system post-disturbance estimated and actual
Thevenin impedance for different cases ................................................... 155
Table 6.11Comparison of the load post-disturbance and pre OLTC operation
estimated and actual power components. P1 & Q1: Induction motor load,
P2&Q2:Constant impedance load P3&Q3:Constant power
load……………………………................................................................ 156
XIV
Table of Symbols and Abbreviations OLTC On-load tap Changer
LTC Load tap changer
E Generator internal voltage
δ Voltage angle
ϕ Load impedance angle
f(.) An objective function f
s Induction motor slip
Δx Perturbation of variable x
x& The derivative of state variable
xf∂
∂ Derivative of the function f with respect to the variable x
thZ Thevenin impedance
thY Thevenin admittance
thE Thevenin voltage
thI Thevenin current
sZ System equivalent impedance
LZ Load impedance
LY Load admittance
pz An internal state variable
pT Active load recovery time constant
sα Steady-state load voltage dependence
tα Transient load voltage dependence
G(s) Transfer function
s Complex frequency variable in Laplace
w(t), d(t) white noise variables
E[] Expected value
W, D variances of w(t) and d(t) ∗
ΔI Complex conjugate of current changes
τ Time lag
)(Rxy τ Cross-correlation between processes x and y
XV
∑=
n
iix
1 Summation of the variable x from its ith to nth elements
TΔ Simulation time step
imag Imaginary component
sec Second
Auto Auto-correlation
Pu Per unit
T Total simulation time
deg Degree
11 QandP Induction motor load active and reactive power
22 QandP Constant impedance load active and reactive power
33 QandP Constant power load active and reactive power
pvK and fpK Voltage and frequency dependency coefficients of the load
active power
qvK and fqK Voltage and frequency dependency coefficients of the load
reactive power
of System nominal frequency
Absolute value
ω Frequency variable in the Fourier domain
h(t) Function in the time domain
)(H ω Function in the Fourier domain
Φ Fourier transform
EST Estimated value
ACT Actual value
ERR Estimation error
n Tap ratio, Tap position
sP VV and Primary and secondary side voltages of OLTC transformer
crn Tap position at voltage collapse point
nΔ Tap step
on Initial tap ratio
minn Tap lower limit
dT OLTC time delay
XVI
fT and mT OLTC intentional and mechanical time delays
mP Induction motor mechanical power
RV Reference voltage
FCL Full Load Current
OXL Over-excitation Limiter
diff Difference and approximate derivative
Xt OLTC transformer reactance
Chapter 1
Introduction For many decades power systems had been concerned of either angle stability or
thermal overload capabilities of lines. However, in the last three decades due to
the difficulties in the building of new transmission and generation facilities,
power systems have been loaded much more heavily than usual in the past. As a
consequence, many power systems around the world are experiencing voltage
problems leading to instabilities [1].
In normal operating conditions, voltage drops in the order of a few percent of the
nominal voltage are accepted between the generation and consumption points.
However, in some cases following a disturbance, voltages in some buses may
experience large and progressive falls. A possible outcome of voltage instability
is loss of load in an area, or tripping of transmission lines and other elements by
their protective systems leading to cascading outages[2, 3]. The term voltage
collapse is also often used. It is the process by which the sequence of events
accompanying voltage instability leads to a blackout or abnormally low voltages
in a significant part of the power system. Many voltage collapses have occurred
throughout the world as shown in Table 1.1. Investigations on the recent major
blackouts indicated that voltage instability is still one of the major causes of the
power system separation (Figure 1.1) [5]. In this event, voltage collapsed 100 sec
after the system fault and was recovered after power system separation.
Chapter 1. Introduction
2
TABLE 1-1 VOLTAGE COLLAPSE INCIDENTS[4]
Date Location Duration
13April1986
Winnipeg, Canada Nelson River HVDC link
1 Second
30Nov.1986 SE Brazil, Paraguay 2 seconds
17May1985 South Florida 4 seconds
22Aug.1987 Western Tennessee 10 seconds
27Dec.1983 Sweden 55 seconds
21May1983 Northern California 2 minutes
2Sep.1982 Florida 1-3 minutes
26Nov.1982 Florida 1-3 minutes
28Dec.1982 Florida 1-3 minutes
30Dec.1982 Florida 1-3 minutes
22Sep.1977 Jacksonville, Florida Few minutes
4Aug.1982 Belgium 4.5 minutes
20May1986 England 5 minutes
12Jan.1987 Western France 6-7 minutes
9Dec.1965 Brittany, France Unknown
10Nov.1976 Brittany, France Unknown
23July1987 Tokyo 20 minutes
19Dec.1978 France 26 minutes
22Aug.1970 Japan 30 minutes
22Sep.1970 New York State Several hours
20July1987 Illinois and Indiana Hours
11June1984 Northeast United States Hours
Voltage instability is driven by the load dynamics. In response to a disturbance,
power consumed by the loads tends to be restored by the action of the motor slip
adjustments (in seconds), tap-changing transformers (in minutes), and
thermostats (in a few hours). Load restoration causes further voltage reduction
on the high voltage network by increasing the reactive power consumption. A
run-down situation causing voltage instability occurs when the load dynamics
Chapter 1. Introduction
3
attempt to restore power consumption beyond the amount that can be provided
by the combined transmission and generation system.
Figure 1.1 The voltage collapse causing grid separation in the blackout in
southern Sweden and eastern Denmark, September 23, 2003 [5].
Although a system protection scheme may integrate and coordinate several types
of actions, action on load is the ultimate countermeasure. Shedding a proper
amount of load, at a proper place, within a proper time can be done indirectly
through a modified control of Load Tap Changers or directly as load shedding.
1.1 Motivation
This work is motivated by on-line identification of time to voltage instability,
that is a critical aspect, and the critical effect that the system and load
representation has on voltage stability studies.
Tracking the proximity of the power system to an insecure voltage condition is
an important element of any protection and control scheme. Many emergency
control measures are based on extensive off-line studies on voltage stability
modeling, computation of voltage collapse point and enhancement of power
system stability. A good overview of these areas is provided in [1, 6-9].
There are a few energy management systems including voltage stability analysis
function in the real-time environment of control centers [10]. In these methods,
Chapter 1. Introduction
4
in order to simplify the calculations, a number of approximations are used, which
introduce high or low uncertainty in the obtained transfer limits, according to the
used assumptions [1]. While, an optimistic approach may lead the system to
unacceptable values (security uncertainty), a pessimistic approach will avoid
risks by introducing larger security margins, but on the other hand it will lead to
a poor utilization of the resources (economic loss). These methods are also based
on off-line models of the system loads. The approaches to load modeling are
either component based or the measurement based, where one test is already
performed on the load [11].
The transfer limits across certain sections of the power system, depends on the
factors such as network topology, loading and generating conditions which vary
by time.
Load behavior is at the heart of voltage instability. The aggregate load seen from
bulk power delivery transformers are the most uncertain power system
components, due to the uncertainty of the participation of individual loads. In
order to develop and implement a true on-line voltage stability analysis method,
the real-time accurate modeling of the higher voltage (supply system) and lower
voltage level (aggregate load) at the load bus is required.
As far as loads restore to constant power due to the on-load tap changer (OLTC)
action, the long-term nature of phenomena together with a control on OLTC’s to
slow down the system degradation, might leave time to a computer to identify
the problem and trigger corrective actions.
1.2 Objectives and contributions
Based on the local measurements this project aimed to develop on-line methods
to predict a long term voltage instability caused by on-load tap changer. The
Chapter 1. Introduction
5
feasibility of the supply system equivalent estimation and load parameters
identification using the changes in the load bus measured voltage and current are
pointed out and then the results are used to identify a possible voltage collapse
resulting from a long-term voltage instability. The arising results may provide a
better understanding of the load dynamics role in the voltage stability, and
improve the reliability and economy of the system operation by making it
possible to decrease uncertainty in security margins and determine accurately the
transfer limits.
The main contributions of this thesis are:
• Correlation Based On-line Power System Thevenin Impedance
Estimation
A method is developed to estimate power system Thevenin impedance that is
based on signal processing on the measured data in the load bus. It is shown that
the cross-correlations of the changes in the load voltage and current with respect
to the changes in the load admittance can be used to estimate the system
Thevenin impedance. The method is validated by simulation. Any dynamic
components in the load voltage and current caused by system and/or load are
removed and the residuals are used for the estimation. Work on real data is also
provided to confirm the method.
• On-line Load Characterization by Sequential Peeling
A method is proposed that is based on the on-line measured load bus voltage and
current phasors during a disturbance. Load is considered as a combination of
induction motor, constant impedance, and constant power loads. The change in
the load active power due to the disturbance is used to identify the active power
of the load different components in a peeling process. Then the induction motor
Chapter 1. Introduction
6
reactive power is estimated using signal processing on the random changes in the
load total power. The other components of the load reactive power are then
estimated using the disturbance data.
• On-line anticipation of the Time to a Long-term Voltage Instability Caused
by the On-Load Tap Changer Operation
It is shown that the local load bus voltage and current can be used to anticipate
possible voltage instability and to estimate the taps to such instability in a system
consisting of constant impedance, induction motor, and constant power loads
behind a tap changer.
1.3 Thesis structure
Introduction (Chapter 1)
In chapter 1 an introduction to voltage stability and description of the facts that
have motivated the realization of the work held in this thesis are provided. The
novel contributions are briefly explained and the publications of the research
work are listed.
Literature Review of Local Data Based Voltage Stability Monitoring
(Chapter 2)
Chapter 2 outlines the fundamentals of voltage stability and some important
aspects such as voltage stability classification, voltage sensitivity of loads and
the role of on-load tap changers in voltage stability are highlighted. The literature
on the on-line voltage stability monitoring and load modeling is also reviewed
and discussed and the shortcomings are pointed out.
Chapter 1. Introduction
7
Correlation Based On-line System Thevenin Impedance Estimation
(Chapter 3)
Chapter 3 covers the achievements on the power system equivalent impedance
estimation from the load bus view point. It is clearly shown that the random
changes in the load bus voltage and current can be extracted and then to be used
to estimate the system Thevenin impedance.
On-line Load Characterization by Sequential Peeling (Chapter 4)
In chapter 4 it is shown that how the measured data in a load bus during a
disturbance can be used to estimate the parameters of the components of a
composite where load is considered as a combination of induction motor,
constant impedance, and constant power loads.
On-line Estimation of the Remaining Time to a Long-term Voltage
Instability (Chapter 5)
Chapter 5 is focused on the on-line identification of the expected time to voltage
instability. Using the local load bus voltage and current, the taps to a possible
voltage collapse in a system consisting of constant impedance, induction motor
and constant power loads behind a tap changer is anticipated. The proposed
method is confirmed by simulation.
Case Study: BPA Test System (Chapter 6)
In order to investigate and validate the proposed methods in this research, they
are applied to the BPA 10-bus test system. The loads are modified and complete
data of the system is provided.
Conclusions (Chapter 7)
Main conclusions and suggestions for future work are outlined.
Chapter 1. Introduction
8
1.4 Publications
Some results of this thesis have already been published in the publications
below. Chapter 3 is describing the contents of [1] and [2]. The results in [3] are
discussed in Chapter 4. The results in Chapter 5 are described in [4] and [5].
[1] M. Bahadornejad, G. Ledwich, “Studies in system Thevenin impedance
estimation from normal operational data”, proceedings of the 6th International
power Engineering Conference, IPEC 2003, 22-24 May 2003, Singapore.
[2] M. Bahadornejad, G. Ledwich, “System Thevenin impedance estimation
using signal processing on load bus data”, proceedings of the IEE Hong Kong
International Conference on Advances in Power System Control, Operation and
Management, APSCOM 2003, 11-14 Nov 2003, Hong Kong.
[3] Bahadornejad, M., and Ledwich, G., “On-line Load Characterization by
Sequential Peeling”, Presented in the 2004 International Conference on Power
System Technology, Powercon2004, 21-24 Nov 2004, Singapore.
[4] Bahadornejad, M., and Ledwich, G. , “Studies in the OLTC Effects on
Voltage Collapse Using Local Load Bus Data”, in Australasian Universities
Power Engineering Conference, AUPEC’2003, 28 Sept-1 Nov 2003,
Christchurch, New Zealand
[5] Bahadornejad, M., and Ledwich, G. , “On-line voltage Collapse Prediction
Considering Composite Load and On Load Tap Changer”, in Australasian
Universities Power Engineering Conference, AUPEC’2004, 26–29 September
2004, The University of Queensland, Brisbane, Australia
[6] Bahadornejad, M., and Ledwich, G., “Correlation Based Power System
Thevenin Impedance Estimation”, under preparation, to be submitted to the
journal “IEEE Transactions on Power Systems”
Chapter 2
Literature Review of Local Data Based Voltage Stability Monitoring
2.1 Introduction
Historically, (rotor) angle instability has been the dominant stability problem on
most power systems. As power systems have evolved, different forms of system
instability have emerged. Phenomena such as frequency stability, interarea
oscillations, and voltage stability have become great concerns to the power
system operation than in the past [8].The amount of power that can be
transferred between different parts of the power system is restricted by the
different stability limits. As an example, Figure 2.1 shows a problem concerning
stability limits [61]. These limits are generally difficult to determine with
sufficient accuracy and reliability due to the high uncertainty related to internal
and external factors, and therefore conservative criteria are often used for their
determination resulting in smaller secure operation areas. As it can be seen from
Figure 2.1 the voltage stability limit is strongly dependent on system loading.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
10
Figure 2.1 Power transfer limits with uncertainty margins [6]
In the past two decades, in the prevailing open access environment, power
systems have been operated under much more stressed conditions than was usual
in the past. As a consequence, many power systems have experienced voltage
instability problems and voltage stability has become a factor leading to limit
power transfers [10]. Environmental pressures on transmission expansion,
increased electricity consumption in heavy load areas, new system loading
patterns due to the transmission open access and maximum profit environment
are some of the responsible factors for these conditions. The unstable behaviour
is characterized by slow (or sudden) voltage drops; sometimes escalating to the
form of a collapse [6].
An important element of any protection and control scheme is to track the
closeness of the system to a collapse in real time [12]. In order to achieve the
acceptability of the true system control/protection schemes, it is necessary to
provide them with “fall-back” position or a safety net based on local data, acting
Secure Operation
Voltage Stability Limit Thermal Limit
Angle Stability Limit
Load (MW)
Power Transfer Limit (MW)
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
11
if the centralized system fails for any reason, such as a possible failure of the
necessary hardware infrastructure [13] . Thus, despite the fact that voltage
instability is a system problem, there is still a need for devices that process only
local measurement [14, 15] .These controls are low cost and simple to build, and
hence, provide an attractive option for the utility industry. They can be used to
send alarms to control centre(s) when local monitoring indicates a locally weak
condition and/or to trip locally if the more severe conditions are encountered
[16] .
In order to create a basis for the next chapters, this chapter outlines some of the
fundamentals of voltage stability and gives a description of the phenomena that
contribute to voltage instability. The literature related to the voltage stability
monitoring and control based on the local measurements is also reviewed and the
existing methods are discussed.
2.2 Definitions and Classification
“Voltage stability refers to the ability of a power system to maintain steady
voltages at all buses in the system after being subjected to a disturbance from a
given initial operating condition” [16]. It depends on the system ability to restore
equilibrium between load demand and load supply from the power system.
The term voltage collapse is also often used. It is the process by which the
sequence of events accompanying voltage instability leads to a blackout or
abnormally low voltages in a significant part of the power system [7, 17, 18].
Stability is a condition of equilibrium between opposing forces. Depending on
the network topology, system operating condition and the form of disturbance,
different sets of opposing forces may experience sustained imbalance leading to
different forms of instability. A classification of the power system stability based
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
12
on time scale is shown in Figure 2.2. While rotor angle and frequency
instabilities are generator-driven phenomena, voltage instability is load-driven. It
should be noted that these terms do not exclude the affect of the other
components to the instability mechanism.
The time scale is divided into short and long-term time scales. When short-term
dynamics are stable they eventually die out some time after disturbance, and the
system enters a slower time frame.
The time frame of angle stability is that of the electromechanical dynamics of the
power system lasting in a few seconds.
During frequency excursions, the characteristic times of the processes and
devices that are activated will range from fraction of seconds, corresponding to
the response of devices such as underfrequency load shedding and generator
controls and protections, to several minutes, corresponding to the response of
devices such as prime mover energy supply systems and load voltage regulators.
Short-term voltage stability involves dynamics of fast acting load components
such as induction motors, electronically controlled loads, and HVDC converters.
The study period of interest is in the order of several seconds, and analysis
requires solution of appropriate system differential equations; this is similar to
analysis of rotor angle stability.
Long-term voltage stability involves slower acting equipments such as tap-
changing transformers, thermostatically controlled loads, and generator current
limiters. The study period of interest may extend to several or many minutes, and
long-term simulations are required for analysis of system dynamic
performance[10, 19, 20].
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
13
Power SystemStability
TransientStability
Small-Disturbance
Angle Stability
Short Term
Rotor AngleStability
Voltage Stability
Large-Disturbance
Voltage Stability
Small-DisturbanceVoltage Stability
Short Term Long Term
FrequencyStability
Long TermShort Term
Power SystemStability
TransientStability
Small-Disturbance
Angle Stability
Short Term
Rotor AngleStability
TransientStability
Small-Disturbance
Angle Stability
Short Term
Rotor AngleStability
Voltage Stability
Large-Disturbance
Voltage Stability
Small-DisturbanceVoltage Stability
Short Term Long Term
Voltage Stability
Large-Disturbance
Voltage Stability
Small-DisturbanceVoltage Stability
Short Term Long Term
FrequencyStability
Long TermShort Term
FrequencyStability
Long TermShort Term
Figure 2.2 Classification of power system stability [8]
2.3 Maximum Load Power: Nose curves One of the primary causes of power system instability is the transmission of
(large amounts of) power over long distances. In voltage stability, attention is
paid to power transfers between generation and load centers.
In the two bus simple system of Figure 2.3, a remote load is supplied by a strong
source (infinite bus) with constant voltage E through a transmission line
modeled as a series reactance that can be the Thevenin equivalent of a power
system seen from one bus.
Figure 2.3 Figure 2.3 Two-bus system
The receiving end voltage V and angle δ depend on the active and reactive
power transmitted through the line which under balanced three-phase, steady-
State sinusoidal conditions can be written as below:
0∠= EE jX δ∠=VV
jQP +
lll jXRZ +=
I
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
14
δsinX
EVP = (2.1)
XVδcos
XEVQ
2−= (2.2)
After eliminating δ using the trigonometric identity we get
02
222
=⎟⎠⎞
⎜⎝⎛−+⎟⎟
⎠
⎞⎜⎜⎝
⎛+
XEVP
XVQ (2.3)
Solving equation (2.3) for V yields
QX
EPXEXXQEV
22
2
42
42−−±−= (2.4)
Figure 2.4 shows how the terminal voltage changes with the load powers
(dimensionless variables are used in the figure). In “normal” conditions, the
operating point lies on the upper part of the surface (corresponding to the
solution with the plus sign in equation (2.4)), with V close to E. Permanent
operation on the lower surface, characterized by a lower voltage and higher
current, is unacceptable.
Figure 2.4 The so-called onion surface as given by equation (2.4)
XV
2EQX
2EPX
020.
40.60.
80.
0
50.
1
20.
20.−
0
0
20.
40.
20.−
40.−
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
15
The condition to have at least one real solution for equation (2.4) is:
222
2
2 ⎟⎟⎠
⎞⎜⎜⎝
⎛≤+
XE
XEQP (2.5)
The above inequality determines all combinations of active and reactive power
that the system can supply to the load. The equality in equation (2.5) corresponds
to the existence of a maximum load power, well-known from circuit theory [21].
More precisely, the Figure 2.4 shows a set of maximum load power points,
located on the “equator” of the surface. The projection of this limit curve onto
the (P,Q) plane is the parabola shown in Figure 2.4. In the (P,Q) load power
space, this parabola bounds the region where operation is feasible. All points
inside the parabola lead to two solutions for the terminal voltage V. For low
loading there are two equilibrium solutions; one with high voltage and the other
with low. The former is the stable equilibrium point(s.e.p.), and the latter one is
the unstable equilibrium point(u.e.p.) These equilibrium points approach each
other as the system is loaded slowly, up to the point where the two solutions in
(2.4) coalesce, i.e., the inner square root vanishes [22]. If the system is loaded
further, all system equilibrium disappears. The last equilibrium has been
identified as the steady-state voltage collapse point. At this bifurcation point, a
real eigenvalue of the load-flow Jacobian becomes zero, i.e., the Jacobian
becomes singular. The consequence of this loss of operating equilibrium is that
the system state changes dynamically. In particular, the dynamics can be such
that the system voltages fall in a voltage collapse [23].
In voltage stability analysis it is common to consider the curves which relate
voltage to (active or reactive) power. Such curves, referred to as P-V (or Q-V)
curves or nose curves are shown in Figure 2.5, for the simple system of Figure
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
16
2.3. These curves are indeed the projection of the onion surface projection onto
the PV-plane. The curves depend on how Q varies with P; in Fig. 2.5, a constant
power factor, i.e., ϕ= tanPQ , has been assumed for each curve.
The vertex of each curve determines the maximum power that can be
transmitted by the system and it is often called the point of maximum loadability
or point of collapse. Using the equality in equation (2.5) and substituting
ϕtanPQ = yields:
02
2222 =⎟
⎟⎠
⎞⎜⎜⎝
⎛−φ+
XEPtan
XEP (2.6)
Solving (2.6) for P will result in:
XE
φsinφcosPmax 21
2
+= (2.7)
Figure 2.5 The onion surface projected onto the PV-plane. The vertex of each
curve is the maximum loadability point of the system.
Substituting maxP from equation (2.7) into equation (2.4) one will get:
φsinEV Pmax +
=12
(2.8)
1EV
2EPX
20.−00.
40tan .φ −=
20.40.
00.00. 10. 20. 30. 40. 50. 60. 70. 80.
50.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
17
Simply stated, voltage instability results from the attempt to operate the system
beyond maximum loadability point. This may result from a severe load increase
or, more realistically, from a large disturbance that increases X and/or decreases
E to the extent that the predisturbance load demand can no longer be satisfied.
When the point of collapse is reached the voltage starts decreasing quickly since
the reactive support of the system under these heavy loaded conditions is not
enough.
The set of the P-V curves in Figure 2.5 may also be considered as the different
load compensation cases, i.e.; by local reactive compensation it is possible to
increase the transfer capacity of the system, but at the same time the system
operates closer to the security margins, since the point of collapse is placed
closer to acceptable voltages.
2.4 Voltage Sensitivity of Loads
Loads are the driving force of voltage instability, and for this reason this
phenomenon has also been called load instability [1]. The term load refers to the
equivalent representation of the aggregate effect of many individual load devices
and the interconnecting distribution and subtransmission systems that are not
clearly represented in the system model.
The voltage dependency of loads is a critical aspect of voltage stability analysis
and the frequency dependence of loads is not of primary importance [1]. The
influence of the load characteristic in voltage stability is studied in this section.
Loads are a complex, time-varying mix of different devices. They were
historically modeled as constant admittances. More recently they have been
modeled as combination of constant impedance, constant current and constant
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
18
power (ZIP model), or in a voltage exponent form. Equation 2.9 describes the
voltage sensitivity, given by the parameter α, for general static models in
exponential form:
α
⎟⎟⎠
⎞⎜⎜⎝
⎛=
oo V
VPP (2.9)
In equation (2.9) oV is a reference voltage and oP is the active power consumed
under this voltage. Constant impedance, constant current and constant power
characteristics are obtained by using the typical values of α, 2, 1 and 0.
We assume that in Figure 2.3 at a certain time a large disturbance (such as
tripping of a transmission line) increases X, causing a significant change in the
system P-V characteristic. Figure 2.6 shows a P-V representation of the supply
Figure 2.6 Influence of the static load characteristic on P-V curves.
Typical values of α set at 2, 1, 0, -0.1, -0.4 and -0.5 [61]
(pu)V
40.−=α10.−=α0=α
2=α
1=α
50.−=α
Pre-disturbance
edisturbancPost −
(pu)P
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
19
system pre-disturbance and post-disturbance situation. Disturbance reduces the
maximum amount of power that can be transmitted by the system. The static
load characteristic for different values of α is also shown.
For a typical case where the load power is not affected by the voltage, i.e.,
constant power, the parameter is equal to zero. Those values, which are higher
than zero, express load voltage dependency that helps the stability of the system,
by providing some system relief. As exemplified in the figure, the larger this
parameter is, the further the new operating point is from the P-V nose. For a
static representation, this parameter expresses the instantaneous load-voltage
dependency.
Figure 2.6 shows an unstable situation when α is equal to –0.5. The negative
values of the parameter that are harmful to the stability are associated with a
combination of a dynamic restoration of the load and the discrete action of tap
changers that is explained in the next section. The admittances of such loads vary
with the supply voltage, either by their inherent design or by control loops
connected to the load devices, in order to consume constant power
2.5 Load Restoration and Voltage Stability
The static models ignore the dynamic behavior of loads that is important in
voltage stability studies. The power consumed by a static load is a function of
voltage only. However, the power consumed by a dynamic load is a function of
both voltage and time. At a voltage change, the dynamics of various load
components and control mechanisms tend to restore load power, at least to a
certain extent. This is referred as load restoration.
Consider that the power consumed by the load at any time depends upon the
instantaneous value of a load state variable, denoted as x:
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
20
( )x,V,zPP t= (2.10)
( )x,V,zQQ t= (2.11)
where, z is the load demand. tt Q,P are called the transient load characteristics.
If the load dynamics are described by the differential equation:
( )x,V,zfx =& (2.12)
Then the steady state of load dynamics is characterized by the following
algebraic equation:
( ) 0=x,V,zf (2.13)
Equation (2.12) can be used to obtain the state variable x as a function of z and
V:
( )V,zhx = (2.14)
Substituting (2.14) into (2.10) and (2.11) we obtain:
( )( ) ( )V,zPV,zh,V,zPP st == (2.15)
( )( ) ( )V,zQV,zh,V,zQQ st == (2.16)
where, ss Q,P , are the steady-state load characteristics.
The transition toward steady-state load characteristics is driven by the load
dynamics.
When loads are subjected to a step change in voltage, they will typically undergo
an initial (transient) step change in power. This will often be followed by a
period where the load recovers back to a new steady state value. This recovery
may be monotonic, or may involve some damped oscillatory behavior [23].
Three of the well-identified load power restoration mechanisms are:
• Induction motors
• Load behind load tap changer(OLTC)
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
21
• Thermostatic Loads
2.5.1 Induction Motors [1, 7, 24-26]
Induction motors are present in many industrial and commercial loads. The
electrical characteristics of single cage induction motors are often represented by
the equivalent circuit shown in Figure 2.7.
Figure 2.7 Per phase equivalent circuit of induction motor
Where, the parameters sR (stator resistance), sX (stator reactance), mX
(magnetizing reactance), rR (rotor resistance) and rX (rotor reactance) are
known. The rotor slip is indicated by s . The slip used in this model is the
frequency of the bus voltage minus the motor speed. Some programs incorrectly
use either average system frequency or 1.0 in place of the frequency [27].
