Islanding detection and classification and load shedding scheme for dispersed generation integrated radial distribution systems.
ISLANDING DETECTION AND CLASSIFICATION AND LOAD SHEDDING
SCHEME FOR DISPERSED GENERATION INTEGRATED
RADIAL DISTRIBUTION SYSTEMS
AZIAH KHAMIS
THESIS SUBMITTED IN FULFILMENT OF THE DEGREE OF
DOCTOR OF PHILOSOPHY
FACULTY OF ENGINEERING AND BUILT ENVIRONMENT
UNIVERSITI KEBANGSAAN MALAYSIA
BANGI
2014
PENGESANAN DAN PENGKELASAN KEPULAUAN DAN SKIM
PENYISIHAN BEBAN BAGI SISTEM PENGAGIHAN
JEJARI TERSEPADU PENJANA TERAGIH
AZIAH KHAMIS
TESIS YANG DIKEMUKAKAN UNTUK MEMPEROLEH IJAZAH
DOKTOR FALSAFAH
FAKULTI KEJURUTERAAN DAN ALAM BINA
UNIVERSITI KEBANGSAAN MALAYSIA
BANGI
2014
iii
DECLARATION
I hereby declare that the work in this thesis is my own except for quotations and
summaries which have been duly acknowledged.
3 Nov 2014
AZIAH KHAMIS
P59755
iv
ACKNOWLEDGMENTS
First and foremost praise be to Almighty Allah for all His blessings for giving me
perseverance and good health throughout the duration of this PhD research.
I would like to express sincere appreciation for the intelligent advice,
encouragement and guidance of my main supervisor, Assoc. Prof. Dr. Hussain
Shareef. Without his tireless assistance, leadership, and confidence in my abilities, this
thesis would not come to its timely completion.
Furthermore, I would like to express my high appreciation to my co-supervisor
Prof. Dr. Hjh. Azah Mohamed for the valued knowledge, ideas, encouragement,
assistance and support received from her during my PhD program.
I would like to acknowledge the financial support from Ministry of Higher
Education, Universiti Teknikal Malaysia Melaka and also Universiti Kebangsaan
Malaysia for making it possible for me to pursue and complete my PhD degree.
I would like to thank all Power System Research group UKM for their help,
friendship, and creating a pleasant working environment throughout my years in
UKM. My heartfelt appreciation goes to all my special friends, especially Hazilah
Abdullah and Abdul Zaini Abdullah for their inseparable support and prayers.
To my dearest husband, Ahmad Zulkarnain Rosli, thanks for your do’as,
patience, understanding and support for all the duration of doing this research. Last
but not least, my special thanks to my family who has been my inspiration since the
primary school until this stage of my education. I dedicate this thesis to my beloved
late mother, Ramlah Binti Karim, who died before I continue my PhD study.
v
ABSTRACT
The high penetration level of distributed generation (DG) provides numerous potential
environmental benefits, such as high reliability, efficiency, and low carbon emissions.
However, the effective detection of islanding and rapid DG disconnection is essential
to prevent safety problems and equipment damage caused by the island mode
operations of DGs. The common islanding protection technology is based on passive
techniques that do not perturb the system but have large nondetection zones.
Therefore, the first part of this thesis attempts to develop a simple and effective
passive islanding detection method with reference to a probabilistic neural networkbased classifier, as well as utilizes the features extracted from three-phase voltages
seen at the DG terminal. This approach enables initial features to be obtained using
the phase-space technique. This technique analyzes the time series in a higher
dimensional space, revealing several hidden features of the original signal.
Meanwhile, the second part of the thesis focuses on the development of an optimal
load shedding scheme after the system experiences an unintentional islanding state to
prevent system collapse due to load-generation mismatch and voltage instability
encountered in the islanded part of the system. To handle this optimization problem, a
constraint multiobjective function that considers the linear static voltage stability
margin and amount of load curtailment was formulated. A novel heuristic
optimization technique based on the backtracking search algorithm (BSA) was
subsequently proposed as an optimization tool for determining the optimum load
shedding based on the proposed objective function. Several test systems, including a
radial distribution system with two DG units and the Institute of Electrical and
Electronics Engineers (IEEE) 33-bus radial distribution system with four DG units,
were utilized to evaluate the effectiveness of the proposed islanding detection method
and the optimal load shedding scheme. The effectiveness of the proposed islanding
detection method was verified by comparing its results with the conventional wavelet
transform (WT)-based technique through intensive simulations conducted with the
DIgSILENT Power Factory® software. The assessment indices, namely, the mean
absolute percentage error (MAPE), mean absolute error (MAE), and root mean square
error (RMSE), obtained a 0% error rate for the proposed method when applied to the
IEEE 33-bus radial distribution system with four DG units. Meanwhile, the MAPE,
MAE, and RMSE obtained error rates of 0.1482%, 0.2752%, and 5.055%, respectively,
for the WT-based technique when applied to the IEEE 33-bus radial distribution
system with four DG units. These indices show that the proposed islanding detection
method is robust and capable of sensing the difference between the islanding
condition and other system disturbances. Meanwhile, the performance of the proposed
load shedding scheme was evaluated through an extensive test conducted on the IEEE
33-bus system considering several scenarios such as load shedding under various
operating points and at various islands using the MATLAB® software. Moreover, the
effectiveness of the proposed scheme was validated by comparing its results with
those obtained using the genetic algorithm (GA). The optimization results indicate that
the proposed BSA technique is more effective in determining the optimal amount of
load to be shed in any islanded system compared with GA.
vi
ABSTRAK
Tahap peningkatan penembusan penjana teragih (PT) yang tinggi memberikan
pelbagai faedah alam sekitar seperti keboleharapan yang tinggi, kecekapan dan
pengeluaran karbon yang rendah. Walau bagaimanapun, pengesanan kepulauan yang
berkesan adalah penting bagi mengelakkan masalah keselamatan dan kerosakan
peralatan disebabkan oleh operasi kepulauan PT. Teknologi perlindungan kepulauan
yang biasa digunakan adalah berasaskan teknik pasif. Oleh itu, bahagian pertama tesis
ini bertujuan untuk membangunkan satu kaedah pengesanan kepulauan pasif yang
mudah dan berkesan dengan menggunakan rangkaian neural kebarangkalian untuk
pengelasan, dan juga mengunakan ciri yang diekstrak dari voltan tiga fasa yang dilihat
pada pengkalan PT. Pendekatan ini membolehkan perolehan ciri awal dengan
menggunakan teknik fasa-ruang. Teknik ini menganalisa siri masa dalam ruang
dimensi yang lebih tinggi, dan mendedahkan beberapa ciri tersembunyi dari isyarat
asal. Sementara itu, bahagian kedua tesis ini memberi tumpuan kepada pembangunan
skim penyisihan beban optimum selepas sistem mengalami kepulauan yang tidak
disengajakan untuk mengelakkan ketakstabilan voltan di bahagian sistem kepulauan.
Untuk menangani masalah pengoptimuman, fungsi objektif berbilang
dipertimbangkan dengan mengambilkira kestabilan jidar voltan statik dan jumlah
pengurangan voltan. Satu teknik pengoptimuman heuristik yang novel, iaitu, algoritma
carian jejak balik (ACJB) telah dicadangkan untuk menentukan penyisihan beban
yang optimum. Beberapa sistem ujian, termasuk sistem jejari dengan dua unit PT dan
juga sistem agihan jejari IEEE 33 bas dengan empat unit PT, masing-masing, telah
digunakan untuk menilai keberkesanan kaedah pengesanan kepulauan dan penyisihan
beban optimum yang dicadangkan. Keberkesanan kaedah pengesanan kepulauan yang
dicadangkan telah disahkan melalui perbandingan dengan teknik anak gelombang
melalui simulasi intensif menggunakan perisian DIgSILENT Power Factory®. Indeks
penilaian, seperti peratusan min ralat mutlak (PMRM), min ralat multak (MRM), dan
ralat punca min kuasa dua (RPMKD ) didapati kadar ralat adalah 0% bagi kaedah
pengesanan kepulauan yang dicadangkan apabila diuji pada sistem agihan jejari IEEE
33 bas dengan empat unit PT. Manakala, PMRM, MRM dan RPMKD memperoleh
kadar ralat sebanyak 0.1482%, 0.2752%, and 5.055%, masing-masing, bagi teknik
berasaskan anak gelombang apabila diuji pada sistem agihan jejari IEEE 33 bas
dengan empat unit PT. Hasil keputusan indeks menunjukkan kaedah pengesanan
kepulauan yang dicadangkan adalah teguh dan mampu mengesan perbezaan antara
kepulauan serta gangguan sistem yang lain. Sementara itu, prestasi skim penyisihan
beban yang dicadangkan telah dinilai dengan ujian yang meluas dijalankan pada
sistem bas IEEE 33 dengan mempertimbangkan beberapa scenario seperti penyisihan
beban di bawah pelbagai titik operasi dan pulau dalam perisian MATLAB®.