When subject to a step drop in voltage, the motor active power first decreases as
the square of the voltage (constant impedance behavior), then recovers close to
its predisturbance value in the time frame of a second. The internal variable x of
this process is the rotor slip. For dynamic voltage stability studies a simplified
first order model with slip being the only state variable may be adequate. In fact,
a motor with constant mechanical torque and negligible stator losses restores to
constant mechanical power (demand variable z). Taking into account these losses
and more realistic torque behaviors, there is a small steady-state dependence of P
rXsXsR
sRR r=
mX
jQP +V
I
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
22
with respect to V. The steady-state dependence of the reactive power is a little
more complex. Q first decreases somewhat quadratically with V, reaches a
minimum, and then increases up to the point where the motor stalls due to low
voltage. In large three-phase industrial motors, the stalling voltage can be as low
as 0.7 pu while in smaller appliances (or heavily loaded motors) it is higher.
Based on the above equivalent circuit several levels of detail may be available,
as below [27]:
1. A dynamic model including the mechanical dynamics but not the flux
dynamics (first-order model),
2. addition of the rotor flux dynamics (third-order model) ,
3. Addition of the stator flux dynamics (fifth-order model).
Stator flux dynamics are normally ignored in stability analysis and for long-term
dynamic analysis the rotor flux dynamics may also be neglected, particularly.
Considering the above assumptions the induction motor state equation is as
below[28]:
ems PPdt
sdT −= (2.17)
In (2.17) mP is mechanical power and,
( )22
2
XsR
sRVPr
re
+= (2.18)
where rs XXX += is the total machine reactance.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
23
2.5.2 On-Load Tap Changers (OLTCs) One of the key mechanisms in the load restoration is the voltage regulation
performed by the tap changing devices of main power transformers. The tap
changer controls the voltage of the distribution by changing the transformer
ratio. Normally the variable tap is on the high voltage side.
An early detailed description of a typical tap changer control system was given
in [29]. The proposed tap changer model was a complex nonlinear dynamic
model that encompassed some inherent time delays. More recent work presented
in [30] explored further the tap changer modeling issue, and presented simpler
yet still accurate models. Depending on the OLTC characteristic (type of time
delay), various discrete state dynamic models and corresponding continuous
approximations were derived.
Figure 2.8 Function of an OLTC control system [31]
The function of a typical OLTC control system is shown in Figure 2.8. The
system remains in the state wait as long as the voltage deviation ( rvv − ) is less
than the function voltage ( functionv ). When the limit is exceeded, a transition to
the state count occurs. Upon entering count, a timer is started and is kept
running until either it reaches the delay time dT , causing a transition to the state
action; or the voltage deviation becomes less than the reset voltage ( resetv ),
causing a transition to the state wait and reset of the timer. When entering the
ready
count
wait
action
functionr uvv >− dcount TT >
resetr uvv <−
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
24
state action, a control pulse to operate the tap changer is given. After the
mechanical delay time ( mT ), the tap operation is completed and the control
system receives a ready signal from the tap changer. The control system then
returns to state wait [31, 32].
The time delay is tuned by the time delay parameter Td0. The actual time delay
can then be either fixed ( dod TT = ), or inversely proportional to the voltage
deviation ( rdod vvTT −∝ ). The variable tap ratio has a limited regulation
range and each step is usually in the range of 0.5%-1.5%.
Load Restoration through LTC
The load restoration by LTC is indirect. In Figure 2.9, when LTC restores 2V to
its reference value the load power, which depends on voltage, is also restored. In
the analysis of LTC dynamics the fast dynamics of the generators and induction
motors can be replaced by their steady-state equations [6].
Figure 2.9 Generator-line –LTC system [6]
Figure 2.10 shows three different load transient characteristics for different
values of n. Following a disturbance due to the initial drop in the primary
voltage operating point will move from point O to point A. Then due to the LTC
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
25
Figure 2.10 P-V curves of generator-line-LTC system [6]
action the secondary side voltage will change and the operating point will move
to point B, causing more drops in primary side voltage. The vertical dashed line
in Figure 2.10 shows the load characteristics when its voltage has restored to its
reference value. When the voltage near the loads (i.e., on the regulated side of
the device) is restored to its pre-disturbance level, the voltage dependence as
viewed from the bulk system is eliminated. While the system voltage may have
changed significantly, the voltage seen by the load has returned to its
predisturbance state. Therefore the power consumed, regardless of the load
voltage dependence, is relatively constant [8, 33]. This is the steady-state load
characteristic seen from the primary side.
LTC Reverse Control Action
In references [34-42] reverse control action of LTC in association with voltage
collapse is investigated. It is shown theoretically that behind a critical condition
of tap position, the secondary voltage drops if the tap position is raised aiming at
raising the secondary voltage. The above critical condition for occurrence of
no
n1
n2
OA
B
P1
V1
Post- disturbance
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
26
reverse action coincides with the power matching condition under which the
power consumed by load is maximized [21]. All practical feeders serve
components of voltage sensitive loads. By changing the tap the effective load
admittance (seen from bulk power side) decreases, so that eventually it can
become smaller than that of the equivalent network feeding the load bus. After
this point the process of load restoration becomes unstable, as successive tap
changes decrease distribution voltage further and further away from its setpoint
[43]. The proposed voltage instability predictor in [12] is based on comparison of
system and load impedance, too.
In other words, the power matching condition is nothing but the voltage collapse
condition. This fact suggests that the reverse action of tap changer is closely
related to the voltage collapse. Hence, tap locking action could be one of the
countermeasures to avoid voltage collapse. There are, however, exceptions such
as a feeder serving almost exclusively heavily loaded induction motors [43]. The
reactive consumption of induction motor is dependent on supply voltage, and
demonstrates a negative slope of reactive power to voltage in heavily loaded,
compensated cases. In such a case increase of distribution side voltage will
decrease reactive consumption and thus will result in increasing the transmission
side voltage as well.
Yorino and Galina in [37] have investigated the sensitivity of load voltage to tap
position with both load exact characteristics and constant power load and have
concluded that it can be used to assess the occurrence of reverse control action .
In [35] Hong and Wang have presented an approach for estimating voltage
stability region concluding the stability region of LTCs.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
27
2.5.3 Thermostatic Loads [1, 7, 44, 45]
An aspect of load behavior that contributes significantly to the voltage stability
problem is the effect of thermostatic controls. The voltage dependence of loads
in a system, particularly loads such as resistive space and water heating, can give
considerable load power relief following a voltage depression induced by a
system disturbance. However, this reduction in power does not remove the need
to deliver energy, e.g., to maintain constant temperature. Eventually, the reduced
power consumption of the individual loads results in thermostats leaving loads
connected longer. The aggregate effect is to push the nominal load power up
towards a level that will produce the pre-disturbance actual power at the
depressed voltage. The state variable (x) and demand variable (z) for this process
are connected equipment and energy requirement, respectively. The time
constant associated with this resetting action is open to investigation, but values
between 10 and 30 minutes have been suggested [33]. However, for a large
enough voltage drop, the aggregate load power does not recover to its
predisturbance value, owing to the fact that the heaters stay on permanently, thus
giving a mere impedance load characteristic in the steady state. It is not always
necessary to take into account “thermostat” characteristics, unless significant
numbers of on-load tap-changers will reach regulation limits following a
disturbance.
2.6 Literature Review on On-line Voltage Stability Analysis
Very few online methods have been proposed in the past to predict voltage
collapse of a power system and to take corrective actions [9, 46-49]. A few such
methods are discussed here which utilities use predominantly to study voltage
collapses.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
28
Voltage stability evaluation using voltage at bus alone will not give a correct
indication of an impending collapse [50]. Sometimes, voltage levels may fall
below limits due to heavy loading, during peak hours. This is only a low voltage
condition and may not lead to voltage collapse. Consequently just monitoring
voltage levels should not be used as a voltage collapse warning index. Voltage
levels have to be studied as function of some other key system parameter such as
real power or reactive power [49].
Voltage stability analysis is still widely studied in industries by computing the P-
V and Q-V curves at selected load buses [1]. Generally, such curves are
generated, by executing large number of load flows, using conventional methods
and models. While such procedures can be automated, they are time-consuming
and do not readily provide information, useful in gaining insight into the causes
of stability problems. In addition, these procedures focus on individual buses;
that is, the stability characteristics are established by stressing each bus
independently. This may unrealistically distort the stability condition of the
system. Also, the buses selected for Q-V and P-V analysis must be chosen very
carefully, and a large number of such curves may be required to obtain complete
information.
The sensitivity analysis using Jacobian matrix evaluates the dV/dQ factor for
voltage stability analysis [51, 52]. When the system is heavily loaded, generators
may reach their reactive limits. Since sensitivity analysis is based on load flows,
a P-V bus in this case will be replaced as P-Q bus and there will be drastic
changes in the sensitivities.
Moreover, the magnitudes of the sensitivities for different system conditions do
not provide a direct measure of the relative degree of stability.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
29
Modal analysis method is an indirect calculation of the dV/dQ sensitivities [53].
When the system is heavily stressed both dV/dQ and dV/dP sensitivities play an
important role in voltage collapse prediction. The implicit assumption is that the
active power dynamics don’t play any role in the voltage collapse, which is not
valid when the system is heavily loaded. A small positive eigenvalue is itself an
indication of danger. This method predicts voltage collapse only when one of the
calculated eigenvalues becomes negative [49].
2.7 Local Data Based Voltage Stability Monitoring
Voltage stability is threatened when a disturbance increases the reactive power
demand beyond the sustainable capacity of the available reactive power
resources [8]. Since the reactive power can not be transferred to areas far away,
voltage instability events usually appear due to localized shortage of reactive
power or voltage control [54]. This makes it possible for decentralized voltage
stability on-line monitoring just with the local bus information, while networks
with long distance to the area being monitored can be equalized.
The local voltage stability monitoring and control is based on a two-bus
equivalent system as shown in the Figure 2.11. The supply system is represented
by its Thevenin equivalent seen from the terminals of the load bus of interest. In
this section the literature related to the system Thevenin impedance identification
is reviewed.
Figure 2.11 Two-bus system
P+jQthhth jXtRZ +=VthE
I
ljXlRlZ +=
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
30
From electric circuit theory it is known that load power is maximized when the
load impedance is the complex conjugate of the transmission impedance. This
case however is not suited for power system applications. The first problem is
that in a transmission system the resistance can be negligible compared to the
reactance, and thus the maximum power goes to infinity. The other problem is
that, a highly capacitive load would be required to match the dominantly
inductive nature of the system impedance. Also, half of the energy must be
consumed in transmission system [6].
A modified derivation, closer to power system applications is made by assuming
that the power factor of the load is specified. Specifying the load power factor
φ cos is equivalent to having a load impedance of the form (Figure 2.11):
φtanjRRjXRZ lllll +=+= (2.19)
Under this assumption, the load active power is given by:
22
2
φ)tanR(X)R(RERP
lthlth
l
+++= (2.20)
Maximizing P over the variable lR , the necessary extremum condition is:
0=∂∂
lRP
(2.21)
The solution of (2.21) provides the following criterion:
( ) ( ) 01 2222 =+−+ φtanRXR lthth (2.22)
which is equivalent to:
thl ZZ = (2.23)
The second derivative is given by:
( )φtanRR
Pl
l
22
212 +−=
∂∂
(2.24)
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
31
which is always negative, indicating that the solution is a maximum. Thus in this
case the load power is maximized when the load impedance becomes equal in
magnitude to the transmission or Thevenin impedance;
Maximal power transfer thl ZZ =⇔
For a lossless transmission the load maximum active power and receiving end
voltage will be, respectively:
thmax X
EφsinφcosP
21
2
+= (2.25)
φsin12 +=
EV Pmax (2.26)
which are the same power limit and voltage magnitude for a lossless line derived
from load flow in equations (2.7) and (2.8). This means that the impedance
matching point is nothing but the voltage collapse point.
Based on this fact, Vu et. al. have proposed a Voltage Instability Predictor(VIP)
[15]. In their work the ratio between the voltage and current phasors measured at
the load bus is used to compute the apparent impedance of the load. Local bus
and the rest of the system are treated as Thevenin equivalent circuit.
The impedance plane is separated into two regions, the Thevenin impedance
circle and the rest of the plane (Figure 2.12). As load varies, its apparent
impedance traces a path in the plane, and voltage instability occurs when the
load impedance crosses the Thevenin circle Indeed, VIP can be viewed as a
voltage Relay with an adaptive set-point [14].
In that work [14], using only local phasor measurements of voltage (V) and
current (I), the Thevenin equivalent is estimated, using Ohm’s law in equation
(2.27).
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
32
IZVE th+= (2.27)
The Thevenin equivalent parameters are estimated by the measurements taken at
two or more different times [15]. It is assumed that source voltage and Thevenin
impedance are constant.
The reports on the feasibility study of the VIP for a realistic 7000 bus system is
provided in[12], where the proximity to voltage collapse is expressed in terms of
the distance between two voltage curves or two impedance curves.
Figure 2.12 Maximal power transfer is reached (voltage instability) when the
apparent impedance of the load bus reaches the Thevenin circle [14]
In [16], a new measure; power margin, is proposed to describe the proximity to
collapse in terms of power.
The proposed VIP can precisely predict voltage collapse if the system Thevenin
impedance be estimated accurately. Using VIP the proximity to voltage collapse
can be described in terms of power and can be looked upon as the power
available to be pushed through the VIP location before the network collapses.
The device has several potential applications. VIP is a voltage relay with an
adaptive set-point. It can be used to impose a limit on the loading at each bus,
and sheds load when the limit is exceeded. It can also be used to enhance
existing voltage controllers such as SVCs. Coordinated control can be obtained if
Circle of radius
thZ
lZ
r
x
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
33
communication links are available; however, in case of emergency, the device
carries out its own decision.
Using the same method for Thevenin equivalent estimation, a method has been
proposed by Haque in [48] to estimate on-line the maximum permissible loading
and voltage stability margin at different buses of a system.
Looking at the voltage and current at the connection point will provide the
Thevenin impedance of the system if there were no variation in the system.
The major difficulty is that there are system variations; such as system dynamics
and dynamic loads, changing the Thevenin equivalent of the supply at the
connection point. It can be shown that in this case the apparent estimated
impedance will be equal to the load impedance, interpreting that the system
voltage is collapsing, even though this is not the case.
Taking the measurements at two different times and using equation (2.27) one
gets:
( )121212 IIZVVEE th −+−=− (2.28)
Assuming constant Thevenin source we will get:
IZV thΔ+Δ=0 (2.29)
Thus:
IVZth Δ
Δ−= (2.30)
It is also obvious that:
( )1212 IIZVV l −=− (2.31)
Thus:
thl ZIVZ −=ΔΔ
= (2.32)
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
34
Warland and Holen [55] tried to overcome the problems arising when the
Thevenin equivalent is not constant(the usual case). They have introduced the
VIP++ that estimates the Thevenin equivalent parameters by adding the
knowledge from the surrounding area, using phasor measurement units on two
buses. The proposed method loses the locality feature of data in the estimation
process.
Beguvic and Novosel in [13] and Soliman et.al, in [56] using the local bus past
data together with the recent measurements have proposed a recursive least
square method to predict Thevenin impedance. However they do not discuss how
the method will work if a disturbance changes the system impedance and/or
there are system dynamics.
The method introduced by Palethorpe and et.al. in [57] employs a power
electronic circuit to inject a small current disturbance onto the energized power
system, and the measurements of the disturbance current and resultant voltage
transients are used to identify the system impedance. In that work the main
problem is that they have not taken into account the system and load changes.
2.8 Literature Review on Aggregate Load Modeling
While data for most transmission and generation elements is well established or
can be readily determined from measurement, good load data and models yet
remain difficult to reliably ascertain [58].
The load seen by a bulk power delivery transformer is an aggregate of many
individual loads, fed through distribution lines and MV/LV transformers,
compensated by switched capacitors, etc. While typical data can be obtained for
individual equipments [59], the real problem is to determine the composition of
the load that varies not only from one bus to another but also with the season, the
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
35
time of the day, etc. Thus the aggregate loads are the most uncertain power
system components.
A typical load response to a step change in voltage is illustrated in Figure 2.13.
Figure 2.13 Load power response to a voltage drop
The load recovery shown in Fig. 2.13 is characterized by three parameters,
steady state load-voltage dependence, transient load-voltage dependence and a
load-recovery time constant [60].
pu
pus
o
oss
pu
put
o
ott
maxpp
minp
oOp
oOs
pp
V)P(
V/VP/Pand
V)P(
V/VP/P
zzzwith
VVPz
VVPP)V(P
dtdz
Tts
Δ
Δ=
ΔΔ
≅αΔ
Δ=
ΔΔ
≅α
⟨⟨
⎟⎟⎠
⎞⎜⎜⎝
⎛−⎟⎟
⎠
⎞⎜⎜⎝
⎛=−=
αα
(2.33)
where pz is an internal state variable(load recovery). oV and oP are the voltage
and power consumption before a voltage change. pT is the active load recovery
time constant and can be obtained from a least-square fit of the time response.
Similar relationships hold for the reactive power [6]. sα and tα are the steady-
state and transient load-voltage dependence, respectively. In [61-63] using the
normal operation data the load parameters time-varying characteristic and their
dependency with weather and season of the year have been studied and have
P
t tPΔ
sPΔ
V VΔ
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
36
been concluded that the active and reactive time constants and transient time
load-voltage dependence exhibit a strong time dependency, even during the day.
It has been proven that load representation has a critical effect on voltage
stability analysis results when loads are subjected to large variations in system
voltages [1]. Consequently, more and more studies have been done to better
understand the nature of load dynamics [11, 27, 28, 60, 64-73].
Load modeling has typically been conducted using two basic approaches:
Component based modeling and measurement based modeling.
2.8.1 Component based and measurement based load modeling In this approach a survey of the individual devices which make up the aggregate
load is made and based on known characteristics of the individual devices a
composite load is synthesized [58]. Some Shortcomings of the method are as
below:
• Large number of individual load devices
• Devices are not connected simultaneously
• Device characteristic varies by the age and manufacturer
Measurement approach to identify the load model and its parameters comes from
real load and may give better modeling results over the component based
approach [11].
The most convenient load model from the point of view of parameter
identification is the input-output transfer function load model (Fig. 2.14).
Transfer function G(s) is of the general form:
Figure 2.14 Block diagram of input-output transfer function model
G(s) VΔ PΔ
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
37
G(s) = K01
201
2
asasbsbs
++
++ (2.34)
where K is the gain constant.
Xu et al. in [69] exploring the determination of load characteristic from field
tests performed on the B.C. Hydro system, have proposed a load test procedure.
It is shown that load response due to disturbances caused by transformer tap
changing are sufficient to capture the dynamic and static load characteristics. It is
also shown that the amount of voltage drop is not very critical for determination
of load parameters and a 5% voltage drop will be sufficient. Their tests indicated
also that the load tap changers and feeder voltage regulators are the main cause
of load recovery dynamics. The influence of voltage variations on the estimation
of the load model parameters has been investigated by Dai et al [74]. It is shown
that first order models are more robust and input-output transfer function model
is not influenced by the delay of the transducers.
In[11] generic algorithms and evolutionary programming based identification is
used by Zhu et al., to identify the power system load model based on data from
field measurement. Beguvic and Mills in [75] proposed a simple recursive
algorithm for real-time identification of the basic components of the composite
load models and, proposing a monitoring system, investigated their effect on the
voltage stability boundary.
Disadvantages of the measurement based models are:
• They are valid only for the particular time and location of measurement
• They are valid only for the small voltage changes
• They do not account the rotational non-linearities of induction motors
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
38
Load models should be able to capture, with acceptable accuracy, the load
behavior. The models should also be physically based, derived from “easy to
obtain” data and suitable for use in both static and dynamic analysis. In a real-
time control, the control effect should be improved if load parameters can be
real-time identified.
2.9 Countermeasure to Long-term Voltage Instability
Although a system protection scheme may integrate and coordinate several types
of actions, action on load is the ultimate countermeasure. Shedding a proper
amount of load, at a proper place, within a proper time can be done indirectly
through a modified control of OLTCs or directly as load shedding.
Emergency control of OLTC can be achieved by OLTC blocking, by bringing
back the taps to predetermined positions, or by reducing OLTC set-points [43].
In the latter case, the sensitivity of load to voltage is eliminated. From the
viewpoint of customer voltage quality, although the distribution voltage remains
low, it is on the average less sensitive to transmission transients [10].
The role of automatic and non-automatic OLTCs in emergency voltage control is
investigated in [43]. It is shown that how tap blocking of bulk power delivery
transformers can prevent an approaching voltage collapse. The problems and
limitations of this countermeasure are also discussed. It is also shown that
OLTCs at higher voltage levels (including those of generator step-up
transformers) can help maximize the loadability margin either by automatic
control, or by selection of tap adjustments using off-line studies.
Chapter 2. Literature Review of Local Data Based Voltage Stability Monitoring
39
2.10 Summary
The fundamental materials that are needed in the following chapters are
reviewed in this chapter. The related works to the on-line voltage stability
monitoring and load modeling were also reviewed and discussed.
In chapter 3 a method is developed to estimate power system Thevenin
impedance that is based on signal processing on the measured data in the load
bus.
Chapter 3
Correlation Based System Thevenin Impedance Estimation
3.1 Introduction As it was stated in chapter 2 the local voltage stability monitoring and control is
based on a two-bus equivalent system where the supply system is represented by
its Thevenin equivalent seen from the terminals of the load bus of interest. In this
chapter a method is developed to estimate power system Thevenin impedance
that is based on signal processing on the random changes of the measured
voltage and current phasors in the load bus. It is shown that the cross-
correlations of the changes in the load voltage and current with respect to the
changes in the load admittance can be used to estimate the system Thevenin
impedance. It is assumed that the power system is linear and the customer load
acts as a random walk process [76]. The required steps are clearly shown.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
41
Dynamic components in the load voltage and current caused by system and/or
load are removed and the residuals are used for the estimation. The method is
validated by simulation and work on real data is also provided to reconfirm the
method.
3.2 Theory of the Correlation Based System Thevenin
Impedance Estimation
There is evidence that the changes in composite customer load in periods up to
10 seconds are unpredictable. The idea that the costumer load changes are
uncorrelated white noise, and hence, the composite load would be the integral of
white noise [77] is confirmed in [76]. Changes in the customer load cause
changes in the load bus voltage and magnitude (Figure 3.1).
Figure 3.1. Time measurement of load voltage and current magnitudes
Chapter 3. Correlation Based System Thevenin Impedance Estimation
42
In the next sections the theory of the system Thevenin impedance estimation
using the changes in the local load bus voltage and current is provided. Two
different methods are used that the first method is based on the block diagram
representation of power system and the other approach uses the system
equivalent circuit to develop the theory.
3.2.1 Block diagram representation The block diagram of the interaction between a power system and a local load is
shown in Figure 3.2. VΔ and IΔ are the changes of the local load bus voltage
and current phasors, respectively. SZ and LY are the system equivalent
impedance and the load admittance, respectively. The system mode is partly
exited by the changes in the local load and partly from changes in the other
loads, here treated as w(t) and d(t) as white noise uncorrelated, respectively.
Figure 3.2. Block diagram representation of proposed method for system
identification
In Figure 3.2, one can write:
( )SL
sLZY
tdZYtwI++
=Δ1)( (3.1)
))()((1
twtdZY
ZVSL
S −+
=Δ (3.2)
Local Load
Power System
w(t)
d(t)
∑∑ sZ+
IΔ
LY
+
_ +
VΔ Local Load
Changes
Other Loads Changes
Chapter 3. Correlation Based System Thevenin Impedance Estimation
43
There is no clear relation between VΔ and IΔ , but, multiplying VΔ by IΔ and
IΔ by itself will result in:
( )21][
SL
sLS
ZY)t(w)t(d)t(d)t(dZY)t(w)t(w)t(d)t(w
ZIV+
−+−=ΔΔ (3.3)
( )222
12
SL
SLSL
ZY)t(d)t(dZY)t(w)t(dZY)t(w)t(w
II+
++=ΔΔ (3.4)
Applying the expectation function (mean value) to equations (3.3) and (3.4) will
yield:
[ ] [ ] [ ] [ ] [ ]{ }( )21 SL
sLS
ZY)t(w)t(dE)t(d)t(dEZY)t(w)t(wE)t(d)t(wE
ZIVE+
−+−=ΔΔ
(3.5)
[ ] [ ] [ ] [ ]( )2
22
12
SL
SLSL
ZY)t(d)t(dEZY)t(w)t(dEZY)t(w)t(wE
IIE+
++=ΔΔ (3.6)
Where, E denotes expected or mean value. Now considering that w(t) and d(t)
are uncorrelated white noises, i.e.; [ ] 0=)t(d)t(wE , equations (3.5) and (3.6)
become:
[ ]( )21 SL
sLS
ZYDZYW
ZIVE+
+−=ΔΔ (3.7)
[ ]( )2
22
1 SL
SL
ZYDZYW
IIE+
+=ΔΔ (3.8)
Where, D and W are variances of d and w, respectively. The ratio of equation
(3.7) with respect to equation (3.8) will have two components as below:
[ ][ ] DZYW
WZDZYW
DYZIIEIVE
SL
s
SL
Ls2222
2
+−
+=
ΔΔΔΔ
(3.9)
In the case when the majority of the modal disturbance is originated in the local
load section being examined, i.e.; 0≈D , equation (3.9) can approximate:
Chapter 3. Correlation Based System Thevenin Impedance Estimation
44
[ ][ ] SZ
IIEIVE
−≈ΔΔΔΔ
(3.10)
The negative sign in equation (3.10) is because current is leaving the supply
system. In the case 0≈W , (3.9) can be used to identify the load. If both W and D
are present equation (3.10) can not be used to estimate the system impedance. In
this case ΔV and ΔI should be correlated with a quantity that is correlated with
w but uncorrelated with d. This quantity for example can be the changes in the
admittance (or impedance) of the local load. Let this quantity be LYΔ , then
considering [ ] 0=Δ )t(dYE L it can easily be shown that:
[ ] [ ]SL
LSL ZY
w(t)Δ(EZΔVΔYE
+−=
1 (3.11)
and,
[ ] [ ]SL
LL ZY
w(t)Δ(EΔIΔYE+
=1
(3.12)
Hence:
[ ][ ] S
L
L ZΔIΔYEΔVΔYE
−= (3.13)
It should be notified that the terms in equations (3.10) and (3.13) are
instantaneous changes in quantities. This means that any possible dynamic
component caused by load and/or system needs to be removed from voltage and
current differences before using the equations.
3.2.2 Using system equivalent circuit
In Figure (3.3) power system is represented by its Thevenin equivalent. It is
assumed that:
1. The system is linear.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
45
2. Load admittance has a fixed real and imaginary components plus
uncorrelated and unpredictable changes in each component.
3. There are random changes in the supply system causing random
changes in the Thevenin equivalent.
The changes in the supply system and local load will give rise to the
changes in the load current and bus voltage.
Figure 3.3 load connected to the equivalent circuit of the system
It was shown in section 3.2 that the changes in the load bus voltage and current
can be used to estimate the system Thevenin impedance. At the first glance it is
expected that the changes in the voltage in phase and out-of-phase components
with respect to current can be respectively used to estimate the real and
imaginary components of the system Thevenin impedance, respectively. This
hypothesis is examined in the next section.
a) Voltage in phase and out-of-phase components with current In Figure 3.3 one can write:
VIZE =− (3.14)
Let δ= jEeE α= jVeV and β= jIeI . Multiplying both sides of equation (3.14)
by β− je1 yields:
)()( α-βjδ-βj eVIZeE =− (3.15)
System equivalent
Ijxrz +=
V
_
+EY
Chapter 3. Correlation Based System Thevenin Impedance Estimation
46
In another form,
ϕ+ϕ=+−−+− sincos)()sin()cos( jVVIjxrβδjEβδE (3.16)
Where RVV =ϕcos and XVV =ϕsin are the voltage in phase and out of phase
components with respect to current, respectively, as shown in Figure 3.4.
Figure 3.4 Voltage components with respect to current
The difference of equation (3.16) at two different times becomes the two
following equations:
12)()cos()cos( 12111222 RR VVIIrβδEβδE −=−−−−− (3.17)
12)()sin()sin( 12111222 XX VVIIxβδEβδE −=−−−−− (3.18)
The two first terms in each of the above equations are dependent on the
magnitude and angle of the system Thevenin voltage and the current angle, and
hence, these terms can not be removed by correlation methods. These equations
can not be used to estimate the system Thevenin impedance components because
there is no information available about the system Thevenin voltage.
b) Random changes in the load bus voltage and current [78] Considering changes to E , V and I equation (3.14) can be rewritten as
following:
V
I
βαϕ −=
ϕcosVVR =
ϕsinVVX =
Im
Re
Chapter 3. Correlation Based System Thevenin Impedance Estimation
47
( ) ( ) VΔVIΔIZEΔE +=+−+ (3.19)
Subtracting equation (3.14)) from equation (3.19) gives:
VΔIZΔEΔ =− (3.20)
Multiplying both sides of equation (3.20) by *IΔ 1, the complex conjugate of IΔ ,
and getting the expectation value yields:
⎥⎦⎤
⎢⎣⎡=⎥⎦
⎤⎢⎣⎡−⎥⎦
⎤⎢⎣⎡ **
S* IΔVΔΕIΔIΔΕZIΔEΔΕ (3.21)
Where E means expectation or mean value and [ ]ZmeanZS = . Solving equation
(3.21) for SZ gives:
⎥⎦⎤
⎢⎣⎡
⎥⎦⎤
⎢⎣⎡−⎥⎦
⎤⎢⎣⎡
=*
**
SIΔIΔΕ
IΔVΔΕIΔEΔΕZ (3.22)
In equation (3.22) the possible small random changes in the system Thevenin
impedance have been cancelled by averaging the impedance over the sampling
time frame. The rest of this section is devoted to the application of equation
(3.22) in different cases in power system.
Case 1: System with Constant Thevenin Voltage, 0EΔ = In this case it can easily be shown that equation (3.22) will reduce to the
following equation:
⎥⎦⎤
⎢⎣⎡ ΔΔ
⎥⎦⎤
⎢⎣⎡ ΔΔ
−=*
*
SIIΕ
IVΕZ (3.23)
1 For the complex processes (t)iZ and (t)jZ cross-correlation functions is defined by
)](t*(t)[)( τ+=τ jiZZ ZZER [peebles].
Chapter 3. Correlation Based System Thevenin Impedance Estimation
48
Being available the data of load voltage and current at different times, in this
case equation (3.23) will be applicable, even though it is not a realistic case.
Case 2: Random Changes in the Thevenin Source, 0EΔ ≠ The random changes in the system, e.g., random changes in the other loads in
the system, will cause random changes in the system Thevenin voltage.
In this case 0≠EΔ , and because E and I are correlated variables, equation
(3.23) can not be used to estimate the system impedance. Correlating the
variables in equation (3.22) with YΔ , the changes in the local load admittance,
can be used to eliminate EΔ from equation because E and Y are not correlated,
i.e.; [ ] 0=YE ΔΔΕ . This means that in this case equation (3.20) will estimate the
system Thevenin impedance if the quantities in the equation are correlated
with YΔ , hence:
[ ][ ]YIΕ
YVΕZS ΔΔΔΔ
−= (3.24)
Case 3: System with Dynamics System variations; such as system dynamics and dynamic loads, will impose
dynamic variations to the Thevenin equivalent of the supply at the connection
point. Consequently, load voltage and load currents will also contain dynamic
components. In equations (3.23) and (3.24) the effect of the dynamic component
in EΔ can not be cancelled by getting its correlation with respect to the changes
in the load current, in equation (3.23), and/or changes in the load admittance, in
equation (3.24), and consequently estimation fails.
Subtracting dynamic components from load voltage and load current differences
can be a useful method for the elimination of the dynamics in equations (3.23)
and (3.24). This can be achieved by fitting a curve to the real and imaginary
Chapter 3. Correlation Based System Thevenin Impedance Estimation
49
components of the measured load voltage and current differences, individually,
and then subtracting the predicted curves from the real ones. The residuals then
may be used for the estimation of the system impedance using equation (3.23)
and/or equation (3.24).
3.3 Algorithm of the Correlation Based System Thevenin
Impedance Estimation
Considering all above cases the system Thevenin impedance can be estimated
using the following steps:
Step1: Providing the initial local load bus data
Compute load current (simulation) and/or load admittance (real data) using
VYI = equation in each time step.
Step2: Refining the local load bus data
Compute the changes in the local load bus voltage, current and admittance using
the following equations:
( )∑−
=+ −=
1
11
n
kkk VVVΔ (3.25)
( )∑−
=+ −−=
1
11
n
kkk IIIΔ (3.26)
( )∑−
=+ −=
1
11
n
kkk YYΔY (3.27)
Where n is the number of the data points. The sign of the current difference is
changed to represent IΔ a current feeding the system.
This differentiation is needed because the estimation method uses the random
changes in the load voltage, current and admittance.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
50
Step3: Removing the dynamic components
Remove any dynamic components, caused by system and/or load,
from IΔVΔ and . As it was already stated in this chapter the method uses only
the instantaneous changes in quantities. In this thesis one-step prediction method
can be used to remove the dynamic components (Ref. to Appendix A).
Step4: Thevenin impedance estimation
Estimate system Thevenin impedance using equations (3.23) and/or (3.24).
Equation (3.23) can only be used when there are no changes in the supply
system. It is noticeable that the estimated thZ is the average of the system
equivalent impedance over the data time frame and the small possible random
changes in system Thevenin impedance are ignored. However, if a disturbance in
the system causes a sudden change in the supply system the estimation process
must be renewed to identify the system new equivalent impedance. This situation
can be recognized by a sudden change in the bus voltage while the load
impedance remains almost constant.
3.4 Simulation Results In this section, the algorithm of the system Thevenin impedance estimation is
confirmed by simulating a test system. The proposed method is applied to the
four bus system shown in Figure 3.5.
Figure 3.5 Four bus test system with two different variable loads
I
2Z1
Load #1 Load #2
23 41Z 3Z
Chapter 3. Correlation Based System Thevenin Impedance Estimation
51
Bus 1 is modelled as an infinite bus. The changes in load admittances are
modelled as unpredictable random processes. These changes give rise to the
changes in the local load bus (bus 3) voltage and current. Also, the changes in
load #2 cause changes in the system Thevenin voltage, seen from the bus 3 view
point.
Dynamic components are also created in the load voltage and current by
choosing an initial value for the angle of the synchronous machine in bus 2
different from its steady state value, and/or, adding dynamic loads; such as
induction motors, to the system. These dynamic components of the load voltage
and current are removed using a “one-step prediction” algorithm.
The changes in the voltage and current of load #1 are then used to estimate the
system Thevenin impedance (seen from the bus 3).
3.4.1 Case 1 System with Constant Thevenin, no dynamics In this case passive load #2 is modelled as a constant admittance. The passive
load #1 admittance has a fixed real and imaginary component (constant power
factor) plus a small separate stochastic change in each component.
The small load changes in each step are considered to be normally distributed
random variables. These changes in the load will give rise to variations in the
voltage and load current in bus 3.
The magnitude of the simulated load admittance for a 0.02 sec sampling time
and 100 sec total simulation time is shown in Figure 3.6. Figure 3.7 indicates the
magnitudes and the angles of the load current and load bus voltage.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
52
Figure 3.6 Simulation of load admittance changes in case 1
Figure 3.7 Magnitudes and phases of the load voltage and current in case 1
Chapter 3. Correlation Based System Thevenin Impedance Estimation
53
Bus 3 voltage and current and the load admittance are differentiated using
equations (3.25) - (3.27). The difference of the real component of the load bus
voltage and its auto-correlation are shown in Figure 3.8.
Figure 3.8 Difference of the real component of the load bus voltage and its auto-
correlation in case 1
Equations (3.23) and (3.24) are used to estimate the system Thevenin impedance.
The results of the simulations with different values for Thevenin impedance thZ
are shown in Table 3.1.
Referring to table 3.1, for both equations the estimated values are very close to
the actual values. The small differences are due to the applying one-step
prediction program.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
54
TABLE 3.1 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE IN CASE 1
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
0.0499 + 0.1065i
0.0983 + 0.1915i
0.0842 + 0.3553i
0.0497 + 0.1064i
0.0977 + 0.1913i
0.0837 + 0.3553
0.0502 + 0.1064i
0.0980 + 0.1927i
0.0843 + 0.3546i
3.4.2 Case 2: Random Changes in the supply system, no dynamics In this case the changes in the components of both load admittances are taken as
uncorrelated random changes (Figure 3.9). The voltage phasor and the current
phasor to the load in bus 3 are measured and used for estimation of the system
Thevenin impedance.
Figure 3.9 Simulation of load admittances changes in case 2
Chapter 3. Correlation Based System Thevenin Impedance Estimation
55
Equations (3.25) - (3.27) were used to find the differences of the bus 3 voltage
and current and the load admittance. System Thevenin impedance was estimated
using (3.23) and (3.24). The estimation results for different values of the ratios of
random changes in two loads are shown in Table 3.2. As it is expected the
achieved results by equation (3.23) are not close to the actual values and even in
the last case (ratio=5) the components of the estimated impedance are negative.
TABLE 3.2 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR A SIMULATION TIME, T=100 SEC AND A TIME STEP ΔT =
0.02 SEC AND DIFFERENT RATIOS OF RANDOM CHANGES IN LOAD #2 WITH RESPECT TO
LOAD #1.
Ratio of changes
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
0.5 1 2 5
0.0499 + 0.1065i 0.0499 + 0.1064i 0.0501 + 0.1064i 0.0505 + 0.1060i
0.0474 + 0.1044i 0.0416 + 0.0991i 0.0181 + 0.0800i -0.1155 - 0.0275i
0.0498 + 0.1057i 0.0497 + 0.1081i 0.0470 + 0.1068i 0.0495 + 0.1118i
3.4.3 Effect of the electrical distance between two loads in estimation In this part, the effect of the electrical distance between changing loads in the
estimation results is investigated by choosing different values for 2Z in Figure
3.6. Results of estimation are shown in Table 3.3. In this simulation the ratio of
the changes in load #2 is chosen to be twice the changes in load #1. In the latter
case in Table 3.3 two loads are connected to the same bus.
TABLE 3.3 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR A SIMULATION TIME T=100 SEC AND A TIME STEP TΔ = 0.04
FOR DIFFERENT VALUES OF 2Z
2Z
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
0.0+0.2i 0.0+0.1i 0.0+0.0i
0.0542 + 0.1324i 0.0508 + 0.1166i 0.0510 + 0.1168i
0.0330 + 0.1143i 0.0232 + 0.0927i 0.0147 + 0.0617i
0.0517 + 0.1313i 0.0507 + 0.1139i 0.0493 + 0.0905i
Chapter 3. Correlation Based System Thevenin Impedance Estimation
56
In this case, equation (3.23) fails to estimate the system Thevenin impedance,
due to the changes in the supply system caused by the changes in load #2.
Equation (3.24) has reasonable accuracy except for the latter case that both loads
are connected to bus 3. Indeed in this case, load #2 is a part of the local load, and
to have a good estimation result, its current also should be included in the load
current.
3.4.4 Case 3a: System with dynamics, no random changes in supply system In this case a combination of passive and induction motor loads are connected to
bus 3 (Figure 3.10).
Figure 3.10. Four bus test system with dynamic load
Bus 1 is modelled as infinite bus. The synchronous generator in bus 2 is
modelled by its electromechanical equation. The changes in the load admittances
are taken as random thus the connected load operates as a random walk process.
The motor size and time constants are taken as random over a range. The passive
and active load admittances are first modelled as a set of random impedances;
they are then scaled by the random walk process to model the random switching
ON and OFF of load elements.
2Z1 23 41Z 3Z
Mz
Load #2
z
I
Load #1
Chapter 3. Correlation Based System Thevenin Impedance Estimation
57
This changing of the connected load give rise to variations in the angle of the
synchronous generator connected to bus 2. The changes in the angle and local
frequency of generator cause changes in voltage and frequency at the load bus.
The bus 3 frequency deviation is calculated as a proportion to the generator bus
frequency deviation (ref. to Appendix A). These changes create a dynamic
response in the induction motor model. This dynamic in turn will cause changes
in the composite load power, voltage and current.
The total load #1 admittance, generator angle and induction motor slip for a 0.04
sec sampling time and 10 sec of simulation time are shown in Figures 3.11, 3.12
and 3.13, respectively. Figure 3.14 shows the magnitudes and angles of the load
#1 voltage and current. The presence of the dynamic oscillation in load voltage
and current can easily be seen in Figure 3.14.
0 1 2 3 4 5 6 7 8 9 101
1.005
1.01
1.015
1.02
1.025
1.03
1.035
1.04
1.045abs(yload # 1)
Time(sec)
Adm
ittan
ce(p
u)
Figure 3.11 Simulation of load #1 admittance changes in case 3a
Chapter 3. Correlation Based System Thevenin Impedance Estimation
58
0 1 2 3 4 5 6 7 8 9 10-7.8
-7.6
-7.4
-7.2
-7
-6.8
-6.6
-6.4
-6.2del2
Time(sec)
Ang
le(d
eg)
Figure 3.12 Variation of generator angle in case 3a
0 1 2 3 4 5 6 7 8 9 100.0256
0.0258
0.026
0.0262
0.0264
0.0266
0.0268Motor slip
Time(sec)
Slip
(pu)
Figure 3.13 Variation of induction motor speed in case 3a
Chapter 3. Correlation Based System Thevenin Impedance Estimation
59
0 2 4 6 8 100.932
0.933
0.934
0.935
0.936load bus voltage magnitude
Vol
tage
(pu)
0 2 4 6 8 100.93
0.94
0.95
0.96
0.97
0.98load current magnitude
Cur
rent
(pu)
0 2 4 6 8 10-8
-7.5
-7
-6.5load bus voltage angle
Time(sec)
Ang
le(d
eg)
0 2 4 6 8 10144.2
144.4
144.6
144.8
145
145.2load current angle
Time(sec)
Ang
le(d
eg)
Figure 3.14 Magnitudes and angles of the load bus voltage and current
Equations (3.25) and (3.26) are applied to the load voltage and current
components. The difference of the voltage imaginary component and its auto-
correlation are shown in Figure 3.15. One- step prediction algorithm is applied
to the voltage difference and the predicted curve is subtracted from the actual
curve. The residual and its auto- correlation are shown in Figure 3.16. Equations
(3.23) and (3.24) were applied to the residuals. The estimated and actual
Thevenin impedances are compared in Table 3.4. The estimated values of the
system impedance are very close to the actual values because in this case there is
no random variation in the supply system. This agrees with the proposed
estimation theory.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
60
0 1 2 3 4 5 6 7 8 9 10-8
-6
-4
-2
0
2
4x 10-4 Voltage imaginary component
Time(sec)
Vol
tage
imag
(pu)
-8 -6 -4 -2 0 2 4 6 8 10-1
-0.5
0
0.5
1x 10-4 Auto(voltage imag)
lag(sec)
Cor
rela
tion
Figure 3.15. Difference of the voltage imaginary component and its auto-
correlation in case 3a
0 1 2 3 4 5 6-1
-0.5
0
0.5
1x 10
-4 voltage imag residual
Time(sec)
Res
idua
l(pu)
-8 -6 -4 -2 0 2 4 6 8 10-5
0
5
10
15
20x 10
-8 Auto(voltage imag residual)
lag(sec)
Cor
rela
tion
Figure 3.16. Difference of the voltage imaginary component and its auto-
correlation in case 3a after removing dynamic component
Chapter 3. Correlation Based System Thevenin Impedance Estimation
61
TABLE 3.4 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR A T=100 SEC SIMULATION TIME AND A TIME STEP TΔ = 0.04
IN CASE 3, NO CHANGES IN LOAD #2
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
0.0334 + 0.0710i
0.0667 + 0.1333i
0.0 + 0.1333i
0.0333 + 0.0707i
0.0665 + 0.1331i
-0.0000 + 0.1332i
0.0337 + 0.0706i
0.0668 + 0.1331i
0.0002 + 0.1330i
3.4.5 Case 3b: System with dynamics and random changes in supply system This case is similar to case 3a but random changes are also added to load #2 in
Figure 3.10. Simulation of load #1 and load #2 admittances is shown in Figure
3.17.
0 1 2 3 4 5 6 7 8 9 101
1.05
1.1
1.15
1.2abs(yload # 1)
Adm
ittan
ce(p
u)
0 1 2 3 4 5 6 7 8 9 100.34
0.36
0.38
0.4
0.42
0.44
0.46abs(yload # 2)
Time(sec)
adm
ittan
ce(p
u)
Figure 3.17. Simulation of load #1 and load #2 admittances in case 3b
The same procedure as case 3a is applied to the load bus voltage and current
components and the residuals, after removing the dynamics, are used to estimate
Chapter 3. Correlation Based System Thevenin Impedance Estimation
62
system Thevenin impedance. The estimated results are shown in Table 3.5.
Equation (3.23) fails but, the estimated values by (3.24) agree with the actual
values of the system impedance. The results confirm the developed theory for
estimation.
TABLE 3.5 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR A T=100 SEC SIMULATION TIME AND A TIME STEP TΔ = 0.04
IN CASE 3B (RANDOM CHANGES IN LOAD #2)
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
0.0334 + 0.0710i
0.0667 + 0.1333i
0.0000 + 0.1333i
0.0277 + 0.0660i
0.0571 + 0.1240i
-0.0057 + 0.1277i
0.0348 + 0.0690i
0.0689 + 0.1300i
0.0016 + 0.1329i
3.4.6 Case 4: Simulation with Swing Reference Bus In this case the angle of voltage in bus1 (reference bus) in Figure 3.5 rotates with
a constant velocity2. Dynamic component is also imposed to the system by
choosing the initial value of the bus 2 voltage angle different from its steady-
state value. A simulation of different phasors angles is shown in Figure 3.18.
0 20 40 60-150
-100
-50
0Reference bus angle
Del
ta1(
pu)
0 20 40 60-150
-100
-50
0
50Generator bus angle
Del
ta2(
pu)
0 20 40 60-150
-100
-50
0Load voltage angle
Del
ta3(
pu)
0 20 40 60-200
-150
-100
-50
0 Current angle
Cur
rent
ang
le(p
u)
Time(sec) Time(sec)
Figure 3.18. Reference bus angle changes and its reflection in other buses
2 This may happen if the measurement unit is not completely synchronized with the system.
Chapter 3. Correlation Based System Thevenin Impedance Estimation
63
As it can be seen in Figure 3.18 the reference bus angle changes are reflected on
the angles of other buses in system. These changes were removed from load bus
voltage and current and the result is shown in Figure 3.19.
0 20 40 60-10
-5
0
5 Voltage angle without rotationD
elta
3(pu
)
Time(sec)0 20 40 60
-42
-40
-38
-36
-34Current angle without rotation
Cur
rent
ang
le(p
u)
Time(sec)
Figure 3.19. Load bus voltage and angle after removing the effect of reference
bus rotation
The real and imaginary components of the load bus voltage and current, with
reference bus rotation effects and after removing these effects, were
differentiated and after removing the dynamic components the residuals were
used to estimate the system Thevenin impedance. The results for different
rotation sizes in the reference bus angle and two different cases, with and
without random changes in supply system, are shown in Tables 3.6 and 3.7,
respectively.
When there are no random changes in the supply system, Table 3.6 illustrates
that, equation (3.23) can be used to estimate the system Thevenin impedance in
either case, with reference bus rotation effects on the measured data and/or
removing those effects from data. However, equation (3.24) can only be used to
estimate the system impedance after removing the reference bus effects from
data.
In the case that there are random changes in the supply system, equation (3.23)
fails to estimate the Thevenin impedance, but, after removing the reference bus
Chapter 3. Correlation Based System Thevenin Impedance Estimation
64
effect, equation (3.24) can be used for most cases. The above simulation results
concede with the proposed theory.
TABLE 3.6 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR A 100 SEC SIMULATION TIME AND 0.04 SEC TIME STEPS FOR
DIFFERENT VALUES OF ROTATIONS IN REFERENCE BUS (BUS 1) ANGLE, WITHOUT
RANDOM CHANGES IN SUPPLY SYSTEM
)secdeg(1
dtdδ
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
-3.60
-36
-72
-180
0.0509 + 0.1167i
0.0509 + 0.1167i
0.0509 + 0.1167i
0.0509 + 0.1167i
*0.0483 + 0.1160i **0.0488 + 0.1151i
0.0507 + 0.1169i 0.0491 + 0.1129i
0.0518 + 0.1158i 0.0517 + 0.1139i
0.0551 + 0.1095i 0.0486 + 0.1052i
0.0025 + 0.1269i 0.0507 + 0.1173i
0.0230 + 0.0481i 0.0503 + 0.1159i
0.1128 + 0.2440i 0.0487 + 0.1149i
0.0178 + 0.0851i 0.0363 + 0.1233i
* Estimated value with reference bus rotation ** Estimated value after removing the effect of reference bus rotation
TABLE 3.7 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR A 100 SEC SIMULATION TIME AND 0.04 SEC TIME STEPS FOR
DIFFERENT VALUES OF ROTATIONS IN REFERENCE BUS (BUS 1) ANGLE, WITH RANDOM
CHANGES IN SUPPLY SYSTEM
)secdeg(1
dtdδ
thZ (actual) Estimated value using (3.23)
Estimated value using (3.24)
-3.60
-36
-72
-180
0.0509 + 0.1168i
0.0509 + 0.1167i
0.0509 + 0.1167i
0.0509 + 0.1169i
*0.0399 + 0.1099i **0.0403 + 0.1083i
0.0413 + 0.1095i 0.0438 + 0.1062i
0.0452 + 0.1065i 0.0442 + 0.1058i
0.0487 + 0.1020i 0.0427 + 0.0960i
0.0391 + 0.0959i 0.0542 + 0.1142i
0.1863 + 0.1388i 0.0572 + 0.1149i
0.0527 + 0.1717i 0.0566 + 0.1130i
-0.1703 + 0.0186i 0.0688 + 0.1098i
* Estimated value with reference bus rotation ** Estimated value after removing the effect of reference bus rotation
Chapter 3. Correlation Based System Thevenin Impedance Estimation
65
3.5 Application of the Proposed Method to Real Data3
The proposed method in this project was applied to the measured real data
obtained from a load bus in Brisbane. The magnitude and phase of the measured
load voltage and current for a sample 100 sec time frame are shown in Figures
3.20 and 3.21, respectively. The angle of the first data point in current is chosen
as reference. The measured current is related to one feeder of six different
feeders connected to the load bus. The load admittance magnitude and angle
connected to this feeder is shown in Figure 3.22.
In this study five other load feeders are considered as a part of the supply system
for which the Thevenin impedance is estimated.
0 10 20 30 40 50 60 70 80 90 1008165
8170
8175
8180
8185
8190Voltage magnitude
Vol
tage
(V)
0 10 20 30 40 50 60 70 80 90 1002490
2495
2500
2505
2510
2515current magnitude
Cur
rent
(A)
Time(sec)
Figure 3.20. Time measurement of load voltage and current magnitudes in a
residential feeder in South Pine substation, Brisbane
3 real data obtained from a phasor measurement unit in the Australian electric power system installed at major load centres by QUT
Chapter 3. Correlation Based System Thevenin Impedance Estimation
66
0 10 20 30 40 50 60 70 80 90 10010
11
12
13
14
15Voltage angle
Vol
tage
ang
le(D
eg)
0 10 20 30 40 50 60 70 80 90 100-3
-2
-1
0
1Current angle
Cur
rent
ang
le(D
eg)
Time(sec)
Figure 3.21. Time measurement of load voltage and current angles, angle of the
first data point in current is chosen as reference
0 10 20 30 40 50 60 70 80 90 1000.304
0.306
0.308
Load admittance magnitude
Adm
ittan
ce(m
ho)
0 10 20 30 40 50 60 70 80 90 100-15
-14
-13
-12Load admittance angle
Time(sec)
Ang
le(D
eg)
Figure 3.22. Real data load admittance magnitude and angle
The changes in the voltage real component and its auto-correlation are shown in
Figure 3.23. Voltage imaginary and current components also have similar
Chapter 3. Correlation Based System Thevenin Impedance Estimation
67
changes. One step prediction method was used to remove dynamic components
of voltage and current changes. The residuals of the changes in the voltage real
component and its auto-correlation are shown in Figure 3.24.
Equations (3.23) and (3.24) were applied to the residuals to estimate the system
Thevenin impedance. Sample results for 100 sec successive time frames of data
are shown in Table 3.8. As it can be seen in Table 3.8, equation (3.23) fails to
estimate the system Thevenin impedance. The real components of the estimated
values by this equation are all negative and the imaginary components are
inconsistence. The estimated results by equation (3.24) are close together and the
differences in the results are mostly due to the random changes in the other loads
connected to the load bus. The results obtained from real data agree with the
theory and simulation results.
0 10 20 30 40 50 60 70 80 90 100-4
-2
0
2
4Voltage real changes
Vr(V
)
Time(sec)
-4 -3 -2 -1 0 1 2 3 4-1000
0
1000
2000
3000 Auto(voltage real changes)
Aut
o of
Vr
Lag(sec)
Figure 3.22. Real data voltage real component changes and its auto-correlation
Chapter 3. Correlation Based System Thevenin Impedance Estimation
68
0 10 20 30 40 50 60 70 80 90 100-4
-2
0
2
4Voltage real residuals
Vr-r
esid
uals
(V)
Time(sec)
-4 -3 -2 -1 0 1 2 3 4-1000
0
1000
2000
3000Auto(voltage real residuals)
Aut
o of
Vr-r
esid
uals
Lag(sec)
Figure 3.23. Residuals of the real data voltage real component changes and its
auto-correlation
TABLE 3.8. ESTIMATED VALUES OF THE SYSTEM THEVENIN IMPEDANCE USING 100 SEC
SUCCESSIVE TIME FRAMES OF BRISBANE LOAD CENTRE MEASURED VOLTAGE AND
CURRENT PHASORS, STARTING AT 9 AM ON 2002/06/06
3.6 Summary
In this chapter, based on signal processing of the measured data in the local load
bus, a method was introduced to estimate power system Thevenin impedance.