Tambahan pula, keberkesanan skim penyisihan yang dicadangkan telah disahkan
dengan membandingkan keputusan yang telah diperolehi dengan keputusan algoritma
genetic (AG). Hasil kajian pengoptimuman menunjukkan bahawa teknik ACJB yang
dicadangkan adalah lebih berkesan dalam menentukan nilai optimum bagi beban yang
akan dikurangkan pada mana-mana sistem kepulauan berbanding AG.
vii
TABLE OF CONTENTS
Page
DECLARATION
iii
ACKNOWLEDGMENTS
iv
ABSTRACT
v
ABSTRAK
vi
CONTENTS
vii
LIST OF TABLES
x
LIST OF FIGURES
xii
LIST OF ABBREVIATIONS
xv
LIST OF SYMBOLS
xvii
CHAPTER I
INTRODUCTION
1.1
Research Background
1
1.2
Problem Statement
4
1.3
Objective of the Research
6
1.4
Scope of the Research
6
1.5
Organization of the Thesis
7
CHAPTER II
LITERATURE REVIEW
2.1
Islanding Detection Methods
8
2.1.1 Central Islanding Detection Techniques
9
2.1.2 Review of the Conventional Local Islanding Detection 12
Technique
2.1.3 Review of the Intelligent Local Islanding Detection
16
Technique
2.2
DG Models for Islanding Detection
23
2.3
Load Shedding Schemes
25
2.3.1 Review of Under-Frequency Load Shedding Schemes 26
2.3.2 Review of Under-Voltage Load Shedding Schemes
29
2.4
DG Models for Optimal Load Shedding
33
2.5
Types of Distribution System Design
35
viii
2.6
Chapter Summary
CHAPTER III
ISLANDING DETECTION USING PHASE-SPACE
AND NEURAL NETWORK
3.1
Introduction
38
3.2
Tools and Methods used in the Proposed Method
39
3.2.1 Phase-Space Technique
3.2.2 Probabilistic Neural Network
39
41
Proposed Islanding Detection Method
44
3.3.1 Data Collection
3.3.2 Phase-Space Feature Extraction
3.3.3 Design of Artificial Intelligent Classifier
44
45
47
Performance Evaluation Methods
52
3.4.1 Performance Evaluation of Conventional Method
3.4.2 Performance Evaluation with Statistical Indices
52
54
3.5
Chapter Summary
55
CHAPTER IV
OPTIMAL LOAD SHEDDING SCHEME USING
BACKTRACKING SEARCH ALGORITHM
4.1
Introduction
4.2
Tools and Methods Used in Proposed Load Shedding Scheme 56
3.3
3.4
36
56
4.2.1 Voltage Stability Margin
4.2.2 Backtracking Search Optimization Algorithm
4.2.3 Newton–Raphson Power Flow Solution
56
58
61
Problem Formulation for Optimal Load Shedding Scheme
62
4.3.1 Operational Constraints
4.3.2 Fitness Function
4.3.3 Application of BSA for Optimal Load Shedding
Scheme
63
64
65
Performance Evaluation Scheme
69
4.4.1 Performance Evaluation with Conventional GA
Method
69
4.5
Chapter Summary
71
CHAPTER V
RESULTS AND DISCUSSION
5.1
The Test System for Islanding Detection
5.1.1 Radial Distribution System with Two Identical DG
4.3
4.4
72
ix
5.2
5.3
Units
5.1.2 IEEE 33-Bus Radial Distribution System with Four
DG Units
72
73
Test Results of the Radial Distribution System with Two
Identical DG Units
74
5.2.1
5.2.2
5.2.3
5.2.4
5.2.5
5.2.6
76
78
79
80
81
81
Input Feature Extraction
Results of RBFNN with Phase-Space Features
Results of RBFNN with Wavelet Transform Features
Results of PNN with Phase-Space Features
Results of PNN with Wavelet Transform Features
Summary of the Result Obtained for All of the Tested
Islanding Detection Methods
Implementation of Islanding Detection on The IEEE 33-bus
Radial Distribution Systems with Four DG Units
84
5.3.1
5.3.2
5.3.3
5.3.4
5.3.5
5.3.6
85
88
89
90
91
91
Input Features Extraction
Results of RBFNN with Phase-Space Features
Results of RBFNN with Wavelet Transform Features
Results of PNN with Phase-Space Features
Results of PNN with Wavelet Transform Features
Summary of Islanding Detection Analysis
5.4
Description of the Test System for Optimal Load
Shedding Scheme
93
5.5
Test Results of Optimal Load Shedding Scheme
96
5.5.1
5.5.2
5.5.3
5.5.4
Optimal Load Shedding for Island A Using BSA
Optimal Load Shedding for Island A Using GA
Optimal Load Shedding for Other Islanded Systems
Summary of Load Shedding Scheme
99
106
113
124
5.6
Chapter Summary
CHAPTER VI
CONCLUSIONS AND RECOMMENDATIONS
6.1
Overall Conclusions
126
6.2
Significant Contributions of the Research
128
6.3
Recommendations for Future Studies
129
REFERENCES
125
130
APPENDIXES
A
IEEE 33-Bus Radial Distribution System
140
B
Wavelet Technique as Features Extraction
141
C
List of Publication
143
x
LIST OF TABLES
Table Number
Page
2.1
Summary of remote islanding detection techniques
12
2.2
Comparison of passive islanding detection techniques
13
2.3
Comparison of active islanding detection techniques
15
2.4
Utilization of SP in islanding detection
20
2.5
Utilization of the AI classifier in islanding detection at DG
23
2.6
Comparison of various computational intelligence load
shedding schemes
33
3.1
Classifier output definition
45
3.2
Selected phase-space features
45
3.3
A scale of judgment of forecasting accuracy
55
5.1
System model description
72
5.2
DG installed node with operating points
74
5.3
Number of samples for training and testing
76
5.4
Parameter settings of the RBFNN and PNN classifiers for
DG1 and DG2
77
5.5
RBFNN classification results with phase-space features
79
5.6
RBFNN classification results with wavelet transform
features
80
5.7
PNN classification results with phase-space features
81
5.8
PNN classification results with wavelet transform features
81
5.9
Comparison of PNN classifier performance with phase-space
and wavelet features
82
5.10
Comparison of MAPE, RMSE, and MAE for various
islanding detection methods
83
5.11
Number of samples for training and testing at IEEE 33 bus
system
85
5.12
Parameter settings of the RBFNN and PNN classifiers for
DG1, DG2, DG3, and DG4
88
5.13
RBFNN classification results with phase-space features
88
5.14
RBFNN classification results with wavelet transform
features
90
5.15
PNN classification results with phase-space features
91
xi
5.16
PNN classification results with wavelet transform features
91
5.17
Comparison of PNN classifier performance with phase-space
and wavelet feature
92
5.18
Comparison of MAPE, RMSE, and MAE for various
islanding detection methods
92
5.19
Rated maximum power of DGs
93
5.20
Percentage load priority limits for the IEEE 33-bus radial
distribution system
94
5.21
Overall power demand and supply in islanded system
96
5.22
GA and BSA parameter settings
96
5.23
Amount of hourly load curtailment at individual buses in
island A
103
5.24
Summary of load shedding performances at hour 9.00
113
5.25
Summary of load shedding performances at hour 15.00
117
5.26
Amount of hourly load curtailment at individual buses in
island B
121
5.27
Amount of hourly load curtailment at individual buses in
island C
122
5.28
Amount of hourly load curtailment at individual buses in
island D
123
5.29
Performance of BSA and GA in terms of fitness, VSM, and
amount of load curtailment at hour 9.00
124
5.30
Performance of BSA and GA in terms of fitness, VSM, and
amount of load curtailment at hour 15.00
124
xii
LIST OF FIGURES
Figure Number
Page
1.1
(a) Traditional distribution
embedded distribution system
system,
(b)
generation
1.2
Power islanding condition
2
1.3
Voltage and frequency response