Estimated value
using Eq. (3.23) (ohms)
Estimated value
using Eq. (3.24) (ohms)
-0.0452 + 0.0368i
-0.0465 + 0.0411i
-0.0345 + 0.0263i
-0.0389 + 0.0261i
0.3324 + 0.4993i
0.3389 + 0.4845i
0.3396 + 0.5083i
0.3433 + 0.5245i
Chapter 3. Correlation Based System Thevenin Impedance Estimation
69
Block diagram representation of the system was firstly used to develop the
theory. Then, different cases were discussed in detail by applying circuit theory
to the Thevenin equivalent circuit of the supply system. It was shown that the
ratio of the correlations of the changes in the load voltage and load admittance
with respect to the correlation of the changes in the load current and load
admittance can be used to estimate the system Thevenin impedance. The
required steps of the estimation algorithm were clearly shown.
The method was confirmed by simulation on a four bus test system. It was
shown that the dynamic components in the load voltage and current can be
removed and the residuals can then be used to estimate the system Thevenin
impedance.
The proposed method was also applied to the real data obtained from a load bus
in Brisbane load centre. The results agreed with the theory and simulation
results.
Chapter 4
On-line Load Characterization by Sequential Peeling
4.1 Introduction Load characteristics play an important role to the power system dynamic
stability. This is particularly true for voltage stability studies in which loads may
be subjected to large variations in system voltages and in which the short term
and long term load characteristics may come into play. Following a disturbance
in system, the dynamics of various load components and control mechanisms
tend to restore load power. The restored consumption may be beyond the supply
system capability causing voltage instability problem to the system [1].
Modeling of generators and transmission system has been well studied and many
models are now provided by manufacturers but the load behaviour is still one of
the biggest uncertainties in the prediction of voltage instability, and in spite of
Chapter 4. On-line Load Characterization by Sequential Peeling
71
significant attention given to load modelling by researchers, it remains a
challenge to properly develop load models which can be used with confidence in
studies.
A load model which is to be used in on-line voltage stability studies has the
following requirements [58]:
• It should be able to capture load behaviour when subjected to practical
variations in system voltages.
• Load model should be able to capture the effects of rotating load
dynamics such as motor stalling which may be the limiting factor in some
cases.
• It must be possible to derive the model from local load bus measured
data.
• In order to reduce the computational burden in the voltage stability
studies, the model should be simple enough.
• The model should be physically based.
In this chapter, considering the above requirements and based on the local load
bus data, a method is developed to characterize on-line the load behaviour. It is
shown that the measured load bus voltage and current during a disturbance in the
supply system can be used to identify parameters of a composite load consisting
induction motor, constant power and constant impedance load. The changes in
the load active power due to the disturbance are used to identify the active power
of the different load components in a peeling process. Then, the induction motor
reactive power is estimated using the random changes in the load active and
reactive power. The other components of the load reactive power are also
estimated using the disturbance data.
Chapter 4. On-line Load Characterization by Sequential Peeling
72
The theory is provided and the peeling algorithm is clearly shown. The proposed
method is validated by simulation on a test system.
4.2 Theory of the On-line Load Characterization Using Load Bus
Data
This section is devoted to the theory of the identification of a composite load
consisting of constant impedance, constant power and induction motor loads
(Figure 4.1). The induction motor can be considered as an equivalent to an
aggregate of induction motors.
Figure 4.1. One-line diagram a simple power system with composite load
The equivalent circuit of the system of Figure 4.1 is shown in Figure 4.2. Supply
system is shown by its Thevenin equivalent. X is the sum of the motor stator and
rotor inductances and the magnetizing circuit is ignored for simplification. rR is
the rotor resistance and s denotes to motor slip. The objective is to identify the
active and reactive power of the load components, using load bus measured
voltage and current [79].
Figure 4.2 Equivalent circuit of a power system with composite load
V
LR
LXconstant33 =+ jQP
X
22 jQP +
11 jQP +
thZ
jQP+
I
sRR r=
M Induction Motor
V
I Constant Power
Transmission
P+jQ Constant Impedance
G
Chapter 4. On-line Load Characterization by Sequential Peeling
73
4.2.1 Load active power peeling
It is assumed that the induction motor is a load with constant mechanical torque
with negligible stator losses. Following a disturbance, motor active power
restores to its predisturbance value[10]. In Figure 4.2, following a sudden
reduction in the magnitude of load bus voltage V ,due to a system disturbance,
the motor slip s increases, resulting reduction in R, until 1P matches its
predisturbance value (motor mechanical power) at time 2t (Figure 4.3). The
constant power load remains unchanged.
Figure 4.3 Active powers of different load components following a disturbance
in supply system
In Figure 4.3, PΔ , the difference of the load total power at times 2t and
−1t (predisturbance), is the final change in constant impedance load power due to
disturbance. We can write:
221
Ptt
Ptt
PP Δ==
−=
=Δ − (4.1)
Chapter 4. On-line Load Characterization by Sequential Peeling
74
Considering that the power of constant impedance load is proportional to the
squared voltage, one can write:
⎟⎟⎠
⎞⎜⎜⎝
⎛
=−
==Δ − 2
2
1
22 2 tt
Vtt
VKP pv (4.2)
Where, 2pvK is the voltage dependency coefficient of the constant impedance
load. Having measured voltage at different times, equations (4.1) and (4.2) can
be used to compute2pvK . Using this coefficient and the bus voltage magnitude,
then, 2P can be computed at any instant of time.
Peeling 2P from the total active power the residual power, resP , will be the sum
of 31 and PP . At the instant of disturbance 1t , the induction motor firstly acts as a
constant impedance load. Thus:
111
Ptt
Ptt
PP resresres Δ==
−=
=Δ +− (4.3)
Where:
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛
=−
==Δ +−
1
2
1
21 1 tt
Vtt
VKP pv (4.4)
Using equations (4.3) and (4.4) one can compute 1pvK coefficient and this
factor in turn can be used to calculate the motor active power at −= 1tt that is pre-
disturbance 1P . Peeling 1P from resP the reminder will be 3P ; the constant power
load active power.
4.2.2 Induction Motor Reactive Power Estimation The electrical characteristics of single cage induction motors are often
represented by the equivalent circuit shown in Figure 4.3, where sR (stator
Chapter 4. On-line Load Characterization by Sequential Peeling
75
resistance), sX (stator reactance), mX (magnetizing reactance), rR (rotor
resistance) and rX (rotor reactance) are motor different parameters. The rotor
slip is indicated by s . It is important to note the “slip” used in this model is the
frequency of the bus voltage minus the motor speed. Some programs incorrectly
use either average system frequency or 1.0 in place of the bus frequency.
Figure 4.4 Per-phase equivalent circuit of induction motor
In Figure 4.4, neglecting magnetizing current and stator resistance the induction
motor current and its complex power are given respectively by:
( )222
2
22 XsR
VXsjRs
Xs
R
VjXs
R
jXs
RVI
r
r
r
r
r +
−=
+⎟⎠⎞
⎜⎝⎛
⎟⎠⎞
⎜⎝⎛ −
=+
= (4.5)
222
22
222
2
XsR
VXsj
XsR
VRsjQPVIS
rr
r
++
+=+== ∗ (4.6)
where rs XXX += , and V is the voltage magnitude. Around the normal
operational region of low slip, i.e., 22 XsRr >> , P and Q in equation (4.6) can
be reduced to:
rRVs
P2
≈ (4.7)
2
22
rR
VXsQ ≈ (4.8)
jQP+V
I
mXsRr
SX rXSR
Chapter 4. On-line Load Characterization by Sequential Peeling
76
Having P already estimated, equations (4.7) and (4.8) could be used to estimate
Q if X was known, but this is not the case.
It was shown in section 3.2.1 that the changes in the local load bus data due to
the random changes in the supply system can be used to identify the load, and
vice versa. In this section it will be shown how the frequency dependent changes
in the local load bus measured data can be used to estimate the induction motor
reactive power.
Induction motor is the only frequency dependent component in the composite
load in Figure 4.4. Changes in the system frequency, f, will cause changes in the
induction motor real power (P) and reactive power (Q). The changes in P and Q
with respect to frequency are defined by:
ff qp kfQ,k
fP
=∂∂
=∂∂ (4.9)
Now using the standard relationship between motor slip and frequency,
fff
s r−= and around the normal operational region of low slip (i.e. ff r ≈ ),
( )
o
rrff
ff
ffffs 1
22≈=
−−=
∂∂ (4.10)
where 0f is the system nominal frequency. From equations (4.9) and (4.10) we
can write:
sP
ffs
sP
fPk
op f ∂
∂≈
∂∂
∂∂
=∂∂
=1 (4.11)
And,
sQ
ffs
sQ
fQk
oq f ∂
∂≈
∂∂
∂∂
=∂∂
=1 (4.12)
Substituting P from equation (4.7) in (4.11) yields:
Chapter 4. On-line Load Characterization by Sequential Peeling
77
21 VfR
kor
p f≈ (4.13)
Using equations (4.8) and (4.12) one can get:
22
2 VfR
Xskor
q f≈ (4.14)
An expression also can be easily found for the ratio of fpk and
fqk as below:
rp
q
RXs
k
k
f
f 2= (4.15)
Now equations (4.13)-(4.15) and the assumption of small slip can be used to
deduce expressions for P and Q from equations (4.7) and (4.8). These estimates
are specified by:
( )r
oqor
orrr R/X
fk
XfR
fRRXVs
RVs
P f
22222
=××
×=≈ (4.16)
( )rp
oqor
rorr R/Xk
fk
XfR
RXs
fR
VXs
R
VXsQ
f
f
4422
2
2
2
2
22
=××=≈ (4.17)
Equations (4.16) and (4.17) can be used to estimate the total induction motor
active and reactive power in a large system if rRX ratio is already known and
fpk and fqk can be estimated. However, having P already estimated, equation
(4.16) may be used to estimate rRX ratio, and then, to estimate Q using (4.17).
Alternatively, the following equation can be used to save the computation time,
and also, to lessen the estimation error. It can easily be shown from equations
(4.16) and (4.17) that:
f
f
p
q
k
kPQ2
= (4.18)
Having the estimated value of P, equation (4.18) can be used to estimate the
induction motor reactive power Q if the ff pq kk ratio is estimated.
Chapter 4. On-line Load Characterization by Sequential Peeling
78
4.2.3 fpk , fqk and ff pq kk Ratio Evaluation Using Load Bus Data
It has been confirmed in [8] that the costumer load changes are uncorrelated
white noise. It was shown in chapter 2 that the changes in the local bus voltage
and current, due to the changes in system, can be used to identify the local load,
and vice versa. Using this idea in chapter 3 the random changes in the local load
bus were used to estimate the system Thevenin impedance. Now it is shown how
the load active and reactive power changes, due to the random changes in the
supply system, can be used to evaluatefpk ,
fqk and the ff pq kk ratio.
In Figure 4.4, the constant impedance and constant power components of load
are not frequency dependent loads. Hence, fpk and
fqk coefficients defined by
(4.9) can respectively be considered as the load total active and reactive power
dependency coefficients with respect to frequency. Considering that the total
load power is a function of voltage and frequency one can write:
vf PPvvPf
fPP Δ+Δ=Δ
∂∂
+Δ∂∂
=Δ (4.19)
vf QQvvQf
fQQ Δ+Δ=Δ
∂∂
+Δ∂∂
=Δ (4.20)
In equations (4.19) and (4.20) the subscripts f and v denote the frequency
dependent and the voltage dependent component, respectively. Equations (4.9),
(4.19) and (4.20) can be used to show:
fP
ff
fP
fPk f
p f Δ
Δ=
Δ×⎟⎟⎠
⎞⎜⎜⎝
⎛Δ×
∂∂
=∂∂
=1 (4.21)
fQ
ff
fQ
fQk f
q f Δ
Δ=
Δ×⎟⎟⎠
⎞⎜⎜⎝
⎛Δ×
∂∂
=∂∂
=1 (4.22)
Chapter 4. On-line Load Characterization by Sequential Peeling
79
f
f
p
q
PQ
ffP
ffQ
fPfQ
k
k
f
f
Δ
Δ=
Δ∂∂
Δ∂∂
=
∂∂∂∂
= (4.23)
Using the random changes in the local load measured data the following
equations can be used to evaluatefpk ,
fqk and the ff pq kk ratio:
][][
ffEfPE
k fp f ΔΔ
ΔΔ= (4.24)
][][
ffEfQE
k fq f ΔΔ
ΔΔ= (4.25)
][][
ff
ff
p
q
PPEPQE
k
k
f
f
ΔΔ
ΔΔ= (4.26)
where E is the expected or mean value. Equations (4.24)-(4.26) can be used to
estimate the coefficients and their ratio if the frequency dependent components
are extracted from PΔ and QΔ , the changes in the load total active and reactive
power.
Correlating PΔ and QΔ with fΔ in equations (4.19) and (4.20) will eliminate the
effect of the voltage dependent components vPΔ and vQΔ if these components
are not correlated with the frequency changes. The problem is that there are
some correlation due to the correlation between fΔ and VΔ . The frequency
changes in the system cause the changes in the induction motor slip that in turn
cause changes in the load bus voltage. The changes in the load bus voltage also
influence the bus frequency through their effect on the bus voltage angle.
Therefore, removal of the voltage dependent components is the only way to use
equations (4.24)-(4.26) in the estimation process.
Chapter 4. On-line Load Characterization by Sequential Peeling
80
One of the methods that can be used to remove the undesired components from
the load bus power and frequency changes is described in the next section.
4.2.3 Removal of the Undesired Components from the Load Bus Data
Consider the system shown in Figure 4.5. A perturbation signal i(t) is applied to
a linear system with impulse response h(t) and the response of the system is y(t).
Figure 4.5 Block diagram of a system
Before y(t) can be measured, it is corrupted with noise n(t) which is due to the
other sources so that the observable signal is
( ) ( ) ( )tntytz += (4.27)
The time domain input-output relation for the linear system of Figure 4.5 is as
below [80]:
( ) ( ) ( ) ( )τ+−τ=τ ∫∞
iniiiz RdttR.thR)
110
1 (4.28)
where R denotes the correlation function. Assuming that i(t) and n(t) are not
correlated and/or their cross-correlation is ignorable comparing to the auto-
correlation of i(t), equation (4.28) reduces to:
( ) ( ) ( ) 110
1 dttR.thtR iiiz −τ= ∫∞
(4.29)
The frequency domain relationship for the system of Figure 4.5 is obtained by
Fourier transformation of equation (4.29):
( ) ( ) ( )ωΦω=ωΦ iiiz .H (4.30)
∑Input i(t)
Noise n(t)
y(t) z(t) + + h(t)
Chapter 4. On-line Load Characterization by Sequential Peeling
81
where H and Φ are the transfer function and the Fourier transform, respectively.
Equation (4.30) may be rewritten as
( ) ( )( )ωΦωΦ
=ωii
izH (4.31)
Equation (4.31) can be used to calculate and remove the undesired components
from the load bus power and frequency changes.
Now, considering vΔ as the input signal, vPΔ and vQΔ as two different outputs,
and PΔ and QΔ as the observable outputs, the following procedure can be used
to calculate and remove the voltage dependent components of the changes in the
load total active and reactive power:
1. Find the transfer functions from V to P and Q using the following
equations :
)()(
)(Hvv
vp
ωΦ
ωΦ=ω1 (4.32)
)()(
)(Hvv
vq
ωΦ
ωΦ=ω2 (4.33)
Where )(vp ωΦ and )(vq ωΦ are the Fourier transforms of cross-correlations of
V to P and Q, respectively and )(vv ωΦ is the Fourier transform of
autocorrelation of V.
2. Using the estimated )(H ω1 and )(H ω2 find vPΔ and vQΔ the load power
changes components associate with vΔ , in the time domain.
3. Subtract vPΔ and vQΔ from PΔ and QΔ , respectively.
vres PPP Δ−Δ=Δ (4.34)
vres QQQ Δ−Δ=Δ (4.35)
Chapter 4. On-line Load Characterization by Sequential Peeling
82
The residuals in equations (4.34) and (4.35) are fPΔ and fQΔ if there are no
changes associate with the local load. Otherwise, these changes should be
removed from the residuals.
4.2.4 Estimation other Components of the Load Reactive Power
A simulation of the load reactive power changes due to a disturbance is shown in
Figure 4.6. The post-disturbance induction motor reactive power 1Q is identified
using the above described method. At the instant of disturbance, 1t , the induction
motor also acts as a constant impedance load. We can write:
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛
=−
==
=−
==Δ +−+−
1
2
1
2
11 ttV
ttVK
ttQ
ttQQ qv (4.36)
Figure 4.6 Reactive power of different load components following a disturbance
in supply system
Equation (4.36) is used to evaluate the reactive power voltage dependency
Chapter 4. On-line Load Characterization by Sequential Peeling
83
coefficient qvK which can then be used in (4.37) to calculate the reactive power
of the constant power load:
−− =−
==
1
2
13
ttVK
ttQQ qv (4.37)
The estimated 1Q and 3Q are then peeled off from the load total reactive power at
the new steady-state time in Figure 4.6 to estimate the constant impedance load
reactive power :2Q
32
122
2 Qtt
Qtt
Qtt
Q −=
−=
==
(4.38)
The estimated active and reactive powers can then be used to obtain the load
parameters in Figure 4.2.
4.3 Algorithm of the Load Characterization by Sequential
Peeling
The procedure of the load characterization using the sequential peeling method is
explained in this section. Load is considered as a combination of induction
motor, constant impedance, and constant power loads.
Step1: Providing the initial local load bus data
Compute load bus active and reactive power and frequency and current
(simulation) and/or load admittance (real data) using VYI = equation in each
time step.
Step2: Peeling load active power
2.1. Use equations (4.1) and (4.2) to find 2pvK the voltage dependency
coefficient of the constant impedance load and then using (4.39) calculate its
Chapter 4. On-line Load Characterization by Sequential Peeling
84
predisturbance and post-disturbance steady-state active power; i.e., −= 12
ttP and
22 tt
P=
in Figure 4.3, respectively.
22 2
VKP pv= (4.39)
2.2. After peeling −= 12
ttP from the load total power, use equations (4.3) and
(4.4) to find 1pvK coefficient. Then, calculate the induction motor active power
using equation (4.40).
−− ==
==
1
2
11 1 tt
VKtt
PP pvm (4.40)
2.3. Use equation (4.41) to find the active power of the constant power load.
( ) −=−−=
1123
ttPPPP (4.41)
Step3: Induction motor reactive power estimation
3.1. Refining the local load bus data
Compute the changes in the load bus voltage, active and reactive power,
frequency and load admittance using the following equations:
( )∑−
=+ −=
1
11
n
kkk VVΔV (4.42)
( )∑−
=+ −=
1
11
n
kkk PPΔP (4.43)
( )∑−
=+ −=
1
11
n
kkk QQΔQ (4.44)
Chapter 4. On-line Load Characterization by Sequential Peeling
85
( )∑−
=+ −=
1
11
n
kkk ffΔf (4.45)
Where n is the number of the data points. This differentiation is required because
the estimation method uses the random changes in the load bus data.
3.2. Removing the dynamic components
The estimation method uses only the instantaneous changes in quantities. In this
thesis one-step prediction method is used to remove the dynamic components
from the data computed in the sub-step 3.1.
3.3. Removing the components associate with the voltage changes ΔV
Estimation method uses only the frequency dependent components of the
changes in the load bus power. Using equations (4.32) and (4.33), estimate the
transfer functions from load bus voltage to the active and reactive power. Then,
use the transfer functions to estimate the load power changes associate with the
voltage changes, in the time domain and subtract them from the load power total
changes.
3.4. Induction motor reactive power estimation
Estimate induction motor reactive power using equation (4.46):
( )( )ff
ff
PPEPQEPQ
ΔΔ
ΔΔ=
2 (4.46)
Where, E is the expectation value.
Step4: Estimation the other components of the load reactive power
Estimate reactive power of the constant power load using equations (4.36) and
(4.37). Then, use equation (4.38) to estimate the reactive power of the constant
impedance load.
Chapter 4. On-line Load Characterization by Sequential Peeling
86
4.4 Simulation
The proposed method is applied to the test system shown in Figure 4.7. Bus 1 is
assumed to be an infinite bus. The synchronous generator dynamics in bus 2 is
simulated by its electromechanical swing equation. Random changes in the
passive load #2 cause random changes in the voltage and current of bus 3.
Figure 4.7 Four bus test system
To simulate a disturbance, at a certain time the system impedances are changed
from one level to another level causing sudden changes in the quantities of load
#1. Figure 4.8 shows a simulation of the load#2 admittance and the supply
system Thevenin impedance from the bus 3 view point.
Load #1 is simulated as a combination of induction motor, constant impedance,
and constant power loads. The induction motor load is simulated as a constant
demand load; i.e., the motor dynamics restores the motor active power
consumption to its per-disturbance value. The bus 3 frequency deviation is
calculated as a proportion to the generator bus frequency deviation (Ref. to
Appendix B). A simulation of the bus 3 voltage and the induction motor slip is
shown in Figure 4.9. Figure 4.10 shows the simulated load total active and
reactive power and their components.
I
Z Load #2
1V 3V 4V 2V1Z 2Z 3Z
Z M Load #1
Infinite
Chapter 4. On-line Load Characterization by Sequential Peeling
87
Equations (4.39)-(4.41) are used to estimate the load active power components.
The comparisons of the estimated and actual active powers for the different
compositions of the loads and a 10% change in the system Thevenin impedance
due to the disturbance are shown in Table 4.1.
Figure 4.8 Simulation of load #2 admittance and system Thevenin impedance
Figure 4.9 Simulation of load bus voltage and induction motor slip
Chapter 4. On-line Load Characterization by Sequential Peeling
88
Figure 4.10 Simulation of load active and reactive powers
A one-step prediction algorithm was applied to the post-disturbance active and
reactive powers of load #1 and the bus 3 voltage and the predicted curves were
subtracted from the actual ones. The residuals of the load bus voltage and active
power and their autocorrelations are shown in Fig. 4.11. The method described in
the section 4.2.3 was applied to the residuals to extract the frequency dependent
components of the random changes in the load active and reactive powers.
Figure 4.12 shows the voltage dependent changes in the load active power and
its autocorrelation. The frequency dependent changes in the load active power
and its autocorrelation are shown in Figure 4.13. Using the extracted frequency
dependent components in the load power (4.) was used to estimate the induction
motor reactive power and the load other reactive power components were
anticipated on the similar way to the identification of the load active power
components. The estimated results and the actual values for different load
Chapter 4. On-line Load Characterization by Sequential Peeling
89
compositions are also compared in Table 4.1.
As it can be seen from Table 4.1, the estimated powers are very close to their
actual values.
TABLE 4.1
COMPARISON OF THE LOAD ESTIMATED AND ACTUAL POWERS FOR DIFFERENT
COMPOSITIONS OF LOADS AND A 10% CHANGE IN THE SYSTEM IMPEDANCE
1P 2P 3P 1Q 2Q 3Q
EST1.(PU)
ACT2.(PU)
ERR3 (%)
0.3007 0.3775 0.0498
0.3000 0.3780 0.0500
0.2254 -0.1336 -0.3428
0.0872 0.1889 0.0199
0.0870 0.1890 0.0200
0.2081 -0.0395 -0.5321
EST.(PU)
ACT.(PU)
ERR (%)
0.3999 0.2781 0.0498
0.4000 0.2778 0.0500
-0.0292 0.0990 -0.3175
0.1756 0.1826 0.0198
0.1728 0.1852 0.0200
1.6159 -1.4201 -0.8137
EST.(PU)
ACT.(PU)
ERR (%)
0.3013 0.2853 0.0996
0.3000 0.2862 0.1000
0.4226 -0.3069 -0.3900
0.0855 0.0962 0.0998
0.0861 0.0954 0.1000
-0.6846 0.8684 -0.2392
EST.(PU)
ACT.(PU)
ERR (%)
0.3004 0.3816 0.0002
0.3000 0.3822 0
0.1187 -0.1524 -
0.0855 0.1914 0.0001
0.0859 0.1911 0
-0.4589 0.1336 -
EST.(PU)
ACT.(PU)
ERR (%)
0.3007 0.2020 0.0496
0.3000 0.2023 0.0500
0.2437 -0.1415 -0.8898
0.0798 0.0008 0.0197
0.0804 0 0.0200
-0.7248 - -1.2701
EST.(PU)
ACT.(PU)
ERR (%)
0.2006 0.0025 0.0020
0.2000 0 0
0.2830 - -
0.0329 -0.0010 -0.0006
0.0325 0 0
1.1613 - -
1 EST: Estimated value 2 ACT: Actual value 3 ERR: Estimation error
Chapter 4. On-line Load Characterization by Sequential Peeling
90
Figure 4.11 (a) and (b): changes in the load bus voltage and its autocorrelation (c) and (d): load active power changes and its autocorrelation
Figure 4.12 (a) and (b): voltage dependent load active power changes and its
autocorrelation
Chapter 4. On-line Load Characterization by Sequential Peeling
91
Figure 4.13 (a) and (b): frequency dependent load active power changes and its
autocorrelation
The results of the estimation for a fixed initial load combination and different
system disturbance sizes are compared to the actual values in Table 4.2. As it is
seen in the Table 4.2, those values estimated based on the disturbance data are
estimated more accurate for the larger disturbances. The induction motor load
reactive power estimation, which is based on the signal processing in the load
bus data random changes, is also of reasonable accuracy.
Knowing load voltage, the estimated load power components can be used to
identify the load different parameters in Figure 4.2.