3
2.1
Classification of islanding detection technique
9
2.2
Remote islanding detection technique
9
2.3
Transfer trip scheme
11
2.4
Basic block of intelligent islanding detection technique and
classification
16
2.5
Block diagram of the inverter
24
2.6
Schematic diagram of mini hydro power for grid connected
operation
24
2.7
Types of load shedding schemes
25
2.8
Generator model
34
2.9
System for load shedding
35
2.10
Simplified illustration of the concept behind three types of
power distribution configuration
36
3.1
Architecture of a PNN
43
3.2
Phase-space feature extraction, (a) three-phase fault, (b)
Euclidean norm (Ex) of the fault signal in (a), (c) selected
region of Euclidean norm (Ex) for feature selections, and
(d) Euclidean norm (Ex) and its features of the fault signal
in (a)
47
3.3
Summary of phase-space based classifier
48
3.4
Implementation steps of phase-space-based islanding
detection scheme
50
3.5
Pseudo-code for phase-space-based islanding detection
scheme
51
3.6
Implementation steps of wavelet transform-based islanding
detection scheme
53
4.1
Typical radial feeder of distribution system
57
4.2
General flow chart of BSA
60
4.3
Optimal load shedding scheme using BSA
67
2
xiii
4.4
Pseudo-code for optimal load shedding scheme using BSA
69
4.5
Optimal load shedding scheme using GA
70
5.1
Power distribution systems with two identical DG units
73
5.2
Single-line diagram of IEEE 33-bus radial distribution
system with four DG units
74
5.3
Possible islands and NDZ regions in the radial distribution
system with two DG units
75
5.4
Samples of selected phase-space features for islanding and
non-islanding events at DG1 and DG2 in the studied
system, (a) grid disconnection events (islanding condition),
(b) capacitor switching events (non-islanding condition),
and (c) three-phase fault event (non-islanding condition)
78
5.5
Regression analyses of RBFNN-based DG1 and DG2
classifier using phase-space features
79
5.6
Regression analyses of RBFNN-based DG1 and DG2
classifier using wavelet transform features
80
5.7
Possible islands and NDZ region in the IEEE 33-bus radial
distribution system with four DG units
84
5.8
Samples of selected phase-space features for islanding and
non-islanding events at DG1, DG2, DG3, and DG4 in the
studied system, (a) grid disconnection events (islanding
condition), (b) capacitor switching events (non-islanding
condition), and (c) three-phase fault event (nonislanding condition)
87
5.9
Regression analyses of RBFNN-based DG1, DG2, DG3,
and DG4 classifier using phase-space features
89
5.10
Regression analyses of RBFNN-based DG1, DG2, DG3,
and DG4 classifier using wavelet transform features
90
5.11
Hourly load profile for individual loads
94
5.12
Hourly PV power production: (a) DG1 (b) DG3
95
5.13
Single line diagram of islanded systems, (a) power island
A, (b) power island B, (c) power island C, and (d) power
island D
98
5.14
Daily load profile and power generation for island A
99
5.15
Proposed load shedding scheme performance for island (a)
generation and load mismatch (b) optimum load profile
with load priority limits
100
5.16
Performance of proposed load shedding scheme at hour
9.00 for island A (a) convergence characteristic, (b)
individual load demand after optimization, and (c) voltage
profile
102
xiv
5.17
Performance of proposed load shedding scheme at hour
15.00 for island A (a) individual load demand after
optimization and(b) voltage profile
106
5.18
Performance of load shedding scheme at hour 9.00 for
island A (a) individual load demand after optimization
using GA, (b) comparison of individual load
demand after optimization between BSA and GA, and (c)
comparison of voltage profiles obtained using BSA and
GA
108
5.19
Performance of load shedding scheme at hour 15.00 for
island A (a) individual load demand after optimization
using GA, (b) comparison of individual load demand after
optimization between BSA and GA, and (c) comparison of
voltage profiles obtained using BSA and GA
110
5.20
Performance comparison of GA and BSA in obtaining
optimal load shedding in island A at hour 9.00
111
5.21
Performance comparison of GA and BSA in obtaining
optimal load shedding in island A at hour 15.00
112
5.22
Comparison of individual load demand after optimization
by BSA and GA at hour 9.00 for (a) Power island B, (b)
Power island C, and (c) Power island D
115
5.23
Comparison of voltage profile before and after load
shedding at hour 9.00 for (a) Power island B, (b) Power
island C, and (c) Power island D
116
5.24
Comparison of individual load demand after optimization
between BSA and GA at hour 15.00 for (a) Power island B,
(b) Power island C, and (c) Power island D
118
5.25
Comparison of voltage profile before and after load
shedding at hour 15.00 for (a) Power island B, (b) Power
island C, and (c) Power island D
120
xv
LIST OF ABBREVIATIONS
AFD
Active Frequency Drift
AI
Artificial Intelligent
ANFIS
Adaptive Neuro-Fuzzy Interference System
ANN
Artificial Neural Network
AVR
Automatic Voltage Regulator
BSA
Backtracking Search Optimization Algorithm
CCP
Common Coupling Point
CF
Correlation Factor
CWT
Continuous Wavelet Transform
DG
Distributed Generation
DMS
Distribution Management System
DT
Decision Tree
DWT
Discrete Wavelet Transform
FCM
Frequency Calculator Module
FFT
Fast Fourier Transforms
FL
Fuzzy Logic
FLC
Fuzzy Logic Control
GA
Genetic Algorithm
IM-SMS
Improved- Slip-Mode Frequency Shift
LSCM
Load Shed Controller Module
MAE
Mean Absolute Deviation
MAPE
Mean Absolute Percent Error
MFs
Membership Functions
MG
Microgrid
MPPT
Maximum Power Point Tracker
MRA
Multi-Resolution Analysis
NDZ
Non-Detection Zones
PCC
Point of Common Coupling
PNN
Probabilistic Neural Network
PSO
Particle Swarm Optimizations
xvi
PV
Photovoltaics
RBFNN
Radial Basis Neural Network
RMSE
Root Mean Squared Error
ROCOF
Rate of Change of Frequency
ROCOV
Rate of Change of Voltage
SCADA
Supervisory Control and Data Acquisition
SFS
Sandia Frequency Shift
SMS
Slip-Mode Frequency Shift
SP
Signal Processing
SVM
Support Vector Machines
SVS
Sandia Voltage Shift
T&D
Transmission and Distribution
TFD
Time-Frequency Distribution
THD
Total Harmonic Distortion
UFLS
Under Frequency Load Shedding
UFP/OFP
Under/Over Frequency Protection
UVLS
Under Voltage Load Shedding
UVP/OVP
Under/Over Voltage Protection
VSC
Voltage-Source Control
VSM
Voltage Stability Margin
VU
Voltage Unbalance
WPT
Wavelet Packet Transform
WT
Wavelet Transform
xvii
LIST OF SYMBOLS
Euclidean Space
ˆx i
Column Vector
Sbus (r)
Residual Power
&
AND Operator
*
Complex Conjugate
:=
Update Operation
[P (x|y)]
Probability of Event x while Event y Given
[P (y)]
Overall Probability of All Events y
[P (y|x)]
Probability of Event y while Event x Given
||
OR Operator
∆P
Power Imbalance
a and b
Random Number Between 0 to 1
d
First Order Differential Equation
D
Dimension
dc
Correlation Dimension
∂f/∂t
rate of Change of Frequency (Hz/s).
diag(Ebus)
Diagonal Bus Voltage Matrix
Ebus
Bus Voltage Matrix
Ebus*
Complex Conjugate of the Bus Voltage Vector
Ex
Euclidean Norm
f
Nominal Frequency (Hz)
F
Algorithm Dependent Parameter Utilized to Control the
Amplitude of the Search-Direction
f
Fitness Function
H
Inertia Constant of Generator
h
Priori Probability of Patterns being in Category A or B
hA, hB, nA, and
nB
Data Patterns
xviii
Hz
Hertz
Ibus
Bus Current Matrix
J
Jacobian Matrix in Complex Form
k
Number of Feeders in The System
kV
Kilovolts
kW
Kilowatts
Lfactor
Load Shedding Vector
Li
Loading Index
ms
Milisecond
MVA
Mega Volt-Ampere
MVar
Mega Volt-Ampere reactive
MW
Mega Watt
n
Number of Sample Points
Ns
Sampling Rate in Each Period
oldP
Historical Population
P
Population
P (x)
Probability of Event x
P di
Active Power Consumed by the Load
Pe
Electric Power in Generator
P gen
Generator Power
P gi
Generated Active Power
Pi
Real Power Entering Bus i
P ij
jth individual element in the problem dimension that falls in ith
position in a population dimension
P loss
Active Power Losses in the Network
Pm
Prime Mover Output Power
P mi
Prime Mover Input Power
P remaining load
Total Remaining Load
Q di
Reactive Power Consumed by the Load
Q gi
Generated Reactive Power
Qi
Reactive Power Entering Bus i
xix
Q loss
Reactive Power Losses in the Network
rand
Random Value Obtained from a Standard Normal Distribution
s
Second
Sbus
Vector of Bus Complex Power Vector
Sl
Apparent Power
Sl-i
Value of Remaining Load Power
Sl-max
Maximum Thermal Limit
Spriority
Priority Load Limit
T
Trial Population
U
Uniform Distribution
up and low
Upper and Lower Boundaries
Vdc
Direct Current Voltage
Vi-max
Maximum Permissible Voltage at Bus i
Vi-min
Minimum Permissible Value of the Voltage at Bus i
Vk
Voltage at Bus k
Vm
Voltage at Bus m
VSMsys
Overall System Voltage Stability Margin
xi
Row Vector
Xi
Original Data
Xi^
Forecasting Data
xwi
Weight Input of xi to Neuron
Ybus
Element of the Bus Admittance Matrix
Ybus*
Complex Conjugate of the Bus Admittance Matrix
Ydgi
Outputs of the Individual Classifier
Youtput
Final Output of the Decision Making
δkm
Angle Between Bus k and Bus m
θ (a )
Heaviside Step Function
Smoothing Parameter
Time Delayed
r
A Set of Branch Constituting the Feeder
1
CHAPTER I
INTRODUCTION
1.1
RESEARCH BACKGROUND
Traditionally, electrical energy in the distribution system is always supplied to the
customer from upstream power resources that are connected to the bulk transmission
system. A small localized power source called distributed generation (DG) becomes
an alternative to bulk electric power generation due to yearly demand growth. These
DGs include wind farms, micro hydro turbines, photovoltaics (PV), and other
generators. These DGs are generally in the range of a few kWs up to a few MWs and
have several advantages, such as environmental benefits, improved reliability,
increased efficiency, prevention of transmission and distribution (T&D) capacity
upgrades, improved power quality, and reduced T&D line losses (Balaguer-álvarez et
al. 2010; Ray et al. 2011). Figure 1.1 shows the difference between traditional and
multiple embedded distribution systems, in which additional DG is commonly
connected near the local load compared with the traditional network system.
Therefore, the traditional approach of energy production and distribution are
changing, introducing new challenges in balancing the power system.
One of the major drawbacks of DGs are island mode operation. Disconnection
of the main source is called islanding as shown in Figure 1.2, which can be either
intentional or unintentional. When disconnection occurs, the active part of the
distribution system should sense the disconnection from the main grid and shut down
the DGs, where island operation is prohibited or control action must be activated to
stabilize the islanded part of the system (Ezzt et al. 2007).
2
External Grid
External Grid
DG
consumers
consumers
Distribution
consumers
consumers
consumers
consumers
DG
consumers
Distribution
consumers
consumers
consumers
consumers
(a)
DG
consumers
(b)
Figure 1.1 (a) Traditional distribution system, (b) generation embedded distribution
system
Tripped Utility Circuit
Breaker
Local
Load
Utility
Network
Load
Power
Island
DG
Industrial Site
Figure 1.2 Power islanding condition
Sources: Ezzt et al. 2007
However, unpredictable behavior due to power mismatch between load and
generation immediately after island creation is a new challenge in controlling the
voltage and frequency response. Normally, large excess load over local generation in
the islanded system could result in a rapid frequency drop. This rapid frequency drop
will cause the load to acquire power from the stored kinetic energy in the rotating
system, thus slowing the rotation (frequency). Moreover, the voltage of the system
decreases rapidly to balance the reactive power in the islanded system. The effect of
the unintentional islanding events described previously is plotted as frequency and
voltage response versus time in Figure 1.3. Any variation in frequency and voltage
may damage the customer equipment. To overcome this problem, various solutions
are being proposed. One of the effective methods of correcting power mismatch is to
quickly shed the load before frequency and voltage stability decline sharply.
Frequency / Voltage
3
Fault occurs, initiating island
Frequency reduces to balance active power
Voltage drops to balance reactive power
Voltage
Frequency
AVR boots
Voltage stabilised excitation current
by generator field
voltage
Steady-state
Grid-connected
Transient period
Governor boosts
power output
Steady-state
islanded
Time
Figure 1.3 Voltage and frequency response
Source: Ecconnect 2001
Generally, automatic load shedding has two types. The first type is underfrequency load shedding (UFLS), which is designed to rebalance load and generation
within an electrical island once the unbalanced system is created. The second type is
under-voltage load shedding (UVLS), which is utilized to prevent local area voltage
collapse and to directly respond to the voltage condition in a local area. The UVLS
scheme aims to shed load to restore reactive power relative to demand, to prevent
voltage collapse, and to contain a voltage problem within a local area rather than
allowing it to spread in geography and magnitude.
By contrast, automatic UFLS is designed for extreme conditions to stabilize
the balance between generation and load after electrical island formation and to drop
sufficient load to allow the frequency to stabilize the island. However, the UFLS is
ineffective if instability or voltage collapse occurs within the island. Moreover, the
most common factor that contributes to power blackout is voltage instability (Laghari
et al. 2013). Thus, effective load shedding is crucial to prevent total system collapse.
Improper load shedding would cause a high number of blackouts.
4
1.2
PROBLEM STATEMENT
Although the islanding operation has some benefits, several drawbacks are still
observed, especially in unintentional islanding events. The unintentional islanding of
DGs may cause problems in terms of power quality, safety, voltage and frequency
stability, and interference (Mahat et al. 2011; Timbus et al. 2010). The Institute of
Electrical and Electronics Engineers (IEEE) 1547-2003 standard specifies a maximum
delay of 2 s for the detection of the unintentional islanding condition, whereas the
IEEE 929-2000 standard requires the disconnection of the DG if islanded (Mahat et al.
2011). To achieve this goal, each DG must be capable of detecting the islanding
condition as quickly as possible. Therefore, the first part of the current study attempts
to develop a simple and effective method that can quickly diagnose the islanding
condition by identifying the islanding and non-islanding conditions in the system.
Several techniques have been developed to accurately identify the islanding
condition; the most economical and effective technique is to use a passive technique
with the application of artificial intelligence (AI). This technique is preferred because
a more accurate online detection is required to monitor the condition of the system.
Moreover, this technique is usually less complex and has high computational
efficiency with good accuracy and reliability. The most common technique being used
nowadays is the combination of signal processing (SP) and neural network.
For instance, Gaing (2004) used discrete wavelet transform (DWT) integrated
probabilistic neural network (PNN) to classify power disturbances. This method used
the multiresolution analysis (MRA) of DWT and Parseval’s theorem to extract energy
distribution features at different resolution levels. The features were subsequently
classified using PNN. Moreover, PNN is incorporated with wavelet transform (WT) to
determine the location and type of the fault (Othman, & Amari 2008). Realizing the
potential of these intelligent methods, G. Yin (2005) presented a combination of fast
Fourier transform (FFT) and artificial neural network (ANN) classifier for detecting
the islanding condition (G. Yin 2005). In this method, the output voltages of the
inverter are sampled and the signal frequency domain is obtained through FFT.