Chapter 4. On-line Load Characterization by Sequential Peeling
92
TABLE 4.2
COMPARISON OF THE LOAD ESTIMATED AND ACTUAL POWERS FOR DIFFERENT SIZES OF
THE SYSTEM DISTURBANCE (K)4
K (%) 1P 2P 3P 1Q 2Q 3Q
1
EST5
ACT
ERR
0.3047 0.2907 0.0492
0.3000 0.2946 0.0500
1.5685 -1.3223 -1.6226
0.0842 0.0977 0.0495
0.0832 0.0982 0.0500
1.1235 -0.4564 -0.9736
5
EST
ACT
ERR
0.3003 0.2920 0.0509
0.3000 0.2931 0.0500
0.0987 -0.3951 1.7245
0.0830 0.0978 0.0505
0.0837 0.0977 0.0500
-0.7673 0.1277 1.0348
10
EST
ACT
ERR
0.3005 0.2917 0.0492
0.3000 0.2914 0.0500
0.1719 0.1048 -1.6421
0.0840 0.0979 0.0495
0.0843 0.0971 0.0500
-0.3301 0.7937 -0.9854
20
EST
ACT
ERR
0.3002 0.2875 0.0501
0.3000 0.2878 0.0500
0.0561 -0.0918 0.1909
0.0848 0.0966 0.0501
0.0855 0.0959 0.0500
-0.7991 0.6526 0.1146
50
EST
ACT
ERR
0.3002 0.2769 0.0499
0.3000 0.2770 0.0500
0.0591 -0.0445 -0.1078
0.0896 0.0922 0.0500
0.0895 0.0923 0.0500
0.1611 -0.1211 -0.0647
100
EST
ACT
ERR
0.3000 0.2587 0.0499
0.3000 0.2586 0.0500
0.0006 0.0276 -0.1434
0.0981 0.0854 0.0500
0.0973 0.0862 0.0500
0.9002 -0.9658 -0.0860
4.5 Summary
Based on the measured load bus voltage and current phasors during a large
disturbance in the power system a method was introduced to identify on-line the
load parameters. It was shown that the measured disturbance data and the
random changes in the local load bus quantities can be used to estimate the real
and reactive powers of the different components of the load, individually. The
change in the load active power due to the disturbance was used to identify the
active power of the load different components in a peeling process. Then the 4 K is the percentage of increase in the power system Thevenin impedance due to the disturbance. 5 The abbreviations in the second column are the same as Table 4.1.
Chapter 4. On-line Load Characterization by Sequential Peeling
93
induction motor reactive power was estimated using the random changes in the
load measured power. The other components of the load reactive power was also
estimated using the disturbance data. The proposed method was confirmed by
simulation.
Chapter 5
On-line Estimation of the Remaining Time to a Long-
term Voltage Instability
5.1 Introduction Load power restoration in constant demand loads, such as TV’s and computers,
induction motors, on-load tap changer (OLTC) controlled loads and constant
energy heating loads, is the main cause of voltage collapse [33]. To stop the
evolution of an unstable scenario before its conclusion to a voltage collapse, the
time to identify the instability is a critical aspect. Many emergency control
measures to deal with voltage collapse are based on extensive off-line studies.
The feasibility of the supply system Thevenin impedance estimation and load
parameter identification using the changes in the load measured voltage and
current were pointed out in chapters 2 and 3, respectively. It was shown in
chapter 2 that the ratio of the cross-correlation of the changes in the load voltage
and current with respect to the changes in the load admittance can be used to
estimate the system Thevenin impedance. Based on the measured load bus
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
95
voltage and current phasors during a disturbance in the power system a method is
introduced in chapter 3 to identify on-line the load parameters. In that method
the change in the load active power due to the disturbance is used to identify the
active power of the load different components in a peeling process. Then the
induction motor reactive power is estimated using the random changes in the
load measured power. The other components of the load reactive power are also
estimated using the disturbance data. The estimated power components can be
used to obtain the load parameters.
In this chapter, methods are developed to anticipate voltage instability for a
system consisting of load behind an OLTC. The analysis theory is provided for
the constant impedance load case, and also, a composite load containing constant
power, constant impedance and induction motor loads. It is shown that the local
load bus measured voltage and current phasors can be used to identify a possible
voltage collapse resulting from a long-term voltage instability caused by the on-
load tap changer, and to estimate the time to such a collapse. It is assumed that
the supply system Thevenin impedance and the load parameters are identified,
prior to the OLTC operation. The proposed methods are confirmed by
simulation.
5.2 Long-term Voltage Instability Prediction Considering
Constant Impedance Load and OLTC
5.2.1 System Description Consider a simple model system shown in Figure 5.1, where the constant
impedance load is supplied through an OLTC. The Tap ratio is shown by n. E
and sZ are the supply system Thevenin equivalent voltage and impedance,
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
96
respectively. The OLTC transformer impedance is ignored. At time T, a major
disturbance causes a significant sudden increase in sZ (Figure 5.2a). This change
in the system impedance, in turn, will cause significant sudden drops
in PV,V andsp , the load primary and secondary voltages and real power,
respectively.
Figure 5.1. Simple system with OLTC and constant impedance load
OLTC attempts to restore the load side voltage to its setpoint value by decreasing
the tap ratio, n. These tap changes decrease the load impedance, seen from the
primary side (Figure 5.2a). Tap changes will stop in two cases: sV reaches its set
point value, and/or, Tap reaches its limit.
0 50 100
0.2
0.4
0.6
0.8
1abs(Zload & Zs)
(a) Time(sec)
Per
uni
t
ZloadZs
0 50 1000.75
0.8
0.85
0.9
0.95Real power
(b) Time(sec)
Per
uni
t
0 50 1000.85
0.9
0.95
1Vs, Vp, Tap position
(c) Time(sec)
Per
uni
t
TapVsVp
Figure 5.2. Changes in impedances, tap ratio, primary and secondary side
voltages and power
n:1VsE
IpVp
P+jQ
Zs ZL
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
97
However, it is possible that the load impedance becomes equal in magnitude to
the impedance of the equivalent system feeding the bus (Figure 5.3a). In this
case any further tap changes beyond this point will have reverse effect causing
decreases in the load side voltage and consequently load power (Figures 3b
&3c). In the critical point of impedance matching, the load voltage and
power, PVs and , will be on their maximum values [81].
Figure 5.3. Changes in impedances, Load voltage and power
Also, the changes in the real power due to tapping in the two sides of the critical
point will have different signs. Thus, each of the following approaches may be
used to identify the remaining time to a possible collapse:
5.2.2 Time to collapse estimation using impedance matching criteria
This approach is based on the comparison of magnitudes of the load primary side
view impedance and system Thevenin impedance.
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
98
For a constant impedance load and two successive tap ratios in Figure 5.1, one
can write:
2221
2
21 11
⎟⎟⎠
⎞⎜⎜⎝
⎛ Δ−=⎟⎟
⎠
⎞⎜⎜⎝
⎛ Δ−=⎟⎟
⎠
⎞⎜⎜⎝
⎛== +++
ii
i
i
i
Li
Li
P
P
nn
nnn
nn
Zn
ZnZ
Z
i
i (5.1)
where in is the tap ratio after ith tapping and nΔ is the size of each tap step and:
iP
PLiP I
VZnZ i
i== 2 (5.2)
In equation (5.2) PP IV and are the OLTC primary side measured voltage and
current. Putting i=0 in equation (5.1) and using equation (5.2), the initial tap ratio
can be computed.
Also it can be shown that after the ith tapping the value of pZ will be:
2
⎟⎟⎠
⎞⎜⎜⎝
⎛=
o
ipp n
nZZ
oi (5.3)
where op nZo
and are the initial values of the primary side view load impedance
and tap ratio, respectively. At the voltage collapse point the load impedance
becomes equal in magnitude to the supply system Thevenin impedance. Putting
SZZi p = in (5.3) we can compute cri nn = , the required tap ratio that
makes load impedance equal in magnitude to the system impedance:
op
Socr
Z
Znn = (5.4)
Voltage will not collapse if mincr nn ≥ where minn is the OLTC lower limit.
The number of the tap changes to collapse cri can be computed using (5.5):
nnn
i crocr Δ
−= (5.5)
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
99
The calculated cri should be rounded to the lower integer if it is between two
integers. The tap changing logic at time instant it is as following:
⎪⎪⎩
⎪⎪⎨
⎧
>−<Δ−
<+>Δ+
=+
otherwiseandif
andif
1
i
minio
ssi
maxio
ssi
in
nndVVnn
nndVVnn
n (5.6)
Where d and osV are half of the OLTC dead band and load voltage reference
value, respectively. There are two modes of OLTC operation. In sequential mode
tap changes starts after an initial fixed time delay and continues at constant time
intervals until the secondary side voltage error is brought back inside the OLTC
deadband, or until the tap limit is reached. The initial time delay is in the range
of 30-60 sec and the subsequent taping time intervals are usually around 10 sec.
The voltage error dead band is usually in the range of ± 1%-2%.
In non-sequential mode of operation there is no distinction between first and
subsequent taps and all time delays are given by the same formula [3].
Now it should be checked to see whether crn will exceed the tap limit or not. If
yes, the collapse will not happen. But if the answer is no, then other quantities
should be checked.
Using cri and on the magnitude of sV for crnn = may be computed and
checked to see whether it exceeds the reference value or not. If yes, collapse will
not happen, otherwise, system will experience a voltage collapse. It can easily be
shown that:
( ) ( )i
o
o
i
Ps
Ps
P
P
i
oosis ZZ
ZZ*
Z
Z*
nn
*VV+
+= (5.7)
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
100
Where i is the number of tap changes and iPZ is defined by (5.2). ( )osV is the
post-disturbance secondary side voltage, before the OLTC operation. Putting
crii = in (5.7) the secondary voltage at critical point can be computed.
In an alternative method one may compute the amount of the load power
(apparent, active or reactive) and compare it to its predisturbance value. For a
constant impedance load:
( )( )
( )( )
2
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛=
os
is
o
iV
V
PP
(5.8)
In equation (5.8) ( )oP is the post-disturbance load real power and before the
OLTC operation.
In the following subsection another method will be introduced that is based on
the changes in the load power due to the OLTC operation.
5.2.3 Time to collapse estimation using load power changes
At the voltage collapse point the load power is in its peak value. Therefore, the
changes in the power prior to this point are positive and after that point the
changes are negative. Hence, the tap ratio related to the sign changing point of
the load power changes will identify the time of voltage collapse.
Using a least-square based method the value of the third change can be estimated
using the values of the first and second changes in load power. The subsequent
changes are also estimated using each pair of the changes until the point that the
change in load power becomes negative.
It can be shown that for the algebraic equation
θ= xy (5.9)
an estimation for θ can readily be as following:
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
101
( ) YXXXˆ T1T −=θ (5.10)
[ ]Tβα=θ can be estimated Putting:
( ) [ ]T341 PPY ΔΔ= and ( )
⎥⎦
⎤⎢⎣
⎡ΔΔΔΔ
=12
231X PPPP
in (5.10). ( ) [ ]T452 PPY ΔΔ= can then
be computed putting ( )⎥⎦
⎤⎢⎣
⎡ΔΔΔΔ
=23
342XPPPP
in (5.9). iPΔ is the change in load active
power due to the ith tapping. This one-step ahead computation uses the estimated
power changes to estimate a new point and it continues until the estimation of
the jth point that jPΔ becomes negative. If at the (j-1)th point of tapping the tap
ratio exceeds the OLTC limit, collapse will not happen, otherwise, using the
estimated power changes the amount of the load power at this point should be
computed and be compared to the load pre-disturbance power. If the estimated
power exceeds the pre-disturbance power, collapse will not happen, because this
means that secondary voltage will reach its reference value prior to this point of
tapping and tapping process will stop.
5.2.4 The Algorithms Based on the theory developed in section 5.2, the procedures of the voltage
instability prediction for a system with a constant power load behind an OLTC
are explained in this section. The first method is based on the impedance
matching criteria and the second method uses the changes in the load power due
to the OLTC operation to predict possible long-term voltage instability.
A: Algorithm Based on the Impedance Matching Criteria
Step1: Estimate the supply system Thevenin impedance using the proposed
method in chapter 3.
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
102
Step 2: Putting i=0 in (5.1) and using (5.2) estimate the tap initial ratio.
Step 3: Estimate crn the required tap ratio and cri related number of taping that
make load impedance equal in magnitude to the system impedance using (5.4)
and (5.5), respectively and rounding the result to the lower integer, if it is
between two integers. If mincr nn ≤ load bus voltage is stable, otherwise go to
step 4.
Step 4: Using (5.7) and (5.8) compute the load power (apparent, active or
reactive) for the critical point crii = and compare it to its predisturbance value:
B: Algorithm Based on the Load Power Changes
Step 1: After the OLTC third operation, estimate the subsequent changes in load
power due to the successive OLTC operation in a one-step ahead method using
equations (5.9) and (5.10) as described in section 5.2.3. The estimation
continues until the estimated power change becomes negative. At the point of the
last positive estimated value if the tap ratio exceeds the OLTC limit, collapse
will not happen, otherwise go to the next step.
Step 2: Using the estimated power changes in step 1, compute the amount of the
load power at the point of the last positive estimated value and compare it to the
load pre-disturbance power. If the estimated power exceeds the pre-disturbance
power, collapse will not happen; otherwise voltage collapse will start by the next
OLTC operation if it is not avoided.
5.2.5 Simulation The proposed methods are applied to the system shown in Figure 5.1. The load
happenwillcollapseotherwise
happennotwillcollapsePnnP edisturbancprerc
→
⇒≥= − if
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
103
admittance is modelled as a constant impedance load plus a small randomly
changing component (Figure 5.4). OLTC quantities are set to the values shown
in Table 5.1. The time interval for all tap changes is assumed to be 10 sec.
Considering the values in Table 1; the tap limit will be reached by ten successive
tap changes. At the time instance T= 10 sec there is a significant increase in the
system impedance, and this in turn, causes sudden drops in the secondary voltage
and load power. OLTC starts to restore load voltage and power by successively
changing the tap ratio. Now let: .The system and load impedances and load real
power for three different values of k are shown in Figures 5.5, 5.6 and 5.7.
Figure 5.4. Simulation of load admittance
TABLE 5.1. OLTC TAP INITIAL RATIO, STEP SIZE, LOWER LIMIT, TIME DELAY AND DEAD
BAND.
on nΔ % minn TΔ (sec) d% 1 1.5 0.85 10 2±
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
104
Figure 5.5 (a): System and load impedances, (b): Load real power, k=2
Figure 5.6. (a): System and load impedances, (b): Load real power, k=2.5
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
105
Figure 5.7. (a): Changes in the system and load impedances,
(b): changes in load real power, k=3
The method described in section 3.3 was used to estimate post-fault system
impedance. Tap initial ratio and the number of taps to collapse were estimated
using equations (5.2) and (5.4), respectively. The results are shown in Table 5.2.
The estimated numbers of taps have been rounded to their lower integer
numbers.
TABLE 5.2. ACTUAL AND ESTIMATED SYSTEM IMPEDANCE, INITIAL TAP RATIO AND TAPS
TO COLLAPSE FOR DIFFERENT VALUES OF K IN (prefault)sk*Zt)(post faulsZ =
sZ (pu) on (pu)
k Actual
Estimated
A
E
Taps to collapse
2
0.1+0.4i
0.1+0.4i
1
0.9999
15
2.5
0.125+0.5i
0.125+0.5i
1
1.0005
10
3
0.15+0.6i
0.15+0.6i
1
0.9994
5
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
106
As it can be seen from Table 5.2, estimated values for the system impedance are
the same as the actual values and the estimated initial tap ratios are very close to
the actual value. In the first case (k=2) collapse will not happen because as it
was stated before OLTC limit will be reached by 10 tap changes. In the second
case (k=2.5) also collapse will not happen because in this case the critical tap
ratio is equal to the tap limit and OLTC operation will be stopped.
In the third case (k=3) the estimated taps to collapse is less than 10. Now it
should be checked to see whether OLTC operation will stop before the critical
point or not. Equation (5.8) is used to compute the load real power in the critical
point. The result is as following:
onncrnnP..P
===<= 0815420
Thus tap operation will continue and voltage will collapse.
The above method of voltage collapse identification is based on the estimated tap
initial position. The size of a tap step is usually in the range of 0.5%-1.5%. So,
any error in estimation of the tap initial ratio may result in an incorrect value
of on , and therefore, invalid identification of voltage collapse.
Now we use equations (5.9) and (5.10) to identify the voltage collapse. The sizes
of the actual and estimated changes in power for different values of k are shown
in Figures 5.8, 5.9 and 5.10. As it can be seen the estimated and actual values are
close together.
Estimated power changes show that the power changes due to tapping for k=2
and k=2.5 are still positive at tenth tapping ( minnn = ). This means that in these
cases collapse will not happen. But for k=3 the sign of the power change
becomes negative at the sixth tap change. Thus, in this case the fifth tap change
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
107
is the critical point.
Figure 5.8. Comparison of estimated and actual values of power changes due to
tapping, k=2
Figure 5.9. Comparison of estimated and actual values of power changes due to
tapping, k=2.5
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
108
Figure 5.10. Comparison of estimated and actual values of power changes due to
tapping, k=3
The values of the power changes at the first five tap changes are added to the
post-fault (before OLTC operation) power value to compute the load power at
the critical tap ratio. The result is as following:
onncrnnP..P
===<= 0815420
Thus tap operation will continue and voltage will collapse. In this method, for
identification of a voltage collapse one should wait until the time of the fourth
change in power. Therefore, it will be useful just in the cases that there is left a
lot of time to collapse.
Simulation results shown in Figures 5.5, 5.6 and 5.7 confirm the above achieved
results by the both proposed methods in this section.
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
109
5.3 On-Line Voltage Collapse Prediction Considering Composite
Load and ON-Load Tap Changer
5.3.1 System Description A simple power system is shown in Figure 5.11, where a combination of the
induction motor load, constant impedance load, and constant power load is
supplied through an OLTC. The supply system is shown by its equivalent
Thevenin circuit. The impedance of the OLTC transformer is ignored. The
induction motor load is modelled as a constant demand load where following any
changes in the load bus voltage the motor slip s is changed and hence the
variable resistance R is adjusted to a new value. X is the sum of the stator and
rotor reactances and the rotor resistance is shown by rR . For simplicity the stator
resistance and the magnetizing circuit are ignored.
Figure 5.11 Simple power system with composite load and on load tap changer
The Tap ratio is shown by n. At time T, a major disturbance causes a significant
sudden increase in the supply system impedance. This change in the system
impedance, in turn, causes significant sudden drops in the load primary and
secondary voltages and real power, respectively (Figure 5.12).
It is assumed that following any changes in the load voltage the constant power
load impedance is immediately adjusted according to equation (5.11).
X
LX
3V
22 jQP +LRI
2V1V
QjP+
1:n
thZ
33
23
3 jQP
VZ
−=
33 jQP +
Rs
Rr=
11 jQP +
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
110
33
23
3 jQPV
Z−
= (5.11)
The motor slip, s , increases, resulting reduction in R, until matches its
predisturbance value (motor demand) (Figure 5.12).
OLTC attempts to restore the load side voltage to its setpoint value by decreasing
the tap ratio. Each tap change causes a sudden change in the load voltage and
hence activates motor dynamics causing changes in the motor slip until motor
active power is recovered at a new steady state condition (Figure 5.12).
Figure 5.12 (a): tap position, primary and secondary voltages, (b): induction
motor slip, (c): load active powers, (d): load reactive powers
The taping process continues until the voltage at bus 3 is maximized. At this
point, while the active powers of the induction motor and constant power loads
are recovered, the active power of the constant impedance load, and hence, the
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
111
load total active power, are also maximized (Figure 5.13). After this point the
OLTC reverse action starts and the process of the load voltage restoration
becomes unstable. Tap changes will stop in two cases: the OLTC secondary side
voltage reaches its set point value, and/or, Tap reaches its limit.
Figure 5.13. Changes in the transformer secondary voltage and load power due
to taping
5.3.2 Tap initial ratio estimation Following any sudden change in the load voltage the induction motor acts
primarily as a constant impedance load, and therefore, right after each taping the
impedances of the induction motor and constant impedance components of load
remain constant but the impedance of the constant power component of load is
adjusted in real time according to (5.11). Assuming that the active and reactive
powers of the constant power load are identified, equation (5.12) can be used to
estimate the pre- and post-taping admittances of this component from the bus 2
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
112
view point [82]:
22
333
V
jQPY P
−= (5.12)
It can easily be shown that the tap initial ratio can be estimated using equation
(5.13):
+=
−=
−
−
=⎟⎟⎠
⎞⎜⎜⎝
⎛ Δ−
1
1
3
32
Tt
Tt
PTP
PTP
o
o
YY
YY
nnn
(5.13)
Where, 2VIYTP = is the total load admittance from the bus 2 view point and
1T is the time instant of the first taping. The size of each tap step nΔ is assumed
to be known, otherwise, using equation (5.13) for two successive taping, on and
nΔ can both be determined.
The tap changing logic is according to equation (5.6). In this paper the tap
operating mode is considered to be in sequence. This mode of operation consists
of a sequence of tap changes starting after an initial fixed or constant time delay
and continuing at constant time intervals. The first tap time delay, if not constant,
can be determined by the following formula:
mfdOLTC TTVdTT ++Δ
= (5.14)
Where, VΔ is the difference between the controlled and reference voltages, d is
half of the OLTC dead-band, dT is the maximum time delay of the inverse-time
characteristic, fT is the intentional time delay, and mT is the mechanical time
delay [6].
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
113
5.3.3 Taps to collapse estimation
Once the tap initial ratio is estimated, the following method can be used to
estimate the time to collapse.
Let thY,Y,Y,Y and321 to be defined in Figure 5.11 as the induction motor load,
constant impedance load, constant power load, and the supply system Thevenin
admittances, respectively, where:
23
12
3
11
V
PRXj
V
PY −= (5.15)
LL jRX
jBGY+
=−=1
222 (5.16)
23
32
3
33
V
Qj
V
PY −= (5.17)
ththth
th jBGZ
Y +==1 (5.18)
It can easily be shown in Figure 5.11 that:
22
222
3BG
InV th
+= (5.19)
Where n is the tap ratio and the other terms are as below:
2VYII thth += (5.20)
23
3122
V
PPGnGG th
+++= (5.21)
2
3
32
3
122
V
Q
V
PRXBnBB th +++= (5.22)
It can also be shown that:
22
23
1XR
VRP
+= (5.23)
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
114
Solving equation (5.23) for R will result in:
1
221
43
23
24
PXPVV
R−±
= (5.24)
It was found by investigation that the above solution is only valid with the plus
sign. Finally, replacing R in equation (5.22) from equation (5.24), and
considering that following each taping the induction motor active power 1P is
recovered in steady state to the motor constant demand mP , equation (5.19) can
be rewritten as follows:
223BG
InV th
+= (5.25)
Where:
23
322
V
PPGnGG m
th+
++= (5.26)
23
324
32
3
23
22
24 V
Q
XPVV
VXPBnBB
m
mth +
−+++= (5.27)
...,,,,ininn o 3210=Δ−= (5.28)
Considering different values for i in equation (5.28), equation (5.25) can be
solved for 3V until this variable is maximized at crii = . At this point if either of
the constraints of OLTC in equation (5.6) is not yet met, collapse will happen.
5.3.4 The Algorithm Assuming that following a disturbance, the system Thevenin impedance and the
load parameters have already been identified by the proposed methods in [6] and
[7], respectively; the following algorithm can be used to anticipate a possible
voltage collapse and to estimate time to such a collapse:
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
115
1. Estimate the tap initial position using equation (3).
2. with different values of n from equation (5.28) solve equation (5.25) for
the voltage magnitude 3V until it is maximized.
3. Check the OLTC constraints in equation (4) for the point which 3V is
maximized. Voltage will start to collapse in the next taping if neither the voltage
is recovered to its set point nor tap has exceeded its limit.
5.3.5 Simulation The proposed method was applied to a four bus test system (Figure 4). Bus 1 is
assumed to be an infinite bus. The synchronous generator dynamics in bus 2 is
simulated by its electromechanical swing equation. The bus 2 frequency
deviation is calculated as a proportion to the generator bus frequency deviation
[7]. The load in bus 3 is a composition of induction motor, constant impedance,
and constant power loads. The induction motor load is simulated as a constant
demand load; i.e., following any changes in the load bus voltage the motor
dynamics restores the motor active power consumption to its previous value in a
few seconds.
Following a disturbance, OLTC; that is simulated by its discrete-time logic
described in equation (5.6), restores load voltage and load power by changing the
tap ratio. Voltage collapse happens when the OLTC reverse control action starts,
i.e., when secondary voltage drops if the tap position n is decreased aiming at
raising the secondary voltage. It is assumed that the voltage and current at the
primary side of OLTC are measured and the load post-disturbance parameters
have been identified before the OLTC starts to operate.
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
116
Figure 5.14. Four bus test system with composite load
TABLE 5.3. OLTC TAP INITIAL RATIO, STEP SIZE, LOWER LIMIT, TIME DELAY, VOLTAGE
REFERENCE, AND DEAD-BAND.
on nΔ % minn OLTCT (sec) (pu)RV d% 1 1.5 0.85 20 0.98 2±
5.3.6 Taps to collapse estimation with different disturbance sizes To simulate a disturbance, at a certain time, 21 and ZZ are changed from one
level to another level causing sudden changes in the quantities of load. Now
defining k as the ratio of the supply system Thevenin impedance to its pre-
disturbance value; the magnitude of the voltage in bus 3 is shown in Figure 5.15
for three different values of k1 and the same pre-disturbance load composition.
In the first case (Figure 5.15a) the tap operation stops because the regulated
voltage reaches its set-point value. In the second case (Figure 5.15b) tap reaches
its lower limit minn and tap operation stops. In the third case (Figure 5.15c)
voltage will collapse if it is not avoided.
Three step algorithm of section 5.3.4 was used to estimate tap initial ratio and the
number of taps to collapse for the above cases. Gauss-Seidel method was used to
1 %e)disturbanc(preZ/et)disturbanc(post Zk thth −−=
Infinite Bus
1n
4V
I
M Z
1V1Z 2Z
3V
2V
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
117
solve equation (5.25). The results are shown in Table 5.4.
Figure 5.15. Simulation of the OLTC primary and secondary voltage changes,
(a): k=1.4, (b): k=2.5, (c): k=3
TABLE 5.4. ESTIMATED INITIAL TAP RATIO AND TAPS TO COLLAPSE FOR DIFFERENT
VALUES OF K IN %e)disturbanc(preZ/et)disturbanc(post Zk thth −−=
K Estimated
on (pu) Taps to
maxV Tap change
limit Voltage
Collapse?
1.40 1 29 10 No 2.50 1 12 10 No 3.00 1 4 10 Yes
Estimated values for the tap initial ratio in Table 5.4 are the same as its actual
value 1=on . In the first and second cases collapse will not happen because as it
Time (Sec)
(a)
(b)
(c)
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
118
was stated before OLTC limit will be reached by 10 tap changes. In the third
case (k=3) the estimated taps to maxV is less than 10. Now it should be checked
to see whether OLTC operation will stop before the critical point or not. The
Estimated secondary (bus 3) voltage for different tap numbers is shown in Figure
5.16. As it can be seen in Figure 5.16 the estimated maxV is not in the OLTC
dead band zone (Table 5.3), and hence, tap operation will continue and voltage
will collapse. Simulation results shown in Figure 5.15 confirm all above
estimated results.
Figure 5.16. Estimated OLTC secondary voltage for 10 successive tapings, k=3
5.3.7 Effects of the load and measurement uncertainties on the estimation In periods up to 10 seconds the changes in the customer composite load are
unpredictable (Figure 5.17). These small random changes and also, measurement
uncertainties may affect the results of the proposed estimation method in this
section. To investigate this matter random changes are added to the load
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
119
admittance (Figure 5.18).
Figure 5.17. Time measurements of load voltage and current magnitudes in the
Brisbane load bus.