However, the algorithm is suitable for stationary waveforms and can be implemented
SCHEME FOR DISPERSED GENERATION INTEGRATED
RADIAL DISTRIBUTION SYSTEMS
AZIAH KHAMIS
THESIS SUBMITTED IN FULFILMENT OF THE DEGREE OF
DOCTOR OF PHILOSOPHY
FACULTY OF ENGINEERING AND BUILT ENVIRONMENT
UNIVERSITI KEBANGSAAN MALAYSIA
BANGI
2014
PENGESANAN DAN PENGKELASAN KEPULAUAN DAN SKIM
PENYISIHAN BEBAN BAGI SISTEM PENGAGIHAN
JEJARI TERSEPADU PENJANA TERAGIH
AZIAH KHAMIS
TESIS YANG DIKEMUKAKAN UNTUK MEMPEROLEH IJAZAH
DOKTOR FALSAFAH
FAKULTI KEJURUTERAAN DAN ALAM BINA
UNIVERSITI KEBANGSAAN MALAYSIA
BANGI
2014
iii
DECLARATION
I hereby declare that the work in this thesis is my own except for quotations and
summaries which have been duly acknowledged.
3 Nov 2014
AZIAH KHAMIS
P59755
iv
ACKNOWLEDGMENTS
First and foremost praise be to Almighty Allah for all His blessings for giving me
perseverance and good health throughout the duration of this PhD research.
I would like to express sincere appreciation for the intelligent advice,
encouragement and guidance of my main supervisor, Assoc. Prof. Dr. Hussain
Shareef. Without his tireless assistance, leadership, and confidence in my abilities, this
thesis would not come to its timely completion.
Furthermore, I would like to express my high appreciation to my co-supervisor
Prof. Dr. Hjh. Azah Mohamed for the valued knowledge, ideas, encouragement,
assistance and support received from her during my PhD program.
I would like to acknowledge the financial support from Ministry of Higher
Education, Universiti Teknikal Malaysia Melaka and also Universiti Kebangsaan
Malaysia for making it possible for me to pursue and complete my PhD degree.
I would like to thank all Power System Research group UKM for their help,
friendship, and creating a pleasant working environment throughout my years in
UKM. My heartfelt appreciation goes to all my special friends, especially Hazilah
Abdullah and Abdul Zaini Abdullah for their inseparable support and prayers.
To my dearest husband, Ahmad Zulkarnain Rosli, thanks for your do’as,
patience, understanding and support for all the duration of doing this research. Last
but not least, my special thanks to my family who has been my inspiration since the
primary school until this stage of my education. I dedicate this thesis to my beloved
late mother, Ramlah Binti Karim, who died before I continue my PhD study.
v
ABSTRACT
The high penetration level of distributed generation (DG) provides numerous potential
environmental benefits, such as high reliability, efficiency, and low carbon emissions.
However, the effective detection of islanding and rapid DG disconnection is essential
to prevent safety problems and equipment damage caused by the island mode
operations of DGs. The common islanding protection technology is based on passive
techniques that do not perturb the system but have large nondetection zones.
Therefore, the first part of this thesis attempts to develop a simple and effective
passive islanding detection method with reference to a probabilistic neural networkbased classifier, as well as utilizes the features extracted from three-phase voltages
seen at the DG terminal. This approach enables initial features to be obtained using
the phase-space technique. This technique analyzes the time series in a higher
dimensional space, revealing several hidden features of the original signal.
Meanwhile, the second part of the thesis focuses on the development of an optimal
load shedding scheme after the system experiences an unintentional islanding state to
prevent system collapse due to load-generation mismatch and voltage instability
encountered in the islanded part of the system. To handle this optimization problem, a
constraint multiobjective function that considers the linear static voltage stability
margin and amount of load curtailment was formulated. A novel heuristic
optimization technique based on the backtracking search algorithm (BSA) was
subsequently proposed as an optimization tool for determining the optimum load
shedding based on the proposed objective function. Several test systems, including a
radial distribution system with two DG units and the Institute of Electrical and
Electronics Engineers (IEEE) 33-bus radial distribution system with four DG units,
were utilized to evaluate the effectiveness of the proposed islanding detection method
and the optimal load shedding scheme. The effectiveness of the proposed islanding
detection method was verified by comparing its results with the conventional wavelet
transform (WT)-based technique through intensive simulations conducted with the
DIgSILENT Power Factory® software. The assessment indices, namely, the mean
absolute percentage error (MAPE), mean absolute error (MAE), and root mean square
error (RMSE), obtained a 0% error rate for the proposed method when applied to the
IEEE 33-bus radial distribution system with four DG units. Meanwhile, the MAPE,
MAE, and RMSE obtained error rates of 0.1482%, 0.2752%, and 5.055%, respectively,
for the WT-based technique when applied to the IEEE 33-bus radial distribution
system with four DG units. These indices show that the proposed islanding detection
method is robust and capable of sensing the difference between the islanding
condition and other system disturbances. Meanwhile, the performance of the proposed
load shedding scheme was evaluated through an extensive test conducted on the IEEE
33-bus system considering several scenarios such as load shedding under various
operating points and at various islands using the MATLAB® software. Moreover, the
effectiveness of the proposed scheme was validated by comparing its results with
those obtained using the genetic algorithm (GA). The optimization results indicate that
the proposed BSA technique is more effective in determining the optimal amount of
load to be shed in any islanded system compared with GA.
vi
ABSTRAK
Tahap peningkatan penembusan penjana teragih (PT) yang tinggi memberikan
pelbagai faedah alam sekitar seperti keboleharapan yang tinggi, kecekapan dan
pengeluaran karbon yang rendah. Walau bagaimanapun, pengesanan kepulauan yang
berkesan adalah penting bagi mengelakkan masalah keselamatan dan kerosakan
peralatan disebabkan oleh operasi kepulauan PT. Teknologi perlindungan kepulauan
yang biasa digunakan adalah berasaskan teknik pasif. Oleh itu, bahagian pertama tesis
ini bertujuan untuk membangunkan satu kaedah pengesanan kepulauan pasif yang
mudah dan berkesan dengan menggunakan rangkaian neural kebarangkalian untuk
pengelasan, dan juga mengunakan ciri yang diekstrak dari voltan tiga fasa yang dilihat
pada pengkalan PT. Pendekatan ini membolehkan perolehan ciri awal dengan
menggunakan teknik fasa-ruang. Teknik ini menganalisa siri masa dalam ruang
dimensi yang lebih tinggi, dan mendedahkan beberapa ciri tersembunyi dari isyarat
asal. Sementara itu, bahagian kedua tesis ini memberi tumpuan kepada pembangunan
skim penyisihan beban optimum selepas sistem mengalami kepulauan yang tidak
disengajakan untuk mengelakkan ketakstabilan voltan di bahagian sistem kepulauan.
Untuk menangani masalah pengoptimuman, fungsi objektif berbilang
dipertimbangkan dengan mengambilkira kestabilan jidar voltan statik dan jumlah
pengurangan voltan. Satu teknik pengoptimuman heuristik yang novel, iaitu, algoritma
carian jejak balik (ACJB) telah dicadangkan untuk menentukan penyisihan beban
yang optimum. Beberapa sistem ujian, termasuk sistem jejari dengan dua unit PT dan
juga sistem agihan jejari IEEE 33 bas dengan empat unit PT, masing-masing, telah
digunakan untuk menilai keberkesanan kaedah pengesanan kepulauan dan penyisihan
beban optimum yang dicadangkan. Keberkesanan kaedah pengesanan kepulauan yang
dicadangkan telah disahkan melalui perbandingan dengan teknik anak gelombang
melalui simulasi intensif menggunakan perisian DIgSILENT Power Factory®. Indeks
penilaian, seperti peratusan min ralat mutlak (PMRM), min ralat multak (MRM), dan
ralat punca min kuasa dua (RPMKD ) didapati kadar ralat adalah 0% bagi kaedah
pengesanan kepulauan yang dicadangkan apabila diuji pada sistem agihan jejari IEEE
33 bas dengan empat unit PT. Manakala, PMRM, MRM dan RPMKD memperoleh
kadar ralat sebanyak 0.1482%, 0.2752%, and 5.055%, masing-masing, bagi teknik
berasaskan anak gelombang apabila diuji pada sistem agihan jejari IEEE 33 bas
dengan empat unit PT. Hasil keputusan indeks menunjukkan kaedah pengesanan
kepulauan yang dicadangkan adalah teguh dan mampu mengesan perbezaan antara
kepulauan serta gangguan sistem yang lain. Sementara itu, prestasi skim penyisihan
beban yang dicadangkan telah dinilai dengan ujian yang meluas dijalankan pada
sistem bas IEEE 33 dengan mempertimbangkan beberapa scenario seperti penyisihan
beban di bawah pelbagai titik operasi dan pulau dalam perisian MATLAB®.