Figure 5.18. Simulation of load admittance with random changes
The estimation results for the same load composition and the same disturbance
sizes in Table 5.4 are shown in Table 5.5. As it can be seen the estimated taps to
collapse are the same as Table 5.4 and the estimated tap initial ratios are very
close to the actual value 1=on .
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
120
TABLE 5.5. ESTIMATED INITIAL TAP RATIO AND TAPS TO COLLAPSE FOR DIFFERENT
DISTURBANCE SIZES AND RANDOM CHANGES IN THE LOAD
k Estimated on (pu) Taps to maxV Collapse
1.40 0.9983 29 No 2.50 0.9991 12 No 3.00 0.9998 4 Yes
5.3.8 Effect of the load composition on the voltage collapse Four different compositions for the total load were chosen. The constant power
load is the same for all compositions. The estimated taps to collapse following a
large disturbance (k=3.6) are compared in Table 5.6. It can be seen in Table 5.6
that, as it is expected, larger induction motor load in the composition makes
safer the system voltage in the long-term, but increases the risk of the short-
term voltage instability. This result can be explained using Figure 5.19.
Following a disturbance, the induction motor active and reactive power
restoration, due to the slip increase, may bring the load total consumption
beyond the supply system capacity, resulting in short-term voltage instability
(case 4). However, if the load voltage is stable in the short-term, then the
reductions in the induction motor reactive power, due to the successive tapings,
improves the voltage stability in the long-term, through cancelling the increase in
the constant impedance load consumption.
TABLE 5.6. COMPARISON OF THE TAPS TO COLLAPSE FOR DIFFERENT LOAD
COMPOSITIONS
Case
Percentage in total load (%)
Taps to Instability
Voltage Collapse
1 P1=15, P2=80, P3=5 3 Yes 2 P1=30, P2=65, P3=5 4 Yes 3 P1=45, P2=50, P3=5 5 Yes 4 P1=60, P2=35, P3=5 0 Yes(Short- term)
Chapter5. On-line Estimation of the Remaining Time to a Long-term Voltage Instability
121
Figure 5.19. Simulation of the load reactive powers, Induction motor: 60%,
Constant impedance: 35%, Constant power: 5%, k=2.9
5.4 Summary
In this chapter, based on the measured data in the local load bus, methods were
introduced to anticipate voltage instability for a system consisting of load behind
an OLTC. Theories were provided for the constant impedance load case, and
also, a composition of the constant power, constant impedance and induction
motor loads. It was shown that the local load bus measured voltage and current
phasors can be used to identify a possible voltage collapse resulting from a long-
term voltage instability caused by the on-load tap changer, and to estimate the
time to such a collapse. Two methods were developed for the first case based on
the impedance matching and the sudden changes in the load power due to the tap
operation, respectively. The load side voltage maximizing criterion was used to
anticipate the voltage instability in the system feeding a composite load behind
an OLTC. It was assumed that the supply system Thevenin impedance and the
load parameters are identified, prior to the OLTC operation. The proposed
methods were confirmed by simulation.
Induction motor load
Constant power load Constant impedance load
Chapter 6
Case Study: BPA Test System
6.1 Introduction Operating of power systems nearer to their physical limits has made the long
term dynamics a significant concern for many utilities in the world. As a
consequence, power system engineers have had increasing interest in advanced
simulations [6, 33, 83, 84]. CIGRE Task Force 38-02-08 tackled this problem
and compared methods and tools on different study cases proposed by the
members.
One the purposes of the task force was to provide the researchers or practitioners
with benchmarks or test systems to be used for the assessment of existing tools
or of new developments in the field of long term dynamics. The Task Force
gathered five test systems, three of them displaying voltage stability problems.
The so called "BPA1 test system" was presented by Canada. This system has
been used by researchers to illustrate mechanisms of voltage instability in time
domain simulation [1, 19, 31, 32, 85, 86] .
1 Bonneville Power Administration (http://www.bpa.gov)
Chapter 6. Case Study: BPA Test System
123
In order to investigate and validate the proposed methods in this research, they
are applied to BPA test system.
6.2 Test System Description Figure 6.1 shows a single line diagram of the BPA test system. It is based on the
system described in [87], with loads and generators outputs modified. A local
area is fed by a remote generation area through five transmission lines. Generator
G3 in the local area, in addition to the supplying a part of the active power,
regulates the voltage at bus 6, to maximize the transfer capacity of the five
transmission lines.
Figure 6.1. The BPA test system [87]
In this study, generator G1 is modeled as an infinite bus, and neglecting the
stator transformer voltages and rotor subtransient dynamics, generators G2 and
G3 are represented by third order models. Considering the time interval of
interest in long term voltage stability, this simplification is acceptable. Excitation
systems and integral type overexcitation limiters are also added to the generation
system. The generators parameters are the same as in [19]. The limiter
characteristic is shown in Figure 6.2. If the field current exceeds fdmax1I the
1
2
34 5 6
7
8 9 10
Local Area
OLTC
G1
G2
G3
Subtransmission Equivalent
Generation Area
Chapter 6. Case Study: BPA Test System
124
current is ramped down to its continuous limit with time delay dependent on the
level of field current. The limiter block diagram and parameters are provided
from [7].
Figure 6.2. Overexcitation limiter characteristic
Load at bus 10 is modeled as a composite of induction motor, constant power
and constant impedance loads. Ignoring the flux dynamics, induction motor
dynamics are modeled with the rotor motion equation [6]. Following any
changes in the load bus voltage, the motor dynamics restore the motor active
power. The other system load at bus 7 is modeled as constant impedance load.
Small random changes are also added to both loads.
The on-load tap changer (OLTC) supplying the composite load is modeled in
detail. OLTC is assumed to be EU1 sequential mode type with 60 seconds time
delay for the first tap movement [6]. In order to reduce the simulation time the
time delay for the subsequent tap movements is set to be 10 seconds. The
deadband is assumed to be %1± p.u. of the controlled bus voltage. Tap range is
15± steps, and the step size is 1%. Complete data of the system is provided in
Appendix C.
fdI
FCL1.05Ifdmax1 ×=
FCL1.6Ifdmax2 ×=
Time (sec) 30 0
FCL=Full Load
Chapter 6. Case Study: BPA Test System
125
The disturbance considered is the sudden change, without a fault, in the
impedance of the transmission corridor between bus 5 and bus 6, for example,
loss of one of the branches.
A simulation of the magnitude of the composite load admittance from the bus 9
view point, including the compensation admittance, for a 0.04 sec sampling time,
a disturbance at 50 sec time (5% increase in the transmission corridor
impedance), 1% tap step size, and 115 sec total simulation time is shown in
Figure 6.3. Figure 6.4 shows the tap ratio and magnitudes of bus 9 and bus 10
voltages.
At time 34 sec in the simulation the overexcitation limiter (OXL) limits the field
current of the generator G3 and voltage support of bus 6, the load area end of
transmission corridor, provided by this generator, is lost. As a consequent,
voltages at OLTC primary and secondary sides are slightly decreased, causing an
increase in the load admittance through increase of the induction motor slip and
constant power load admittance.
At time 50 sec in the simulation, a disturbance in the system causes sudden
decrease in the OLTC primary and secondary side voltages. The load admittance
has increased in response to the voltage reduction. The small increase in the load
voltages after disturbance is due to the voltage control applied by generator G2.
At time 110 sec, the OLTC has operated to recover the secondary side voltage
causing further decrease in the primary side voltage and an increase in the load
admittance, seen from the OLTC primary side. Any further operation of the
OLTC will reduce load voltage stability margin by decreasing the load
impedance magnitude and bringing it closer to the value of the system
impedance after tripping of the line.
Chapter 6. Case Study: BPA Test System
126
20 30 40 50 60 70 80 90 100 110
2.93
2.94
2.95
2.96
2.97
2.98
2.99
abs(Load addmittance)-bus 9
addm
ittan
ce(p
u)
Time (sec)
OLTC operation
Disturbance
overexcitationlimiter of G3
Figure 6.3. Simulation of load admittance changes in bus 10.
20 30 40 50 60 70 80 90 100 110
0.975
0.98
0.985
0.99
0.995
1
tap ratio,primary and secondary voltages
Time(sec)
tap ratioabs(V10)abs(V9)
OLTC operation
Disturbance
OXL(G3)
Figure 6.4. Tap ratio, magnitudes of bus 9 and bus 10 voltages
Chapter 6. Case Study: BPA Test System
127
6.3 Validation of the Proposed “System Thevenin Impedance
Estimation” Method
In this section, the proposed method for the system Thevenin impedance
estimation in this research work is applied to the BPA test system. The supply
system is viewed from bus 9, i.e., the primary side of OLTC.
A simulation of the magnitudes and angles of the bus 9 (OLTC primary side)
voltage and current is shown in Figure 6.5. The disturbance considered at time
50 sec is an increase, without fault, in the transmission corridor impedance
between bus 5 and bus 6. At time 110 sec OLTC operates to restore the load side
voltage. The dynamics and random changes in the magnitudes and angles of the
voltage and current are caused by supply system and load. These signals, for the
time span between disturbance and tap operation, are highlighted in Figure 6.6.
For the same time period, the real and imaginary components of the changes in
the bus 9 voltage and current phasors and their autocorrelations are shown in
Figures 6.7 and 6.8, respectively. The autocorrelations confirm the presence of
electromechanical dynamics in the signals. A simulation of the changes in the
load admittance and its autocorrelation is also indicated in Figure 6.9. The DC
offset shown by the autocorrelation in Figure 6.9 is due to the induction motor
load slip change in response to the voltage step caused by the disturbance in the
system.
A one-step ahead prediction algorithm was applied to the post-disturbance
changes in the real and imaginary components of bus 9 voltage and current
phasors and the estimated signals were subtracted from the actual ones. The
residuals are shown in Figure 6.10. The autocorrelations of residuals in Figure
6.11 confirm that the residuals are free from dynamic components.
Chapter 6. Case Study: BPA Test System
128
40 60 80 1000.98
0.985
0.99
0.995
1abs(voltage)-bus 9
Per
uni
t
40 60 80 1002.88
2.9
2.92
2.94
2.96abs(current)-bus 9
Per
uni
t
40 60 80 100-32.4
-32.3
-32.2
-32.1
-32
-31.9angle(voltage)-bus 9
Deg
Time (sec)40 60 80 100
-41.6
-41.4
-41.2
-41
-40.8angle(current)-bus 9
Deg
Time (sec)
Figure 6.5 Simulation of the magnitudes and angles of the bus 9 voltage and
current
60 80 100
0.985
0.9852
0.9854
0.9856
0.9858
Per
uni
t
abs(voltage)
60 70 80 90 1002.886
2.888
2.89
2.892
2.894
2.896
60 80 100-32.08
-32.07
-32.06
-32.05
-32.04
Deg
60 80 100
-41.02
-41
-40.98
-40.96
Time (sec)
abs(current)
angle(voltage) angle(current)
Time (sec)
Figure 6.6 Magnitudes and angles in bus 9 voltage and current from disturbance
until start of OLTC operation.
Chapter 6. Case Study: BPA Test System
129
20 40 60-4
-2
0
2
x 10-5imag(diff(V9))
20 40 60
-2
-1
0
1
2
x 10-4
20 40 60
-2
-1
0
1
2x 10-4
20 40 60
-4
-2
0
2
4
x 10-5 real(diff(I9))
imag(diff(I9))
Time (sec)Time (sec)
real(diff(V9))
Figure 6.7 Changes in the components of the bus 9 voltage and current from the
line trip disturbance until start of OLTC operation
-10 0 10-1
0
1
2
3
4x 10-7 Auto(voltage real)
lag(sec)-10 0 10-5
0
5
10
15x 10-8 Auto(current real)
lag(sec)
-10 0 10-2
0
2
4
6x 10-6 Auto(voltage imag)
lag(sec)-10 0 10-1
0
1
2
3x 10-5 Auto(current imag)
lag(sec)
Figure 6.8 Auto correlations of the post-disturbance changes in the components
of the bus 9 voltage and current phasors
Chapter 6. Case Study: BPA Test System
130
10 20 30 40 50 60
-1
0
1
2
3x 10-4
Time(sec)
diff(abs(Yload))
-8 -6 -4 -2 0 2 4 6 81
2
3
4x 10-5 Auto(Yload changes)
Lag(sec)
Figure 6.9 Post-disturbance changes in the load admittance magnitude and its
autocorrelation until start of OLTC operation
0 20 40 60-4
-2
0
2
4x 10-5 Residuals(voltage real)
0 20 40 60-4
-2
0
2
4x 10-4
0 20 40 60-4
-2
0
2
4x 10-5
Time(sec) 0 20 40 60-4
-2
0
2
4x 10-4 Residuals(current imagl)
Residuals(current real)
Residuals(voltage imag)
Time(sec)
Figure 6.10 Residuals of the post-disturbance changes in the components of the
load voltage and current until OLTC operation
Chapter 6. Case Study: BPA Test System
131
-10 0 10-5
0
5
10
15x 10-8
Auto(voltage real residu)
-10 0 10-2
0
2
4
6
8x 10-8 Auto(current real residu)
-10 0 10-1
0
1
2
3
4x 10-6 Auto(voltage imag residu)
lag(sec)-10 0 10-2
0
2
4
6x 10-6 Auto(current imag residu)
lag(sec)
Figure 6.11 Autocorrelations of the residuals of the post-disturbance changes in
the load admittance magnitude until OLTC operation
The above method of the residual extraction will be applied to the signals in
different simulations through this section.
The system Thevenin impedance estimation method is validated by choosing
different values for the system Thevenin impedance. This impedance may be
changed by:
• Applying a disturbance, for example losing one or two of the lines in the
transmission corridor between bus 5 and bus 6. Other sizes of disturbance
are also applicable.
• Changing the ratio of different tap changers, except the tap changer
between bus 9 and bus 10.
• Changing the size of the reactive compensations in buses 6, 7, and 8.
• Changing the load connected to bus 7.
Chapter 6. Case Study: BPA Test System
132
6.3.1 Supply System without Random Changes
In this case the load in bus 7 is simulated as a constant impedance load without
random changes. A simulation of the system Thevenin impedance and the
composite load impedance, connected to bus 10, both from the bus 9 view point,
is shown in Figure 6.12. The proposed algorithm is section 3.3 is applied to the
60 sec post-disturbance bus 9 voltage and current phasors to estimate the system
impedance. The results of the simulations for a 0.04 sec time step and different
values of Thevenin impedance are shown in Table 6.1. as it can be seen, in this
case the estimated system Thevenin impedances are very close to the actual
values.
20 40 60 80 100 120 140
0.15
0.2
0.25
0.3
0.35
Time(sec)
Per
uni
t
Magnitudes of system Thevenin and load impedances
Thevenin ImpedanceLoad Impedance
Figure 6.12 Simulation of the magnitudes of the system Thevenin and load
impedances from the bus 9 view point
Chapter 6. Case Study: BPA Test System
133
TABLE 6.1 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE, SUPPLY SYSTEM WITHOUT RANDOM CHANGES, TIME STEP=0.04
SEC
thZ (actual) Estimated value using equation (3.23)
Estimated value using equation (3.24)
0.0338 + 0.1513i
0.0355 + 0.1543i
0.0384 + 0.1656i
0.0271 + 0.1651i
0.0328 + 0.1508i
0.0315 + 0.1540i
0.0348 + 0.1651i
0.0244 + 0.1640i
0.0345 + 0.1515i
0.0349 + 0.1549i
0.0375 + 0.1674i
0.0285 + 0.1645i
6.3.2 Supply System with Random Changes
In this case random changes are added to the load in bus 7. Consequently, the
random changes in the bus 9 voltage and current are partly due to these random
changes in supply system, and partly due to the random changes in the composite
load connected to bus 10. A simulation of the admittances of both loads is
shown in Figure 6.13. The results of the simulations for a 0.04 sec time step and
the same values of Thevenin impedance in Table 1 are shown in Table 6.2.
20 40 60 80 100
2.92
2.94
2.96
2.98
3
3.02
3.04
3.06
3.08
Time(sec)
Per
uni
t
Magnitudes of local and remote load addmittances
abs(Yload)-Bus7
abs(Yload)-Bus10
Disturbance
OLTC operation
Figure 6.13 Simulation of local and remote load admittances
Chapter 6. Case Study: BPA Test System
134
TABLE 6.2 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE, SUPPLY SYSTEM WITH RANDOM CHANGES, TIME STEP=0.04 SEC
thZ (actual) Estimated value using equation (3.23)
Estimated value using equation (3.24)
0.0338 + 0.1513i
0.0355 + 0.1543i
0.0384 + 0.1656i
0.0271 + 0.1650i
0.0257 + 0.1444i
0.0229 + 0.1487i
0.0276 + 0.1594i
0.0150 + 0.1588i
0.0338 + 0.1491i
0.0332 + 0.1563i
0.0367 + 0.1682i
0.0263 + 0.1660i
As it can be seen from Table 6.2, the estimated result using equation (3.24) are
still of reasonable accuracy. However, the obtained results from equation (3.23)
are not quantify to the actual values, as it was expected.
The number of data points was doubled by reducing the sampling time to half.
The estimation results for system Thevenin impedance are compared to the
actual values in Table 6.3. As it was expected, the results are improved,
comparing to the estimated results in Table 6.2.
TABLE 6.3 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE, SUPPLY SYSTEM WITH RANDOM CHANGES, TIME STEP=0.02 SEC
thZ (actual) Estimated value using equation (3.23)
Estimated value using equation (3.24)
0.0338 + 0.1513i
0.0355 + 0.1543i
0.0384 + 0.1656i
0.0271 + 0.1651i
0.0258 + 0.1464i
0.0265 + 0.1487i
0.0288 + 0.1603i
0.0162 + 0.1571i
0.0344 + 0.1505i
0.0353 + 0.1566i
0.0376 + 0.1661i
0.0255 + 0.1659i
For the final simulation in this section, by choosing different values of the ratios
of random changes in two loads, the effect of the size of the random changes in
the supply system in estimation result is investigated. The results are shown in
Table 6.4.
Chapter 6. Case Study: BPA Test System
135
TABLE 6.4 COMPARISON OF THE ESTIMATED AND ACTUAL VALUES OF THE SYSTEM
THEVENIN IMPEDANCE FOR DIFFERENT RANDOM CHANGE SIZES FOR REMOTE LOAD (K),
SUPPLY SYSTEM WITH RANDOM CHANGES, TIME STEP=0.02 SEC
K
thZ (actual)
Estimated value using equation
(3.23)
Estimated value using equation
(3.24) 0.1
0.2
0.5
1
1.5
2
0.0355 + 0.1543i
0.0355 + 0.1543i
0.0355 + 0.1543i
0.0355 + 0.1544i
0.0355 + 0.1544i
0.0355 + 0.1544i
0.0326 + 0.1539i
0.0324 + 0.1532i
0.0255 + 0.1487i
0.0019 + 0.1335i
-0.0287 + 0.1133i
-0.0608 + 0.0900i
0.0343 + 0.1549i
0.0350 + 0.1535i
0.0358 + 0.1545i
0.0324 + 0.1553i
0.0372 + 0.1518i
0.0353 + 0.1485i K is the change in the remote load as a fraction of the change in the local load.
Table 6.4 indicates that the estimated results are affected by the size of the
supply system random changes. For large values of K, as it was expected,
equation 3.23 completely fails to estimate the Thevenin impedance and even for
some cases the components of the estimated values are negative. In fact,
increasing the size of the random changes in the remote load will increase their
contribution to the changes in the bus 9 voltage and current phasors.
The estimation error for equation 3.24, in the last case in Table 6.4 is less than
4% that is still of reasonable accuracy.
It is noticeable that the electrical distance between the remote load and the local
bus is also an important factor in the contribution of the supply system random
changes to the local load bus changes, i.e., the smaller the electrical distance the
bigger the contribution. In the BPA test system, the load connected to bus 7 is
separated of bus 9 by only two transformers and the subtransmission line
between bus 8 and bus 9, i.e., the remote load is electrically close to the local
load bus. In such cases, the estimation result can be improved by using more data
points by increasing the length of the data time frame, and/or, decreasing the
Chapter 6. Case Study: BPA Test System
136
sampling time. In this research work, the time frame is restricted to the time
difference between disturbance occurring time and the OLTC operation.
However, in the cases that the local load random changes are highly dominated
by the remote load random changes, the changes in the bus 9 voltage and current
caused by the OLTC operation may be used as an alternative to estimate the
system Thevenin impedance from the bus 9 view point.
6.3.3 Statistical Evaluation of the Estimated Thevenin Impedance
The proposed method in this thesis uses the normal random changes in the load
voltage and current to estimate the system Thevenin impedance. Hence, the
estimation result, imitating the data, should be normally distributed around the
actual value. To evaluate the integrity of the method, in this section different
results for one case are statistically evaluated. The histograms of the real and
imaginary components of the 200 estimated Thevenin impedances, for a 0.04 sec
simulation time step, are shown in Figure 6.14. Different parameters of the
histograms are shown in Table 6.5.
TABLE 6.5 STATISTICAL PARAMETERS OF THE COMPONENTS OF THE ESTIMATED
THEVENIN IMPEDANCES
Estimated component of Thevenin impedance
Mean value (pu)
Median (pu)
Standard deviation (pu)
Coefficient of variation (%)
Real component 0.0519 0.0517 0.0039 7.56 Imaginary component 0.1630 0.1631 0.0053 3.22
The coefficient of variation is defined to be: %100mean value
deviationstandardCV ×= .
As it is seen in Table 6.5, the mean values are very close to the medians, i.e., the
estimated values are almost normally distributed. The standard deviation of the
imaginary component is 3.22% of the mean value. Considering that the
imaginary component is the dominant component in the power system Thevenin
Chapter 6. Case Study: BPA Test System
137
impedance, the standard deviation of the estimated real component also can be
acceptable. However, both standard deviations can be decreased by improving
the estimation accuracy through increasing the length of the data window, as it
was shown in section 6.3.2.
0.04 0.045 0.05 0.055 0.060
10
20
30
40
50
0.145 0.15 0.155 0.16 0.165 0.17 0.175 0.180
20
40
60
Real component of the estimated thevenin impedance (pu)
Imaginary component of the estimated thevenin impedance (pu)
Num
ber o
f dat
a po
ints
Num
ber o
f dat
a po
ints
Figure 6.14 Histograms of the components of the estimated Thevenin impedance
The system actual impedance in this case is 0.0541 + 0.1640i. The mean values
of the errors in the real and imaginary components of the estimated impedances
are - 4.07% and - 0.61%, respectively. This implies a 1% mean error in the
magnitude of the estimated impedance that can still be decreased by decreasing
the simulation time step.
6.4 Validation of the Proposed “Load Characterization by
Sequential Peeling” Method
This section is devoted to the validation of the proposed load characterization
method in chapter 4. The so called peeling method is applied to the BPA test
Chapter 6. Case Study: BPA Test System
138
system. The composite load is connected to bus 10, i.e., the secondary side of
OLTC, but the voltage and current measurements are taken at bus 9, i.e., OLTC
primary side. However, to characterize the load, the voltage and current at bus
10 are required; otherwise OLTC should be included in the load model that is not
intended in this research work. Hence, it is assumed that OLTC parameters are
known.
In this section, the initial tap position in bus 9 is set to 1 and the load is
characterized by applying a disturbance, without a fault, to the system and using
the measured data at bus 9, from disturbance time to the OLTC first operation.
The load connected to bus 10 is simulated as a combination of induction motor,
constant impedance, and constant power loads. The induction motor load is
simulated as a constant shaft power demand load; i.e., the motor dynamics
restores the motor active power consumption to its per-disturbance value. Small
random changes are included in both system loads.
To get the voltage at bus 10, for each data point the voltage drop across the
OLTC transformer is subtracted from the measured voltage. Using this voltage
and the measured current the load power is calculated at each data point. A
simulation of the bus 10 voltage and the measured current magnitudes, for a 0.04
sec time step and 250 sec total simulation time, is shown in Figure 6.15. The
disturbance is considered to be the loss of one of the branches between bus 5 and
bus 6 at time 50 sec. Figures 6.16-6.19 indicate the simulated load total active
and reactive powers and their components. The effects of the disturbance and
OLTC operations on the load voltage, current and power can easily be seen in
Figures 6.15-6.19.
Chapter 6. Case Study: BPA Test System
139
50 100 150 200 2500.9
0.95
1
1.05Magnitude of load voltage-bus 10
Vol
tage
(pu)
0 50 100 150 200 2502.8
3
3.2
3.4
3.6
3.8Magnitude of load current-bus 10
Time(sec)
Cur
rent
(pu)
Disturbance OXL (G3)
Figure 6.15 Simulation of the load voltage and current magnitudes
0 50 100 150 200 2502.4
2.6
2.8
3
3.2Load total active power
Per
uni
t
0 50 100 150 200 2501.1
1.2
1.3
1.4
1.5Load total reactive power
Per
uni
t
Time(sec)
Figure 6.16 Simulation of the load total active and reactive powers
Chapter 6. Case Study: BPA Test System
140
Figure 6.17 Simulation of induction motor load power
0 50 100 150 200 2502
2.2
2.4
2.6
2.8Constant impedance load active power
Per
uni
t
0 50 100 150 200 2501
1.1
1.2
1.3
1.4Constant impedance load reactive power
Per
uni
t
Time(sec)
Figure 6.18 Simulation of the constant impedance load power
50 100 150 200
0.26
0.28
0.3
Induction motor load active power
Per
uni
t
50 100 150 200 2500.03
0.04
0.05
0.06
0.07
Induction motor load reative power
Per
uni
t
Time(sec)
Chapter 6. Case Study: BPA Test System
141
0 50 100 150 200 250-1
-0.5
0
0.5
1
1.5Constant power load active power
Per
uni
t
0 50 100 150 200 250-1
-0.5
0
0.5
1
1.5Constant power load reactive power
Per
uni
t
Time(sec)
Figure 6.19 Simulation of the constant power load
Equations (4.39)-(4.41) are used to estimate the load active power components.
The comparisons of the estimated and actual active powers for the different
compositions of the load are shown in Table 6.6.
To estimate the induction motor load reactive power, the frequency dependent
random changes in the load active and reactive powers are required. A one-step
prediction algorithm and the method described in the section 4.2.3 are used to
extract the frequency dependent components of the random changes in the load
post-disturbance active and reactive power. Changes in the load total power, the
random components after removing the dynamics and the voltage and frequency
dependent components of the active and reactive powers are shown in Figures
6.20 and 6.21, respectively.