Tambahan pula, keberkesanan skim penyisihan yang dicadangkan telah disahkan
dengan membandingkan keputusan yang telah diperolehi dengan keputusan algoritma
genetic (AG). Hasil kajian pengoptimuman menunjukkan bahawa teknik ACJB yang
dicadangkan adalah lebih berkesan dalam menentukan nilai optimum bagi beban yang
akan dikurangkan pada mana-mana sistem kepulauan berbanding AG.
vii
TABLE OF CONTENTS
Page
DECLARATION
iii
ACKNOWLEDGMENTS
iv
ABSTRACT
v
ABSTRAK
vi
CONTENTS
vii
LIST OF TABLES
x
LIST OF FIGURES
xii
LIST OF ABBREVIATIONS
xv
LIST OF SYMBOLS
xvii
CHAPTER I
INTRODUCTION
1.1
Research Background
1
1.2
Problem Statement
4
1.3
Objective of the Research
6
1.4
Scope of the Research
6
1.5
Organization of the Thesis
7
CHAPTER II
LITERATURE REVIEW
2.1
Islanding Detection Methods
8
2.1.1 Central Islanding Detection Techniques
9
2.1.2 Review of the Conventional Local Islanding Detection 12
Technique
2.1.3 Review of the Intelligent Local Islanding Detection
16
Technique
2.2
DG Models for Islanding Detection
23
2.3
Load Shedding Schemes
25
2.3.1 Review of Under-Frequency Load Shedding Schemes 26
2.3.2 Review of Under-Voltage Load Shedding Schemes
29
2.4
DG Models for Optimal Load Shedding
33
2.5
Types of Distribution System Design
35
viii
2.6
Chapter Summary
CHAPTER III
ISLANDING DETECTION USING PHASE-SPACE
AND NEURAL NETWORK
3.1
Introduction
38
3.2
Tools and Methods used in the Proposed Method
39
3.2.1 Phase-Space Technique
3.2.2 Probabilistic Neural Network
39
41
Proposed Islanding Detection Method
44
3.3.1 Data Collection
3.3.2 Phase-Space Feature Extraction
3.3.3 Design of Artificial Intelligent Classifier
44
45
47
Performance Evaluation Methods
52
3.4.1 Performance Evaluation of Conventional Method
3.4.2 Performance Evaluation with Statistical Indices
52
54
3.5
Chapter Summary
55
CHAPTER IV
OPTIMAL LOAD SHEDDING SCHEME USING
BACKTRACKING SEARCH ALGORITHM
4.1
Introduction
4.2
Tools and Methods Used in Proposed Load Shedding Scheme 56
3.3
3.4
36
56
4.2.1 Voltage Stability Margin
4.2.2 Backtracking Search Optimization Algorithm
4.2.3 Newton–Raphson Power Flow Solution
56
58
61
Problem Formulation for Optimal Load Shedding Scheme
62
4.3.1 Operational Constraints
4.3.2 Fitness Function
4.3.3 Application of BSA for Optimal Load Shedding
Scheme
63
64
65
Performance Evaluation Scheme
69
4.4.1 Performance Evaluation with Conventional GA
Method
69
4.5
Chapter Summary
71
CHAPTER V
RESULTS AND DISCUSSION
5.1
The Test System for Islanding Detection
5.1.1 Radial Distribution System with Two Identical DG
4.3
4.4
72
ix
5.2
5.3
Units
5.1.2 IEEE 33-Bus Radial Distribution System with Four
DG Units
72
73
Test Results of the Radial Distribution System with Two
Identical DG Units
74
5.2.1
5.2.2
5.2.3
5.2.4
5.2.5
5.2.6
76
78
79
80
81
81
Input Feature Extraction
Results of RBFNN with Phase-Space Features
Results of RBFNN with Wavelet Transform Features
Results of PNN with Phase-Space Features
Results of PNN with Wavelet Transform Features
Summary of the Result Obtained for All of the Tested
Islanding Detection Methods
Implementation of Islanding Detection on The IEEE 33-bus
Radial Distribution Systems with Four DG Units
84
5.3.1
5.3.2
5.3.3
5.3.4
5.3.5
5.3.6
85
88
89
90
91
91
Input Features Extraction
Results of RBFNN with Phase-Space Features
Results of RBFNN with Wavelet Transform Features
Results of PNN with Phase-Space Features
Results of PNN with Wavelet Transform Features
Summary of Islanding Detection Analysis
5.4
Description of the Test System for Optimal Load
Shedding Scheme
93
5.5
Test Results of Optimal Load Shedding Scheme
96
5.5.1
5.5.2
5.5.3
5.5.4
Optimal Load Shedding for Island A Using BSA
Optimal Load Shedding for Island A Using GA
Optimal Load Shedding for Other Islanded Systems
Summary of Load Shedding Scheme
99
106
113
124
5.6
Chapter Summary
CHAPTER VI
CONCLUSIONS AND RECOMMENDATIONS
6.1
Overall Conclusions
126
6.2
Significant Contributions of the Research
128
6.3
Recommendations for Future Studies
129
REFERENCES
125
130
APPENDIXES
A
IEEE 33-Bus Radial Distribution System
140
B
Wavelet Technique as Features Extraction
141
C
List of Publication
143
x
LIST OF TABLES
Table Number
Page
2.1
Summary of remote islanding detection techniques
12
2.2
Comparison of passive islanding detection techniques
13
2.3
Comparison of active islanding detection techniques
15
2.4
Utilization of SP in islanding detection
20
2.5
Utilization of the AI classifier in islanding detection at DG
23
2.6
Comparison of various computational intelligence load
shedding schemes
33
3.1
Classifier output definition
45
3.2
Selected phase-space features
45
3.3
A scale of judgment of forecasting accuracy
55
5.1
System model description
72
5.2
DG installed node with operating points
74
5.3
Number of samples for training and testing
76
5.4
Parameter settings of the RBFNN and PNN classifiers for
DG1 and DG2
77
5.5
RBFNN classification results with phase-space features
79
5.6
RBFNN classification results with wavelet transform
features
80
5.7
PNN classification results with phase-space features
81
5.8
PNN classification results with wavelet transform features
81
5.9
Comparison of PNN classifier performance with phase-space
and wavelet features
82
5.10
Comparison of MAPE, RMSE, and MAE for various
islanding detection methods
83
5.11
Number of samples for training and testing at IEEE 33 bus
system
85
5.12
Parameter settings of the RBFNN and PNN classifiers for
DG1, DG2, DG3, and DG4
88
5.13
RBFNN classification results with phase-space features
88
5.14
RBFNN classification results with wavelet transform
features
90
5.15
PNN classification results with phase-space features
91
xi
5.16
PNN classification results with wavelet transform features
91
5.17
Comparison of PNN classifier performance with phase-space
and wavelet feature
92
5.18
Comparison of MAPE, RMSE, and MAE for various
islanding detection methods
92
5.19
Rated maximum power of DGs
93
5.20
Percentage load priority limits for the IEEE 33-bus radial
distribution system
94
5.21
Overall power demand and supply in islanded system
96
5.22
GA and BSA parameter settings
96
5.23
Amount of hourly load curtailment at individual buses in
island A
103
5.24
Summary of load shedding performances at hour 9.00
113
5.25
Summary of load shedding performances at hour 15.00
117
5.26
Amount of hourly load curtailment at individual buses in
island B
121
5.27
Amount of hourly load curtailment at individual buses in
island C
122
5.28
Amount of hourly load curtailment at individual buses in
island D
123
5.29
Performance of BSA and GA in terms of fitness, VSM, and
amount of load curtailment at hour 9.00
124
5.30
Performance of BSA and GA in terms of fitness, VSM, and
amount of load curtailment at hour 15.00
124
xii
LIST OF FIGURES
Figure Number
Page
1.1
(a) Traditional distribution
embedded distribution system
system,
(b)
generation
1.2
Power islanding condition
2
1.3
Voltage and frequency response
3
2.1
Classification of islanding detection technique
9
2.2
Remote islanding detection technique
9
2.3
Transfer trip scheme
11
2.4
Basic block of intelligent islanding detection technique and
classification
16
2.5
Block diagram of the inverter
24
2.6
Schematic diagram of mini hydro power for grid connected
operation
24
2.7
Types of load shedding schemes
25
2.8
Generator model
34
2.9
System for load shedding
35
2.10
Simplified illustration of the concept behind three types of
power distribution configuration
36
3.1
Architecture of a PNN
43
3.2
Phase-space feature extraction, (a) three-phase fault, (b)
Euclidean norm (Ex) of the fault signal in (a), (c) selected
region of Euclidean norm (Ex) for feature selections, and
(d) Euclidean norm (Ex) and its features of the fault signal
in (a)
47
3.3
Summary of phase-space based classifier
48
3.4
Implementation steps of phase-space-based islanding
detection scheme
50
3.5
Pseudo-code for phase-space-based islanding detection
scheme
51
3.6
Implementation steps of wavelet transform-based islanding
detection scheme
53
4.1
Typical radial feeder of distribution system
57
4.2
General flow chart of BSA
60
4.3
Optimal load shedding scheme using BSA
67
2
xiii
4.4
Pseudo-code for optimal load shedding scheme using BSA
69
4.5
Optimal load shedding scheme using GA
70
5.1
Power distribution systems with two identical DG units
73
5.2
Single-line diagram of IEEE 33-bus radial distribution
system with four DG units
74
5.3
Possible islands and NDZ regions in the radial distribution
system with two DG units
75
5.4
Samples of selected phase-space features for islanding and
non-islanding events at DG1 and DG2 in the studied
system, (a) grid disconnection events (islanding condition),
(b) capacitor switching events (non-islanding condition),
and (c) three-phase fault event (non-islanding condition)
78
5.5
Regression analyses of RBFNN-based DG1 and DG2
classifier using phase-space features
79
5.6
Regression analyses of RBFNN-based DG1 and DG2
classifier using wavelet transform features
80
5.7
Possible islands and NDZ region in the IEEE 33-bus radial
distribution system with four DG units
84
5.8
Samples of selected phase-space features for islanding and
non-islanding events at DG1, DG2, DG3, and DG4 in the
studied system, (a) grid disconnection events (islanding
condition), (b) capacitor switching events (non-islanding
condition), and (c) three-phase fault event (nonislanding condition)
87
5.9
Regression analyses of RBFNN-based DG1, DG2, DG3,
and DG4 classifier using phase-space features
89
5.10
Regression analyses of RBFNN-based DG1, DG2, DG3,
and DG4 classifier using wavelet transform features
90
5.11
Hourly load profile for individual loads
94
5.12
Hourly PV power production: (a) DG1 (b) DG3
95
5.13
Single line diagram of islanded systems, (a) power island
A, (b) power island B, (c) power island C, and (d) power
island D
98
5.14
Daily load profile and power generation for island A
99
5.15
Proposed load shedding scheme performance for island (a)
generation and load mismatch (b) optimum load profile
with load priority limits
100
5.16
Performance of proposed load shedding scheme at hour
9.00 for island A (a) convergence characteristic, (b)
individual load demand after optimization, and (c) voltage
profile
102
xiv
5.17
Performance of proposed load shedding scheme at hour
15.00 for island A (a) individual load demand after
optimization and(b) voltage profile
106
5.18
Performance of load shedding scheme at hour 9.00 for
island A (a) individual load demand after optimization
using GA, (b) comparison of individual load
demand after optimization between BSA and GA, and (c)
comparison of voltage profiles obtained using BSA and
GA
108
5.19
Performance of load shedding scheme at hour 15.00 for
island A (a) individual load demand after optimization
using GA, (b) comparison of individual load demand after
optimization between BSA and GA, and (c) comparison of
voltage profiles obtained using BSA and GA
110
5.20
Performance comparison of GA and BSA in obtaining
optimal load shedding in island A at hour 9.00
111
5.21
Performance comparison of GA and BSA in obtaining
optimal load shedding in island A at hour 15.00
112
5.22
Comparison of individual load demand after optimization
by BSA and GA at hour 9.00 for (a) Power island B, (b)
Power island C, and (c) Power island D
115
5.23
Comparison of voltage profile before and after load
shedding at hour 9.00 for (a) Power island B, (b) Power
island C, and (c) Power island D
116
5.24
Comparison of individual load demand after optimization
between BSA and GA at hour 15.00 for (a) Power island B,
(b) Power island C, and (c) Power island D
118
5.25
Comparison of voltage profile before and after load
shedding at hour 15.00 for (a) Power island B, (b) Power
island C, and (c) Power island D
120
xv
LIST OF ABBREVIATIONS
AFD
Active Frequency Drift
AI
Artificial Intelligent
ANFIS
Adaptive Neuro-Fuzzy Interference System
ANN
Artificial Neural Network
AVR
Automatic Voltage Regulator
BSA
Backtracking Search Optimization Algorithm
CCP
Common Coupling Point
CF
Correlation Factor
CWT
Continuous Wavelet Transform
DG
Distributed Generation
DMS
Distribution Management System
DT
Decision Tree
DWT
Discrete Wavelet Transform
FCM
Frequency Calculator Module
FFT
Fast Fourier Transforms
FL
Fuzzy Logic
FLC
Fuzzy Logic Control
GA
Genetic Algorithm
IM-SMS
Improved- Slip-Mode Frequency Shift
LSCM
Load Shed Controller Module
MAE
Mean Absolute Deviation
MAPE
Mean Absolute Percent Error
MFs
Membership Functions
MG
Microgrid
MPPT
Maximum Power Point Tracker
MRA
Multi-Resolution Analysis
NDZ
Non-Detection Zones
PCC
Point of Common Coupling
PNN
Probabilistic Neural Network
PSO
Particle Swarm Optimizations
xvi
PV
Photovoltaics
RBFNN
Radial Basis Neural Network
RMSE
Root Mean Squared Error
ROCOF
Rate of Change of Frequency
ROCOV
Rate of Change of Voltage
SCADA
Supervisory Control and Data Acquisition
SFS
Sandia Frequency Shift
SMS
Slip-Mode Frequency Shift
SP
Signal Processing
SVM
Support Vector Machines
SVS
Sandia Voltage Shift
T&D
Transmission and Distribution
TFD
Time-Frequency Distribution
THD
Total Harmonic Distortion
UFLS
Under Frequency Load Shedding
UFP/OFP
Under/Over Frequency Protection
UVLS
Under Voltage Load Shedding
UVP/OVP
Under/Over Voltage Protection
VSC
Voltage-Source Control
VSM
Voltage Stability Margin
VU
Voltage Unbalance
WPT
Wavelet Packet Transform
WT
Wavelet Transform
xvii
LIST OF SYMBOLS
Euclidean Space
ˆx i
Column Vector
Sbus (r)
Residual Power
&
AND Operator
*
Complex Conjugate
:=
Update Operation
[P (x|y)]
Probability of Event x while Event y Given
[P (y)]
Overall Probability of All Events y
[P (y|x)]
Probability of Event y while Event x Given
||
OR Operator
∆P
Power Imbalance
a and b
Random Number Between 0 to 1
d
First Order Differential Equation
D
Dimension
dc
Correlation Dimension
∂f/∂t
rate of Change of Frequency (Hz/s).