Chapter 6. Case Study: BPA Test System
142
20 40 60-2
-1
0
1
2x 10-5
20 40 60-1
-0.5
0
0.5
1
1.5x 10-5
20 40 60-2
-1
0
1
2x 10-5
20 40 60-1
-0.5
0
0.5
1x 10-5
Active power changes Random component
Voltage dependen random component
Frequency dependent random component
Time(sec)Time(sec)
Figure 6.20 Changes in the load post-disturbance active power
20 40 60
-5
0
5
x 10-6 Reactive power changes
20 40 60
-4
-2
0
2
4
6x 10-6
20 40 60
-5
0
5
x 10-6
Random component
Time(sec)20 40 60
-10
-5
0
5
x 10-6Frequency dependent
random component
Time(sec)
Voltage dependent random component
Figure 6.21 Changes in the load post-disturbance reactive power
Chapter 6. Case Study: BPA Test System
143
Equation (4.46) is used to estimate the induction motor reactive power. The other
components of the load reactive power are anticipated on the similar way to the
components of the load active power. The estimated results and the actual values
for different load compositions are also compared in Table 6.6. As it can be seen,
the estimated results for most of the components are quantified to the actual
values. The induction motor load reactive power estimation, which is based on
the signal processing in the load bus data random changes, is also quantified.
TABLE 6.6 COMPARISON OF THE LOAD ESTIMATED AND ACTUAL POWERS FOR
DIFFERENT COMPOSITIONS OF LOADS, DISTURBANCE IS THE LOSS OF ONE OF THE
BRANCHES BETWEEN BUS 5 AND BUS 6, P1 & Q1: INDUCTION MOTOR LOAD,
P2 & Q2: CONSTANT IMPEDANCE LOAD, P3&Q3: CONSTANT POWER LOAD
1P 2P 3P 1Q 2Q 3Q
EST2.(PU)
ACT3.(PU)
ERR4 (%)
0.4006 1.7338 0.2000
0.4000 1.7344 0.2000
0.1519 -0.0332 -0.0164
0.1463 0.8624 0.0000
0.1414 0.8673 0
3.4563 -0.5650 -
EST.(PU)
ACT.(PU)
ERR (%)
0.2996 1.7435 0.3000
0.3000 1.7431 0.3000
-0.1497 0.0263 -0.0034
0.0726 0.8734 0.0500
0.0747 0.8712 0.0500
-2.8695 0.2448 0.0224
EST.(PU)
ACT.(PU)
ERR (%)
0.5012 2.1248 -0.0000
0.5000 2.1261 0
0.2449 -0.0569 -
0.2377 1.0711 0.0000
0.2458 1.0630 0
-3.3095 0.7625 -
EST.(PU)
ACT.(PU)
ERR (%)
0.6015 1.4998 0.1000
0.6000 1.5013 0.1000
0.2554 -0.1032 0.0182
0.3562 0.7677 0.1000
0.3732 0.7506 0.1000
-4.5676 2.2742 -0.0243
EST.(PU)
ACT.(PU)
ERR (%)
0.2996 1.7435 0.3000
0.3000 1.7431 0.3000
-0.1497 0.0263 -0.0034
0.0726 0.8734 0.0500
0.0747 0.8712 0.0500
-2.8695 0.2448 0.0224
The results of the estimation for a fixed initial load combination and different
2 EST: Estimated value 3 ACT: Actual value 4 ERR: Estimation error
Chapter 6. Case Study: BPA Test System
144
system disturbance sizes are compared to the actual values in Table 6.7. As it is
seen, those values estimated based on the disturbance data are estimated more
accurate for the larger disturbances. The estimated induction motor reactive
powers are also close enough to the actual values, considering its direct
proportion to the already estimated induction motor active power, according to
equation (4.46).
TABLE 6.7 COMPARISON OF THE LOAD ESTIMATED AND ACTUAL POWERS FOR
DIFFERENT SIZES OF THE SYSTEM DISTURBANCE.
Voltage step (%)
1P 2P 3P 1Q 2Q 3Q
1.5
EST
ACT
ERR
0.3066 1.9704 0.2999
0.3000 1.9768 0.3000
2.1837 -0.3258 -0.0369
0.0682 0.9850 0.0501
0.0650 0.9883 0.0500
4.9212 -0.3305 0.2131
3
EST
ACT
ERR
0.2966 1.9229 0.3002
0.3000 1.9196 0.3000
-1.1392 0.1702 0.0502
0.0679 0.9587 0.0500
0.0671 0.9596 0.0500
1.2443 -0.0887 0.0332
4.5
EST
ACT
ERR
0.3028 1.8590 0.3000
0.3000 1.8618 0.3000
0.9224 -0.1502 0.0106
0.0707 0.9294 0.0500
0.0694 0.9307 0.0500
1.8481 -0.1366 -0.0223
6
EST
ACT
ERR
0.3014 1.8016 0.3000
0.3000 1.8030 0.3000
0.4744 -0.0786 -0.0021
0.0730 0.8666 0.0500
0.0720 0.9014 0.0500
1.3948 -3.8598 0.0345
7.5
EST
ACT
ERR
0.2996 1.7435 0.3000
0.3000 1.7431 0.3000
-0.1497 0.0263 -0.0034
0.0726 0.8734 0.0500
0.0747 0.8712 0.0500
-2.8695 0.2448 0.0224
6.4.1 Statistical Evaluation of the Estimated Power Components
In the proposed method of load power peeling, the induction motor load reactive
power is estimated using the correlation of the random changes in the load
reactive and active power and the other load power components are estimated
Chapter 6. Case Study: BPA Test System
145
using the changes in the load total active and reactive power caused by the
disturbance. However, due to the unpredictable random components in the load
voltage and current, the load active and reactive power also contain such
random components, as it can be seen from Figures 6.20 and 6.21. Hence, all
estimated power components are statistically random results. In this section, the
statistical characteristics of the estimated load power components are
investigated. The system loading condition is similar to the last case in table 6.6
and the disturbance considered is loss of a transmission line between bus 5 and
bus 6. The histograms of the 200 estimations of the load power components are
shown in Figures 6.22-6.24. Different parameters of the histograms are show in
Table 6.8.
0.297 0.298 0.299 0.3 0.301 0.3020
10
20
30
40
50
0.0716 0.0718 0.072 0.0722 0.0724 0.07260
10
20
30
Num
ber o
f dat
a po
ints
Num
ber o
f dat
a po
ints
Induction motor load estimated reactive power (pu)
Induction motor load estimated active power (pu)
Figure 6.22 Histograms of the estimated induction motor load active and reactive
power
Chapter 6. Case Study: BPA Test System
146
1.7655 1.7661.7665 1.7671.7675 1.768 1.7685 1.7691.7695 1.770
10
20
30
40
0.885 0.8852 0.8854 0.8856 0.8858 0.886 0.88620
10
20
30
40
Num
ber o
f dat
a po
ints
Constant impedance load estimated active power (pu)
Num
ber o
f dat
a po
ints
Constant impedance load estimated reactive power (pu)
Figure 6.23 Histograms of the estimated constant impedance load active and
reactive power
0.2999 0.2999 0.3 0.3 0.30010
20
40
60
80
0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.050
20
40
60
Num
ber o
f dat
a po
ints
Constant power load estimated active power (pu)
Num
ber o
f dat
a po
ints
Constant power load estimated reactive power (pu)
Figure 6.24 Histograms of the estimated constant power load active and reactive
power
Chapter 6. Case Study: BPA Test System
147
TABLE 6.8 STATISTICAL PARAMETERS OF THE HISTOGRAMS OF THE COMPONENTS OF
THE ESTIMATED THEVENIN IMPEDANCES
Estimated power component
Mean value (pu)
Median (pu)
Standard deviation (pu)
Coefficient of variation (%)
P-induction motor 0.3000 0.2999 9.7829e-004 0.3261 Q-induction motor 0.0720 0.0720 2.3269e-004 0.3231
P-constant impedance 1.7680 1.7680 8.9899e-004 0.0508 Q-constant impedance 0.8855 0.8855 2.5128e-004 0.0284
P-constant power 0.3000 0.3000 2.5960e-005 0.0087 Q-constant power 0.0500 0.0500 1.1468e-005 0.0229
The coefficient of variation is defined to be: %100mean value
deviationstandardCV ×= .
Referring to Table 6.8, all histograms are normally distributed with very small
standard deviations.
The mean values of the estimation errors in this experiment are shown in Table
6.9. The accuracy of the induction motor load reactive power estimation can be
improved by increasing the length of the data window through decreasing the
simulation time step.
TABLE 6.9 MEAN VALUES OF THE ESTIMATION ERROR IN THE DIFFERENT COMPONENTS
OF THE LOAD ACTIVE AND REACTIVE POWER
Load component Active power (%)
Reactive power (%)
Induction motor load -0.0142 -2.0849 Constant impedance load 0.0051 0.1784 Constant power load 0.0004 0.0002
6.5 Validation of the Proposed “On-line Estimation of the Time
to a Long-term Voltage Instability” Methods
In this section the proposed methods in chapter 5 for the on-line identification of
possible long-term voltage instability are validated by applying the methods to
the BPA test system. The load connected to bus 10 is considered to be a
composite of the induction motor, constant impedance and constant power loads.
Chapter 6. Case Study: BPA Test System
148
At time 50 sec, a disturbance is applied to the system and the post-disturbance
load voltage and current phasors are measured at bus 9, i.e., the OLTC primary
side. At time 110 sec OLTC operation starts and the time delay for the
subsequent tap movements is set to be 10 seconds.
6.5.1 Maximizing OLTC Secondary Side Voltage Criterion
Consider the equivalent system in Figure 6.25. It was shown that how the post-
disturbance measured voltage and current at bus 10 can be used to estimate the
system Thevenin impedance and the load power components. Once the load
power is peeled off, the load parameters are identified using the bus 10 voltage,
which is computed using equation (6.1):
InjXnV
V oto
×−= 910 (6.1)
where, on is the tap initial position.
Figure 6.25 Equivalent of the BPA test system with OLTC transformer and
composite load in bus 10.
Following the disturbance, OLTC successive operations tend to restore the
voltage at bus 10 its set point value. Voltage at bus 9 becomes unstable if OLTC
inverse control action starts. One of the proposed methods in this research work
uses the load bus voltage maximizing as the criterion of the voltage instability
starting point.
9V
I
thE
QjP+
thZ
OLTC transformer
1:n tXjjX
LjX
10V
22 jQP +LR
R srR
=
11 jQP +
33
210
3 jQP
VZ
−=
33 jQP +
Chapter 6. Case Study: BPA Test System
149
The voltage at bus 10 is a function of the Thevenin voltage, tap ratio and load
impedance. Taking into account the OLTC impedance, some changes are
required in equations (5.25)-(5.27). The system equations are as below:
ts IYYYYV ×+++= 32110 (6.2)
where,
210
12
10
11
V
PRXj
V
PY −= (6.3)
LL jRXY
+=
12 (6.4)
210
32
10
33
V
Qj
V
PY −= (6.5)
tths
XjnZnY 2
2
+= (6.6)
tth
tht
XjnZ
nEI 2+= (6.7)
IZVE thth += 9 (6.8)
Following each tap operation the induction motor active power 1P is recovered
in steady state to the motor constant demand mP . It was shown in chapter 5 that:
m
mP
XPVVR
24 224
102
10 −+= (6. 9)
...,,,,ininn o 3210=Δ−= (6.10)
In equation (6.10), on and nΔ are the tap initial ratio and tap step size,
respectively. It can easily be shown that setting the transformer reactance tX to
zero, equation (6.2) changes to equation (5.25).
Considering different values for i in equation (6.10), equation (6.2) can be solved
Chapter 6. Case Study: BPA Test System
150
for 10V until this variable is maximized if the magnitude of thE in equation (6.8)
remains constant through the OLTC operation. However, due to the voltage
control provided by the generators, the magnitude of thE changes, causing
problem to the voltage instability anticipation. A simulation of the thE
magnitude is shown in Figure 6.26.
50 100 150 200
1.2
1.22
1.24
1.26
1.28
1.3
1.32
Time(sec)
Vol
tage
(pu)
Magnitude of Thevenin Voltage
Figure 6.26 Simulation of the Thevenin voltage magnitude
To solve the problem, using the measured voltage and current at bus 9, the
magnitudes of thE , at steady state, are computed for the tap initial position and
its first and second operations. Then, evaluating a curve fitted to the computed
values, the thE magnitude is estimated for the other OLTC operations. The
estimated and actual values of the thE magnitudes are compared in Figure 6.27.
Chapter 6. Case Study: BPA Test System
151
2 4 6 8 10 12 141.223
1.224
1.225
1.226
1.227
1.228
1.229
1.23Thevenin Voltage - estimated and actual values
Tap number
Vol
atge
(pu)
ActualEstiamted
Figure 6.27 Comparison of the estimated and actual values of the system
Thevenin impedance
Using the estimated voltage Thevenin magnitudes, the Gauss-Seidel method is
used to solve equation (6.2) for the successive tap positions until the magnitude
of voltage at bus 10 is maximized. The voltage instability at bus 9 will start in
the next tap operation if neither the voltage is recovered to its set point nor tap
has exceeded its limit. In any case, the estimated voltage tracing is limited to the
tap position limit point. The voltage is stable if the estimated voltage is still
increasing at this point.
6.5.2 Impedance Matching Criterion In this section, the magnitude of the estimated load impedance, from the bus 9
view, is compared to the magnitude of the system Thevenin impedance as a
method to determine voltage stability. At the point of instability, the magnitude
of the load impedance is equal to and/or smaller than the system impedance
Chapter 6. Case Study: BPA Test System
152
magnitude. Equation (6.11) is used to estimate the load impedance:
⎟⎟⎠
⎞⎜⎜⎝
⎛++
+=321
2 1YYY
jXnZ tL (6.11)
where, 321 YandY,Y are computed using equations (6.2)-(6.10). The comparison
of the load and system impedances is limited to the OLTC limit point.
6.5.3 Simulation Results
Three cases with different pre-disturbance conditions are considered. The
disturbance considered at time 50 sec is a sudden change in the transmission
corridor between bus 5 and bus 6. At time 110 sec OLTC operation starts. In all
cases, the initial tap position and the tap step size are 1, and 1%, respectively.
The reference voltage at bus 10 is set to 1, and the deadband is 010.± p.u. The
tap range is 15± steps for the two first cases and 20± steps in the third case. It is
assumed that the OLTC transformer impedance and the initial tap position are
known.
Figures 6.28-6.30 indicate the time responses of the voltages at bus 9 and bus 10,
and the tap ratio of transformer between the two buses. Referring to Figure 6.28,
in first case the OLTC secondary side voltage is restored to the dead band of the
reference voltage by 8 tap successive movements and OLTC operation has
stopped. In the second case, OLTC operation is limited by the tap lower limit
and the voltage of bus 10 is still out of the reference voltage dead band, as it can
be seen from Figure 6.29. Figure 6.30 shows voltage instability that starts after
tap 13th movement. As seen in the figure, at this point the bus 10 voltage is
maximized.
Chapter 6. Case Study: BPA Test System
153
0 50 100 150 200 250 3000.9
0.92
0.94
0.96
0.98
1
1.02
1.04Tap position, OLTC primary and secondary volatges
Time (sec)
TapBus 9Bus 10
Figure 6.28. Simulation of the OLTC primary and secondary side voltages and
tap ratio in case1
0 50 100 150 200 250 3000.8
0.85
0.9
0.95
1
1.05
Time(sec)
Tap position, OLTC primary and secondary volatges
TapBus 9Bus10
Figure 6.29. Simulation of the OLTC primary and secondary side voltages and
tap ratio in case2
Chapter 6. Case Study: BPA Test System
154
50 100 150 200 250 300
0.76
0.78
0.8
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
Time(sec)
OLTC primary and secondary volatgesV
olta
ge(p
u)
Bus 9Bus10
Figure 6.30. Simulation of the OLTC primary and secondary side voltages in
case3
The 60 sec post-disturbance measured voltage and current phasors at bus 9 are
used to estimate the pre OLTC operation system Thevenin impedance and to
peel off the load power components. The results are compared to the actual
values in Tables 6.10 and 6.11, respectively.
TABLE 6.10
COMPARISON OF THE SYSTEM POST-DISTURBANCE ESTIMATED AND ACTUAL THEVENIN
IMPEDANCE FOR DIFFERENT CASES
thZ (actual) Estimated value
Case 1 0.0437 + 0.1645i 0.0420 + 0.1669i Case 2 0.0455 + 0.1671i 0.0393 + 0.1646i Case 3 0.0948 + 0.1843i 0.0926 + 0.1828i
Chapter 6. Case Study: BPA Test System
155
TABLE 6.11
COMPARISON OF THE LOAD POST-DISTURBANCE AND PRE OLTC OPERATION ESTIMATED
AND ACTUAL POWER COMPONENTS. P1 & Q1: INDUCTION MOTOR LOAD, P2 & Q2:
CONSTANT IMPEDANCE LOAD P3&Q3: CONSTANT POWER LOAD
1P 2P 3P 1Q 2Q 3Q
Case 1
EST5.(PU)
ACT6.(PU)
ERR7 (%)
0.3041 1.7273 0.1000
0.3000 1.7314 0.1000
1.3764 -0.2374 -0.0187
0.0735 0.8622 0.1000
0.0704 0.8654 0.1000
4.5274 -0.3720 0.0341
Case 2
EST.(PU)
ACT.(PU)
ERR (%)
0.3001 1.5701 0.1000
0.3000 1.5702 0.1000
0.0333 -0.0064 -0.0020
0.0795 0.7840 0.1000
0.0786 0.7849 0.1000
1.1450 -0.1147 0.0400
Case 3
EST.(PU)
ACT.(PU)
ERR (%)
0.5006 2.6128 0.1000
0.5000 2.6135 0.1000
0.1235 -0.0241 0.0086
0.2342 0.6678 0.1000
0.2486 0.6534 0.1000
-5.7944 2.2038 0.0079
Having the load power components peeled off, the bus 10 voltage computed by
equation (6.1) at time 110 sec is used to calculate the constant impedance load
parameters and the constant reactive component of the induction motor load
impedance in any case.
Using equation (6.8), the system Thevenin voltage is computed for the tap initial
position and its first and second operations for three cases of simulation, and
then, the results are used to estimate the system Thevenin voltages for the other
tap positions, using the method described in section 6.5.1.
All estimated powers, voltages, and impedances are used to solve the equations
of sections 6.5.1 and 6.5.2 for the bus 10 voltages, and the load impedance from
the bus 9 view point, for different cases and at different tap positions. The
impedance of the compensation equipment in bus 9 is also included in the load
5 EST: Estimated value 6 ACT: Actual value 7 ERR: Estimation error
Chapter 6. Case Study: BPA Test System
156
impedance. The comparisons of the system Thevenin and load estimated
impedances, and the estimated bus 10 voltages to the actual values are shown in
Figures 6.31-6.34.
2 4 6 8 10 12 14 16 18 200.95
1
1.05Load estimated and actual volatges
Vol
tage
(pu)
EstimatedActualReference
2 4 6 8 10 12 14 16 18 200.15
0.2
0.25
0.3
0.35
0.4
Tap number
Impe
danc
e(pu
)
Estimated load and thevenin impedances
LoadSystem
Tap limit
Figure 6.31. Estimated OLTC secondary side voltages, system Thevenin
impedance, and load impedance in case1
Referring to Figure 6.31, the 6 initial estimated load bus voltages are very close
to the actual values. As a general assessment, the estimated voltages up to the tap
limit point are increasing, i.e., voltage is stable. The estimated system Thevenin
Impedances are also above the load impedance for all those points and thus the
same conclusion of voltage being stable is obtained using the impedance
matching criterion. The particular outcome is that the estimated voltage for the
8th tap movement exceeds the voltage reference, and hence, tap movement stops
at this position. The result agrees with the simulation result.
In the second case, the estimated voltage is of reasonable accuracy, however, as
Chapter 6. Case Study: BPA Test System
157
it can be seen from Figure 6.32, there is a consistent overestimation which gives
rise to conservative estimation of stability. The estimated voltages are increasing
and the estimated system Thevenin impedances are above the estimated load
impedances, i.e., voltage instability will not happen. All the estimated voltages
are below the reference voltage, i.e., tap reaches the limit and stops. The results
agree with the simulation results.
2 4 6 8 10 12 14 16 18 200.9
0.95
1
1.05
1.1Load estimated and actual volatges
Vol
tage
(pu)
EstiamtedActualReference
2 4 6 8 10 12 14 16 18 200.15
0.2
0.25
0.3
0.35
0.4
Tap number
Impe
danc
e(pu
)
Estimated load and thevenin impedances
LoadSystem
Tap limit
Figure 6.32 Estimated OLTC secondary side voltages, system Thevenin
impedance, and load impedance in case2
Figure 6.33 shows that in case 3, there is a small difference between the
estimated and actual voltages; however, this creates less difficulty for the voltage
stability assessment in this case, because in such cases, that the system is facing
large disturbances, the post-disturbance voltages are too far from the reference
value and the risk of the estimated voltage to be above the reference value;
implying stable voltage, is too low.
Chapter 6. Case Study: BPA Test System
158
2 4 6 8 10 12 14 16 18 200.92
0.925
0.93
0.935
0.94Load estimated and actual volatges
Vol
tage
(pu)
Tap number
EstimatedActual
Figure 6.33 Comparison of the estimated OLTC secondary side voltages to the
actual values in case 3
120 140 160 180 200 220 240 260 280 300
0.18
0.2
0.22
0.24
0.26
Time (sec)
Impe
danc
e (p
u)
Actual load and thevenin impedances
systemload
2 4 6 8 10 12 14 16 18 200.16
0.18
0.2
0.22
0.24
0.26
Tap number
Impe
danc
e(pu
)
Estimated load and thevenin impedances
loadsystem
Figure 6.34. Estimated system Thevenin impedance and load impedance and
their comparison to the actual values in case 3
Chapter 6. Case Study: BPA Test System
159
In Figure 6.33, the estimated voltage is maximized at the tap 13th movement and
the voltage is still below the reference value and tap limit has not been
reached, i.e., voltage instability is anticipated at the next tap movement. At the
same point of tap position, the estimated load impedance becomes smaller than
the estimated system Thevenin impedance, as it is seen in Figure 6.34.
Considering the above examples, the OLTC secondary side voltage tracing is a
better index, comparing to the impedance matching criterion, because of the
capability of anticipation of the stopping point of the OLTC operation.
6.6 Summary
In this chapter, the developed methods for the voltage instability anticipation
were applied to the BPA standard test system. In the first part, the validity of the
system Thevenin impedance estimation method was investigated. The estimated
results were confirmed by the simulation results. The load peeling method was
also applied to the test system. The estimated results were of reasonable
accuracy.
For a chosen load composite and disturbance size, the different estimation results
were statistically evaluated using the histograms. All results were normally
distributed and the mean values and the standard deviations were also
acceptable.
Finally, these tools were combined to infer voltage stability by applying them to
the test system in three different cases. It was confirmed by simulation that the
proposed methods are capable for the anticipation of a long-term voltage
instability caused by the OLTC operation following a large disturbance in the
system.
Chapter 7
Conclusions Motivated by finding local measurement based on-line methods, the work
presented aimed to estimate the remained time to a possible long term voltage
instability caused by the on-load tap changer operation, following a large
disturbance in the power system.
In order to develop and implement a true on-line voltage stability analysis
method, the on-line accurate modeling of the supply system and the load, based
on the local measurements, is required. The shortcomings of the current
approaches in the system equivalent identification and load modeling were
discussed in chapter 2.
Based on the mentioned objective and the requirements, novel methods were
developed to:
• Estimate on-line the supply system Thevenin impedance using signal
processing on the random changes in the load bus voltage and current
caused by the load.
Chapter 7. Conclusions
161
• Characterize load in an on-line peeling process using changes in the load
bus measured data caused by the disturbance and the random changes in
the supply system.
• Anticipate on-line the expected time to a long-term voltage instability
caused by the OLTC operation.
7.1 Summary of the Results In this section, the main conclusions of the thesis are summarized.
On-line System Thevenin Impedance Estimation
In chapter 3, the idea that the load bus voltage and current changes caused by the
local load can be used to identify the supply system was used to introduce a
method to estimate the power system Thevenin impedance.
Using a one-step ahead prediction method, the dynamic components in the load
voltage and current were removed and the residuals (the random changes) were
then used to estimate the system Thevenin impedance. In order to cancel the
effect of the supply system changes in the load bus voltage and current changes,
they were correlated with the load admittance changes. It was shown that the
ratio of the correlations of the random changes in the load voltage and load
admittance with respect to the correlation of the random changes in the load
current and load admittance can be used to estimate the system Thevenin
impedance.
The proposed method was applied to a four bus test system in different cases and
it was concluded that the estimation result improves with an increase in the
portion of the local load related changes in the bus voltage and current changes
and/or increases in the length of the data window.
Chapter 7. Conclusions
162
The invalidity of the current approach in literature, i.e., the ratio of the changes
in the load voltage with respect to the changes in the load bus current, when the
changes are present in the supply system, was also investigated.
The proposed method was also applied to the real data from Brisbane load
centre. The results were consistent across subsequent time slices; however, the
judgment on the results is still open to the availability of the system actual
Thevenin impedance.
Load Characterization by Sequential Peeling
In chapter 4, considering the requirements of a load model in the on-line voltage
stability study, a method was introduced to characterize the behaviour a
composite of an induction motor, constant impedance and constant power loads,
following a large disturbance in the supply system.
The changes in the load total active power, caused by the disturbance, were used
to identify the active power of the different components of the load. Then, the
electromechanical dynamics and the voltage related random changes in the post-
disturbance load active and reactive power were removed and the residuals, i.e.,
the random changes caused by the load bus frequency changes, together with the
induction motor estimated active power were used to estimate the post-
disturbance induction motor reactive power, and then peel it off from the load
total reactive power. Finally, the reactive powers of the constant impedance load
and the constant power load were estimated using the changes in the load
measured reactive power, caused by the disturbance.
The proposed load peeling method was applied to a four bus test system in two
different cases:
Chapter 7. Conclusions
163
• Different pre-disturbance load compositions with the same disturbance
size
• The same pre-disturbance load composition with different disturbance
sizes
The estimated values showed less than 1.61% error in both cases. However, it
was found from the second experiment that the larger the disturbance, the better
the estimated results. The reason is that, in the proposed method, the estimation
of the five out of the six load power components is based on the load power
changes due to the disturbance. However, they are also affected by the random
changes contribution in the load power changes, and the larger the disturbance
size, the lesser the relative share of the random changes in the load power
changes.
Anticipation of the Expected Time to a Long-term Voltage Instability
In chapter 5, methods were introduced to anticipate voltage instability for a
system consisting of a constant impedance load case, and/or, a composition of
the constant power, constant impedance and induction motor loads behind an
OLTC. The OLTC transformer assumed to be ideal and the changes in the load
bus voltage and current due to the OLTC first operation were used to anticipate
possible voltage instability and the remained time to such instability. It was also
assumed that the system Thevenin impedance and the load parameter/s are
already identified using the proposed methods in chapter3 and chapter 4,
respectively.
Two criteria were used to anticipate the instability: impedance matching between
load and system criterion, and OLTC secondary side voltage maximizing
criterion. The changing of the sign of the changes in the load power due to the
Chapter 7. Conclusions
164
tap operation, at the instability point, was also used as a second index in the
constant impedance load case.
The effect of the load composition was also determined. It was shown that how
the larger induction motor component in the load composite makes the system
voltage safer in the long-term, but increases the risk of the short-term voltage
instability.