diag(Ebus)
Diagonal Bus Voltage Matrix
Ebus
Bus Voltage Matrix
Ebus*
Complex Conjugate of the Bus Voltage Vector
Ex
Euclidean Norm
f
Nominal Frequency (Hz)
F
Algorithm Dependent Parameter Utilized to Control the
Amplitude of the Search-Direction
f
Fitness Function
H
Inertia Constant of Generator
h
Priori Probability of Patterns being in Category A or B
hA, hB, nA, and
nB
Data Patterns
xviii
Hz
Hertz
Ibus
Bus Current Matrix
J
Jacobian Matrix in Complex Form
k
Number of Feeders in The System
kV
Kilovolts
kW
Kilowatts
Lfactor
Load Shedding Vector
Li
Loading Index
ms
Milisecond
MVA
Mega Volt-Ampere
MVar
Mega Volt-Ampere reactive
MW
Mega Watt
n
Number of Sample Points
Ns
Sampling Rate in Each Period
oldP
Historical Population
P
Population
P (x)
Probability of Event x
P di
Active Power Consumed by the Load
Pe
Electric Power in Generator
P gen
Generator Power
P gi
Generated Active Power
Pi
Real Power Entering Bus i
P ij
jth individual element in the problem dimension that falls in ith
position in a population dimension
P loss
Active Power Losses in the Network
Pm
Prime Mover Output Power
P mi
Prime Mover Input Power
P remaining load
Total Remaining Load
Q di
Reactive Power Consumed by the Load
Q gi
Generated Reactive Power
Qi
Reactive Power Entering Bus i
xix
Q loss
Reactive Power Losses in the Network
rand
Random Value Obtained from a Standard Normal Distribution
s
Second
Sbus
Vector of Bus Complex Power Vector
Sl
Apparent Power
Sl-i
Value of Remaining Load Power
Sl-max
Maximum Thermal Limit
Spriority
Priority Load Limit
T
Trial Population
U
Uniform Distribution
up and low
Upper and Lower Boundaries
Vdc
Direct Current Voltage
Vi-max
Maximum Permissible Voltage at Bus i
Vi-min
Minimum Permissible Value of the Voltage at Bus i
Vk
Voltage at Bus k
Vm
Voltage at Bus m
VSMsys
Overall System Voltage Stability Margin
xi
Row Vector
Xi
Original Data
Xi^
Forecasting Data
xwi
Weight Input of xi to Neuron
Ybus
Element of the Bus Admittance Matrix
Ybus*
Complex Conjugate of the Bus Admittance Matrix
Ydgi
Outputs of the Individual Classifier
Youtput
Final Output of the Decision Making
δkm
Angle Between Bus k and Bus m
θ (a )
Heaviside Step Function
Smoothing Parameter
Time Delayed
r
A Set of Branch Constituting the Feeder
1
CHAPTER I
INTRODUCTION
1.1
RESEARCH BACKGROUND
Traditionally, electrical energy in the distribution system is always supplied to the
customer from upstream power resources that are connected to the bulk transmission
system. A small localized power source called distributed generation (DG) becomes
an alternative to bulk electric power generation due to yearly demand growth. These
DGs include wind farms, micro hydro turbines, photovoltaics (PV), and other
generators. These DGs are generally in the range of a few kWs up to a few MWs and
have several advantages, such as environmental benefits, improved reliability,
increased efficiency, prevention of transmission and distribution (T&D) capacity
upgrades, improved power quality, and reduced T&D line losses (Balaguer-álvarez et
al. 2010; Ray et al. 2011). Figure 1.1 shows the difference between traditional and
multiple embedded distribution systems, in which additional DG is commonly
connected near the local load compared with the traditional network system.
Therefore, the traditional approach of energy production and distribution are
changing, introducing new challenges in balancing the power system.
One of the major drawbacks of DGs are island mode operation. Disconnection
of the main source is called islanding as shown in Figure 1.2, which can be either
intentional or unintentional. When disconnection occurs, the active part of the
distribution system should sense the disconnection from the main grid and shut down
the DGs, where island operation is prohibited or control action must be activated to
stabilize the islanded part of the system (Ezzt et al. 2007).
2
External Grid
External Grid
DG
consumers
consumers
Distribution
consumers
consumers
consumers
consumers
DG
consumers
Distribution
consumers
consumers
consumers
consumers
(a)
DG
consumers
(b)
Figure 1.1 (a) Traditional distribution system, (b) generation embedded distribution
system
Tripped Utility Circuit
Breaker
Local
Load
Utility
Network
Load
Power
Island
DG
Industrial Site
Figure 1.2 Power islanding condition
Sources: Ezzt et al. 2007
However, unpredictable behavior due to power mismatch between load and
generation immediately after island creation is a new challenge in controlling the
voltage and frequency response. Normally, large excess load over local generation in
the islanded system could result in a rapid frequency drop. This rapid frequency drop
will cause the load to acquire power from the stored kinetic energy in the rotating
system, thus slowing the rotation (frequency). Moreover, the voltage of the system
decreases rapidly to balance the reactive power in the islanded system. The effect of
the unintentional islanding events described previously is plotted as frequency and
voltage response versus time in Figure 1.3. Any variation in frequency and voltage
may damage the customer equipment. To overcome this problem, various solutions
are being proposed. One of the effective methods of correcting power mismatch is to
quickly shed the load before frequency and voltage stability decline sharply.
Frequency / Voltage
3
Fault occurs, initiating island
Frequency reduces to balance active power
Voltage drops to balance reactive power
Voltage
Frequency
AVR boots
Voltage stabilised excitation current
by generator field
voltage
Steady-state
Grid-connected
Transient period
Governor boosts
power output
Steady-state
islanded
Time
Figure 1.3 Voltage and frequency response
Source: Ecconnect 2001
Generally, automatic load shedding has two types. The first type is underfrequency load shedding (UFLS), which is designed to rebalance load and generation
within an electrical island once the unbalanced system is created. The second type is
under-voltage load shedding (UVLS), which is utilized to prevent local area voltage
collapse and to directly respond to the voltage condition in a local area. The UVLS
scheme aims to shed load to restore reactive power relative to demand, to prevent
voltage collapse, and to contain a voltage problem within a local area rather than
allowing it to spread in geography and magnitude.
By contrast, automatic UFLS is designed for extreme conditions to stabilize
the balance between generation and load after electrical island formation and to drop
sufficient load to allow the frequency to stabilize the island. However, the UFLS is
ineffective if instability or voltage collapse occurs within the island. Moreover, the
most common factor that contributes to power blackout is voltage instability (Laghari
et al. 2013). Thus, effective load shedding is crucial to prevent total system collapse.
Improper load shedding would cause a high number of blackouts.
4
1.2
PROBLEM STATEMENT
Although the islanding operation has some benefits, several drawbacks are still
observed, especially in unintentional islanding events. The unintentional islanding of
DGs may cause problems in terms of power quality, safety, voltage and frequency
stability, and interference (Mahat et al. 2011; Timbus et al. 2010). The Institute of
Electrical and Electronics Engineers (IEEE) 1547-2003 standard specifies a maximum
delay of 2 s for the detection of the unintentional islanding condition, whereas the
IEEE 929-2000 standard requires the disconnection of the DG if islanded (Mahat et al.
2011). To achieve this goal, each DG must be capable of detecting the islanding
condition as quickly as possible. Therefore, the first part of the current study attempts
to develop a simple and effective method that can quickly diagnose the islanding
condition by identifying the islanding and non-islanding conditions in the system.
Several techniques have been developed to accurately identify the islanding
condition; the most economical and effective technique is to use a passive technique
with the application of artificial intelligence (AI). This technique is preferred because
a more accurate online detection is required to monitor the condition of the system.
Moreover, this technique is usually less complex and has high computational
efficiency with good accuracy and reliability. The most common technique being used
nowadays is the combination of signal processing (SP) and neural network.
For instance, Gaing (2004) used discrete wavelet transform (DWT) integrated
probabilistic neural network (PNN) to classify power disturbances. This method used
the multiresolution analysis (MRA) of DWT and Parseval’s theorem to extract energy
distribution features at different resolution levels. The features were subsequently
classified using PNN. Moreover, PNN is incorporated with wavelet transform (WT) to
determine the location and type of the fault (Othman, & Amari 2008). Realizing the
potential of these intelligent methods, G. Yin (2005) presented a combination of fast
Fourier transform (FFT) and artificial neural network (ANN) classifier for detecting
the islanding condition (G. Yin 2005). In this method, the output voltages of the
inverter are sampled and the signal frequency domain is obtained through FFT.
However, the algorithm is suitable for stationary waveforms and can be implemented