To investigate the effect of the measurement uncertainties in the estimation
result, random changes were added to the system and the results were still of
acceptable accuracy. The methods were confirmed by simulation.
Simulation Results on the Standard Test System
To validate the proposed methods in this thesis in voltage stability prediction,
they were applied to the BPA test system. In the simulations, the generators were
presented by their third order models and excitation system and the
overexcitation limiters were also included. The local load and the other load
were modelled as a composite load and constant impedance load, respectively.
A disturbance was applied to the system and the measured load bus voltage and
current from disturbance time until the start of OLTC operation were used to
estimate the system Thevenin impedance and to peel the load power components
off. The estimated results for the different system and load conditions showed
less than 4% and 5% errors for the components of the Thevenin impedance and
the components of the load power, respectively. It was found that the larger the
disturbance, the better the estimated results for the components of the load
power.
The histograms of the different estimations for a chosen load composition were
used to evaluate statistically the estimation methods. The investigation indicated
Chapter 7. Conclusions
165
that all results were normally distributed and the mean values of the estimation
errors were less than 4.1% and 2.08% for the components of the Thevenin
impedance and the components of the load power, respectively. The standard
deviations for the estimated values for the components of the load power were
less that 0.33% of the mean values. The standard deviation of the estimated
imaginary component of the system Thevenin impedance was 3.2% of the mean
value. This coefficient for the estimated real component was around 7.5% which
is acceptable, considering the small value of the real component of the system
Thevenin impedance. However, the statistical characteristics can be improved by
using larger data windows in the estimation process.
Finally, all methods were validated by applying them to the test system in three
different cases. It was confirmed by the simulation results that the proposed
methods are capable of anticipation of a long-term voltage instability caused by
the OLTC operation following a large disturbance in the system. The voltage
tracing criterion was found a better index because of the capability of
anticipation of the stopping point of the OLTC operation.
7.2 Future Research
The proposed load characterization method was confirmed by simulation on the
test systems. The first objective in the continuing work is to apply the method to
a composite load in laboratory. A disturbance can be simulated by inserting
series impedance in the circuit. The random changes in the load voltage and
current caused by the supply system can then be used to investigate the accuracy
of the method and also to quantify the similarities and differences with the
results obtained in this thesis.
Chapter 7. Conclusions
166
A second objective is to apply the peeling method to the real data obtained from
a large disturbance in the system, followed by the OLTC operation. The
accuracy of the estimation result can be investigated by calculating the load bus
voltage at different tap operation points, using the estimated load parameters, and
comparing them to the measured voltages.
The induction motor load in this thesis is considered to be the aggregation of a
group of induction motor loads. Even though there are aggregation methods
applicable to the dissimilar motors, but it is not recommended in general [89].
Hence, a suggestion for future work is to consider different types of motor load
and to try to peel the load off. It is possible that the peeling can be achieved
using frequency domain analysis methods. The theory provided in this thesis
may be extended to the frequency domain to find the load transfer function and
then to break it into single time constant portions, each representing a different
induction motor model.
Estimation of the post-disturbance dynamic status of the system, determination
of the stability margins, the sensitivity of the transfer limits to those margins, and
implementation of a control process to ensure security in the system are the main
parts of an on-line security assessment package. On-line security assessment
packages are commercially available today but their accuracy is limited by the
load model uncertainties [33]. A further suggestion for future work is to improve
the system security margins by developing new control strategies and
determining accurately the system transfer capacity based on the achievements in
this thesis.
My final suggestion for the future work is more focusing on the OLTC role in
the system dynamic behaviour. Sitting between load and system, OLTC
Chapter 7. Conclusions
167
transformer has a dual role. From the primary side it is a part of the load and
from the secondary side it is a part of the supply system. Hence, the changes of
the OLTC primary side and the secondary side voltage and current caused by its
operation may be used to identify the system and the load, respectively.
References
[1] C. W. Taylor, "Power system voltage stability". New York: McGraw-
Hill, 1994.
[2] L. Vargas D., V. H. Quintana, and R. Marinda D., "Voltage collapse
scenario in the Chilean interconnected system," IEEE Transactions on
Power Systems, vol. 14, pp. 1415-1421, 1999.
[3] S. Corsi and C. Sabelli, "General blackout in Italy Sunday September 28,
2003, h. 03:28:00," presented at Power Engineering Society General
Meeting, 2004. IEEE, 2004.
[4] A. AL-Hinai, "Voltage Collapse Prediction for Interconnected Power
Systems," in College of Engineering. Morgantown: West Virginia, 2000.
[5] S. Larsson, "The Black-out in southern Sweden and eastern Denmark,
September 23, 2003," presented at Transmission Network Reliability in
Competitive Electricity Markets (Workshop), IEA, Paris, 29-30 March,
2004.
[6] T. Van Cutsem and C. D. Vournas, "Voltage stability of electric power
systems". Boston,USA: Kluwer Academic Publisher, 1998.
[7] P. Kundur, Power system stability and control. New York: McGraw-Hill,
1994.
[8] P. Kundur, J. Paserba, V. Ajjarapu, G. Andersson, A. Bose, C. Canizares,
N. Hatziargyriou, D. Hill, A. Stankovic, C. Taylor, T. Van Cutsem, and
V. Vittal, "Definition and classification of power system stability
IEEE/CIGRE joint task force on stability terms and definitions," Power
Systems, IEEE Transactions on, vol. 19, pp. 1387-1401, 2004.
[9] A. C. Zambroni de Souza, J. C. Stacchini de Souza, and A. M. Leite da
Silva, "On-Line voltage stability monitoring," IEEE Transactions on
Power Systems, vol. 15, pp. 1300-1305, 2000.
[10] T. Van Cutsem, "Voltage instability: phenomena, countermeasures, and
analysis methods," Proceedings of the IEEE, vol. 88, pp. 208-227, 2000.
[11] S. Z. Zhu, Z. Y. Dong, K. P. Wong, and Z. H. Wang, "Power system
dynamic load identification and stability," presented at International
Conference on Power System Technology, 2000.
References
169
[12] K. Vu, D. Julian, J. Ove gjerde, N. Bhatt, B. Laios, and R. Schulz,
"Voltage instability predictor (VIP) and its applications," presented at
13th PSCC, Trondheim, 1999.
[13] M. Begovic, B. Milosevic, and D. Novosel, "A novel method for voltage
instability protection," presented at System Sciences, 2002. HICSS.
Proceedings of the 35th Annual Hawaii International Conference on,
2002.
[14] k. Vu, M. M. Beguvic, and D. Novosel, "Grids get smart protection and
control," IEEE Computer Applications in Power, pp. 40-44, 1997.
[15] K. Vu, M. M. Beguvic, D. Novosel, and M. Mohan Saha, "Use of local
measurements to estimate voltage-stability margin," IEEE Transactions
on Power Systems, vol. 14, pp. 1029-1035, 1997.
[16] D. E. Julian, R. P. Schulz, K. T. Vu, W. H. Quaintance, N. B. Bhatt, and
D. Novosel, "Quantifying proximity to voltage collapse using the voltage
instability predictor (VIP)," presented at Proceedings of the IEEE-PES
Summer Meeting, Seattle,WA, 2000.
[17] C. W. Taylor, "The future in on-line security assessment and wide-area
stability control," presented at IEEE/PES Winter Meeting, 2000.
[18] I. S. P. 90TH0358-2-PWR, "Voltage Stability of power systems:
Concepts, Analytical tools, and Industry Experience," IEEE 1990.
[19] G. K. Morison, B. Gao, and P. Kundur, "Voltage stability analysis using
static and dynamic approaches," Power Systems, IEEE Transactions on,
vol. 8, pp. 1159-1171, 1993.
[20] D. J. Hill, A. P. Lof, and A. G., "Analysis of long-term voltage stability,"
presented at 10th Power System Computation Conference, Aug. 1990.
[21] C. K. Alexander and M. N. O. Sadiku, "Fundamentals of Electric
Circuits". Singapour: McGraw-Hill, 2003.
[22] C. A. Canizares, A. C. Z. DE Souza, and V. H. Quintana, "Comparison of
performance indices for detection of proximity to voltage collapse," IEEE
Transaction on Power Systems, vol. 11, pp. 1441-1450, 1996.
[23] M. M. Beguvic, D. Fulton, M. R. Gonzalez, J. Goossens, E. A. Guro, R.
W. Haas, C. F. Henville, G. Manchur, G. L. Michel, R. C. Pastore, J.
Postforoosh, G. L. Schmitt, J. B. Williams, and K. Zimmerman,
References
170
"Summary of system protection and voltage stability," IEEE Transaction
on Power Delivery, vol. 10, pp. 631-638, 1995.
[24] A. E. Hammad and M. Z. El-Zadek, "Prevention of transient voltage
instabilities due to induction motor loads by static var compensators,"
IEEE Transaction on Power Systems, vol. 4, pp. 1182-1190, 1989.
[25] Y. Sekine and H. Ohtsuki, "Cascaded voltage collapse," IEEE
Transaction on Power Systems, vol. 5, pp. 250-256, 1990.
[26] C. D. Vournas and G. A. Manos, "Modelling of stalling motors during
voltage stability studies," IEEE Transaction on Power Systems, vol. 13,
pp. 775-781, 1998.
[27] I. T. Force, "Load representation for dynamic performance simulation,"
IEEE Transaction on Power Systems, vol. 8, pp. 472-482, 1993.
[28] Q. Liu, Y. Chen, and D. Duan, "The load modeling and parameters
identification for voltage stability analysis," presented at Power System
Technology, 2002. Proceedings. PowerCon 2002. International
Conference on, 2002.
[29] M. S. Calovic, "Modelling and analysis of under-load tap-changing
transformer control system," IEEE Transaction on Power Apparatus and
Systems, vol. PAS-103, pp. 1909-15, 1984.
[30] P. W. Sauer and M. A. Pai, "A comparison of discrete vs. continuous
dynamic models of tap-changing-under-load-transformers.," presented at
Bulk Power System Voltage Phenomena-III: Voltage Stability,Security
and Control, Davos, Switzerland, 1994.
[31] M. Larsson, " Coordinated voltage control in power systems." Lund:
Lund University, 2000.
[32] M. Larsson, "The ABB Power Transmission Test Case," ABB Schweiz
AG, Tech. report, 2002.
[33] IEEE/PES, "Voltage Stability Assessment: Concepts, Procedures and
Tools": IEEE Catalog Number SP101PSS, August, 2002.
[34] H. Ohtsuki, A. Yokoyama, and Y. Sekine, "Reverse action of on-load tap
changer in association with voltage collapse," IEEE Transaction on
Power Systems, vol. 6, pp. 300-306, 1991.
References
171
[35] Y.-Y. Hong and H.-Y. Wang, "Investigation of the voltage stability
region involving on-load tap changers," Electric Power Systems
Research, vol. 32, pp. 45-54, 1995.
[36] C.-C. Liu and K. T. Vu, "Analysis of tap-changer dynamics and
construction of Voltage Stability regions," IEEE Transaction on Circuits
and Systems, vol. 36, pp. 575-590, 1989.
[37] N. Yorino, A. Funahashi, and H. Sasaki, "On reverse control action of
on-load tap-changers," Electric Power & Energy Systems, vol. 19, pp.
541-548, 1997.
[38] K. T. Vu and C.-C. Liu, "Shrinking stability regions and voltage collapse
in power systems," Circuits and Systems I: Fundamental Theory and
Applications, IEEE Transactions on [see also Circuits and Systems I:
Regular Papers, IEEE Transactions on], vol. 39, pp. 271-289, 1992.
[39] L. Bao, X. Duan, and Y. He, "Dynamical analysis of voltage stability for
a simple power system," Electric Power & Energy Systems, vol. 23, pp.
557-564, 2001.
[40] C. C. Liu, "An extended study of dynamic voltage collapse mechanisms,"
presented at Proceedings of 28th Conference on Decision and Control,
Tampa, Florida, 1989.
[41] J. A. Momoh, Y. Zhang, and G. J. Young, "Effects of under-load tap
changer (LTC) on power system voltage stability," presented at Power
Symposium, 1990. Proceedings of the Twenty-Second Annual North
American, 1990.
[42] V. Venkatasubramanian, H. Schattler, and J. Zaborszky, "Analysis of the
tap changer related voltage collapse phenomena for the large electric
power system," presented at Decision and Control, 1992., Proceedings of
the 31st IEEE Conference on, 1992.
[43] C. D. Vournas, "On the role of LTCs in emergency and preventive
voltage stability control," presented at IEEE/PES Power System Stability
Controls Subcommittee Meeting, New York, USA, 2002.
[44] H. K. Clark, "Voltage control and reactive supply problems," Ieee
Tutorial on Reactive Power: Basics,Problems and solutions, pp. 17-27,
1987.
References
172
[45] D. Karlsson, "Voltage stability simulation using detalied models based on
field measurements," Chalmers University of Technology, 1992.
[46] V. Ajjarapu and B. Lee, "Bibliography on voltage stability," IEEE
Transactions on Power Systems, vol. 13, pp. 115-125, 1998.
[47] G. M. Huang and l. Zhao, "Measurement based voltage stability
monotoring of power system," presented at 34th Annual Frontiers of
Power Conference, Oklahoma State University,USA, 2001.
[48] M. H. Haque, "On-line monitoring of maximum permissible loading of a
power system within voltage stability limits," Generation, Transmission
and Distribution, IEE Proceedings-, vol. 150, pp. 107-112, 2003.
[49] t. S. Sidhu, V. Balamourougan, and M. S. Sachdev, "On-line Prediction
of Voltage Collapse," presented at The sixth International power
Engineering conference(IPEC2003), Singapore, 27-29 November 2003.
[50] N. Yorino, H. Sasaki, Y. Masuda, Y. Tamura, M. Kitagawa, and A.
Oshimo, "An investigation of voltage instability problems," Power
Systems, IEEE Transactions on, vol. 7, pp. 600-611, 1992.
[51] N. Flatabo, R. Ognedal, and T. Carlsen, "Voltage stability condition in a
power transmission system calculated by sensitivity methods," Power
Systems, IEEE Transactions on, vol. 5, pp. 1286-1293, 1990.
[52] O. B. Fosso, N. Flatabo, B. Hahavik, and A. T. Holen, "Comparison of
Methods for Calculation of Margins to Voltage Instability," presented at
Athens Power Tech, 1993. APT 93. Proceedings. Joint International
Power Conference, 1993.
[53] B. Gao, G. K. Morison, and P. Kundur, "Voltage stability evaluation
using modal analysis," Power Systems, IEEE Transactions on, vol. 7, pp.
1529-1542, 1992.
[54] H. Dai, C. Wang, Y. Yu, Y. Ni, and F. F. Wu, "Decetralized power
system voltage stability proximity indicator based on local bus
information," presented at 5th International Conference on Advances in
Power System Control, Operation and Management, Hong Kong, 2000.
[55] L. Warland and A. T. Holen, "A voltage instability predictor using local
measurements(VIP++)," presented at Power Tech Conference, Porto,
Portugal, 2001.
References
173
[56] S. A. Soliman, H. K. Temraz, and S. M. El-Khodary, "Power system
voltage stability margin identification using local measurements,"
presented at Power Engineering, 2003 Large Engineering Systems
Conference on, 2003.
[57] B. Palethorpe, M. Sumner, and D. W. P. Thomas, "System impedance
measurement for use with active filter control," presented at Power
Electronics and Variable Speed Drives, 2000. Eighth International
Conference on (IEE Conf. Publ. No. 475), 2000.
[58] K. Morison, H. Hamadani, and L. Wang, "Practical issues in load
modeling for voltage stability studies," presented at Power Engineering
Society General Meeting, 2003, IEEE, 2003.
[59] L. M. Hajagos and B. Danai, "Laboratory measurements and models of
modern loads and their effect on voltage stability studies," Power
Systems, IEEE Transactions on, vol. 13, pp. 584-592, 1998.
[60] D. J. Hill, "Nonlinear dynamic load models with recovery for voltage
stability studies," IEEE Transaction on Power Systems, vol. 8, pp. 166-
176, 1993.
[61] I. R. Navarro, "Dynamic Load Models for Power Systems," in
Department of Industrial Electrical Engineering. Lund: Lund, 2002, pp.
157.
[62] I. R. Navarro, O. Samuelsson, and S. Lindahl, "Automatic determination
of parameters in dynamic load models from normal operation data,"
presented at Power Engineering Society General Meeting, 2003, IEEE,
2003.
[63] I. R. Navarro, O. Samuelsson, and S. Lindahl, "Influence of
normalization in dynamic reactive load models," Power Systems, IEEE
Transactions on, vol. 18, pp. 972-973, 2003.
[64] IEEE Task Force on Load Representation for Dynamic Performance,
"Standard load models for power flow and dynamic performance
simulation," IEEE Transactions on Power Systems, vol. 10, pp. 1302-
1313, 1995.
[65] "Bibliography on load models for power flow and dynamic performance
simulation," Power Systems, IEEE Transactions on, vol. 10, pp. 523-538,
1995.
References
174
[66] B. W. Kennedy and R. H. Fletcher, "Conservation voltage reduction
(CVR) at Snohomish County PUD," Power Systems, IEEE Transactions
on, vol. 6, pp. 986-998, 1991.
[67] d. Karlsson and D. J. Hill, "Modeling and identification of nonlinear
dynamic loads in power systems," IEEE Transaction on Power Systems,
vol. 9, pp. 157-166, 1994.
[68] W. Xu and Y. Mansour, "Voltage stability analysis using generic
dynamic load models," IEEE Transaction on Power Systems, vol. 9, pp.
479-493, 1994.
[69] W. Xu, E. Vaheddi, Y. Mansour, and J. Tamby, "Voltage stability load
parameter determination from field tests on B.C. Hydro's system," IEEE
Transaction on Power Systems, vol. 12, pp. 1290-1297, 1997.
[70] A. Borghetti, R. Caldon, A. Mari, and C. A. Nucci, "On dynamic load
models for voltage stability studies," IEEE Transaction on Power
Systems, vol. 12, pp. 293-303, 1997.
[71] A. Borghetti, R. Caldon, and C. A. Nucci, "Generic dynamic load models
in long-term voltage stability studies," Electric Power System Research,
vol. 22, pp. 291-301, 2000.
[72] Y. Wang and N. C. Pahalawaththa, "Power system load modelling,"
presented at International Conference on Power system
Technology,Power CON'98, 1998.
[73] M. A. Pai, P. W. Sauer, and B. C. Lesieutre, "Static and dynamic
nonlinear loads and structural stability in power systems," Proceedings of
IEEE, vol. 83, pp. 1562-1572, 1995.
[74] F. T. Dai, J. V. Milanovic, N. Jenkins, and V. Roberts, "The influence of
voltage variations on estimated load parameters," presented at Electricity
Distribution, 2001. Part 1: Contributions. CIRED. 16th International
Conference and Exhibition on (IEE Conf. Publ No. 482), 2001.
[75] M. M. Begovic and R. Q. Mills, "Load identification and voltage stability
monitoring," Power Systems, IEEE Transactions on, vol. 10, pp. 109-
116, 1995.
[76] G. Ledwich and E. Palmer, "Modal estimates from normal operation of
power systems," presented at IEEE/PES Winter Meeting, 2000.
References
175
[77] Z. Peyton and J. Peebles, " Probability, random variables, and random
signal principles". New York: McGraw-Hill, 2000.
[78] M. Bahadornejad and G. Ledwich, "System Thevenin impedance
estimation using signal processing on load bus data," presented at IEE
Hong Kong International Conference on Advances in Power System
Control, Operation and Management, APSCOM 2003, Hong Kong, 11-
14 November 2003.
[79] M. Bahadornejad and G. Ledwich, "On-line Load Characterization by
Sequential Peeling," presented at 2004 International Conference on
Power System Technology, Powercon2004, Singapore, 21-24 Nov 2004.
[80] K. R. Godfrey, "Correlation methods," Automatica, vol. 16, pp. 527–534,
1980.
[81] M. Bahadornejad and G. Ledwich, "Studies in the OLTC Effects on
Voltage Collapse Using Local Load Bus Data," presented at Australasian
Universities Power Engineering Conference, AUPEC'2003, Christchurch,
New Zealand, 28 Sept-1 Nov, 2003.
[82] M. Bahadornejad and G. Ledwich, "On-line voltage Collapse Prediction
Considering Composite Load and On Load Tap Changer," presented at
Australasian Universities Power Engineering Conference, AUPEC'2004,
Brisbane, Australia, 26-29 September 2004.
[83] I. A. Hiskens, "Analysis tools for power systems- contending with
nonlinearities," Proceedings of IEEE, vol. 83, pp. 1573-1587, 1995.
[84] J. E. O. Pessanha, V. Leonardo Paucar, and M. J. Rider, "A review of
power system voltage and angular stability dynamics," presented at
Power System Technology, 2002. Proceedings. PowerCon 2002.
International Conference on, 2002.
[85] C. W. Taylor, "Concepts of undervoltage load shedding for voltage
stability," IEEE Transaction on Power Delivery, vol. 7, pp. 480-488,
1992.
[86] V. H. Quintana and L. Vargas, "Voltage stability as affected by discrete
changes in the topology of power networks," Generation, Transmission
and Distribution, IEE Proceedings-, vol. 141, pp. 346-352, 1994.
[87] C. T. Force, "Tools for Simulating Long Term Dynamics," Cigre,
Technical report 1995.
References
176
[88] P. M. Anderson and A. A. Fouad, "Power System Control and Stability".
New York: IEEE, 1994.
[89] D. C. Franklin and A. Morelato, "Improving dynamic aggregation of
induction motor models," IEEE Transaction on Power Systems, vol. 9,
pp. 1934-1941, 1994.
[90] J. P. Norton, "An introduction to identification" New York: Academic
Press, 1986
Appendix A
One-step-ahead Prediction for Removing Dynamic Component The one-step-ahead prediction method is calculating the response of the system
one step in the future to an input sequence while the process outputs are known
up to some time instant k-1 [90]. Consider the process
)k()k()mk(ub)k(ub)nk(ya)k(ya)k(y Tmn θϕ=−+⋅⋅⋅+−+−−⋅⋅⋅−−−= 11 11
(A-1)
where u and y are the system input and output, respectively and
pT R,)mk(u)k(u)nk(y)k(y)k( ∈ϕ−⋅⋅⋅−−⋅⋅⋅−−=ϕ ]11[ M , ;mnp +=
and T11 ][ mn bbaa ⋅⋅⋅⋅⋅⋅=θ M , pR∈θ .
To identify y(k) the unknown parameter θ should be identified which can be
estimated by least squares (LS) technique. For N available data samples the
model can be written in vector/matrix form with N-m equations for k = m + 1; …
; N. Assuming n=m, it can be shown that the estimation error is minimized if:
YXXX TT 1-)(=θ∧
(A-2)
where T)]( )2( )1([ NymymyY ⋅⋅⋅++= and the regression matrix X is :
( ) ( ) ( ) ( )( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢
⎣
⎡
−⋅⋅⋅−−−⋅⋅⋅−−
⋅⋅⋅+−⋅⋅⋅+−⋅⋅⋅−⋅⋅⋅−
=
mNuNumNyNy
umuymyumuymy
X
11
212111
MMMM (A-3)
This is an autoregressive moving average (ARMA) process that is used in this
thesis to estimate the noise free dynamic component of the signal using the
original signal as output and an arbitrary signal as input. The random component
of the signal is extracted by subtracting the estimated signal from the original
one.
178
Appendix B
Frequency Relation between Buses In the three bus system in Figure B.1 we can write:
0331232131 =++ VYVYVY (B-1) Thus:
( )23213133
31 VYVY
YV +−= (B-2)
1V = 1V ∠ 1δ 3V = 3V ∠ 3δ 2V = 2V ∠ 2δ
1Z 2Z I LZ
Figure B.1 Three bus system
Putting 222δ= jeVV and 333
δ= jeVV in Equation 2 we will have:
=33
δjeV ( )223213133
1 δ+− jeVYVYY
(B-3)
Taking the derivatives of both sides in Equation (B-3) with respect to time and considering that bus 1 is an infinite bus, supposing 2V is constant, and
neglecting the changes in 3V ( 03 =dtVd
); we will have:
=ω δ333jeV ( )22322
33
1 δω− jeVYY
(B-4)
Where, dtdδω = . Now it can be shown simply that:
23
232
333
1ω⎟⎟⎠
⎞⎜⎜⎝
⎛−=ω
VV
YY
(B-5)
Considering L
3
ZV
I = and using the current division law, we can write:
179
⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛+
+=LZ
VZZ
ZZVV 3
21
1232 (B-6)
Replacing 2V in Equation (B-5) from Equation (B-6) and considering that 2
321
ZY −=
andLZZZ
Y 111
2133 ++= ; we will have:
( ) 221
21
21
23 1
111
1
ωω ⎟⎟⎠
⎞⎜⎜⎝
⎛+
+++
=L
L
ZZZZZ
ZZZ
Z (B-7)
After some simplification it can be shown that:
221
13 ωω ⎟⎟
⎠
⎞⎜⎜⎝
⎛+
=ZZ
Z (B-8)
Equation (B-8) is valid just when both impedances; 1Z and, 2Z have the same
arguments. For the transmission lines normally the resistive component of
impedance is ignored in comparison to the reactive component. Putting 11 jXZ =
and 22 jXZ = in Equation (B-8) we will have:
221
13 ω⎟⎟
⎠
⎞⎜⎜⎝
⎛+
=ωXX
X (B-9)
If there is also any changes in 1δ (i.e. 011 ≠
δ=ω
dtd
), then in can be easily shown
that:
21
21123 XX
ωXωXω
+
+= (B-10)
180
Appendix C
BPA Test System Data
The BPA test system data are as below [19]: Transmission lines (R, X & B in pu on 100 MVA Base) R X B 4-5 0.0000 0.0040 0.0000 5-6 0.0015 0.0288 1.1730 8-9 0.0010 0.0030 0.0000 Transformers (R & X in pu on 100 MVA) R X 1-4 0.0000 0.0020 2-5 0.0000 0.0045 3-6 0.0000 0.0125 6-7 0.0000 0.0030 6-8 0.0000 0.0026 3-6 0.0000 0.0010 Machine Parameters Machine 1: Infinite Bus Machine 2: H=2.09, MVA Base = 2200 MVA Machine 3: H=2.33, MVA Base = 1400 MVA Stator and rotor parameters (Machine 2 &3) Ra=0.0046 Xd=Xq=2.07 X’
d=0.28 T’do=4.10 Exciters Both machine 2 and 3 have ALTHYREX exciters and the data are based on Unit F18 in Appendix D of Anderson and Fouad [88] KA=200 TA=0.3575 KE=0.0 TE=0.0 Field current limiters Data are provided from Kundur [7] Ifdmax1=3.02 pu (or 1.05 x full load current) Ifdmax2=4.60 pu (or 1.6x full load current) K1=150 K2=0.248 K3=12.6 K4=140 ILIM=3.85 Induction motor load data Rs=0.0 Xs =0.092 Xm =2.14 Rr =0.059 Xr =0.075 H=0.342 MVA Base=2440