Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
WOKORO WOYENGIPREYE GEORGE
(PG/M.ENG/10/52889)
POWER SYSTEM RESTORATION USING ARTIFICIAL
NEURAL NETWORKS
ENGINEERING
A THESIS SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ENGINEERING, FACULTY OF ENGINEERING, UNIVERSITY OF NIGERIA NSUKKA
kkdkjd
IJEOMAH CLARA
Digitally Signed by: Content manager’s Name
DN : CN = Webmaster’s name
O= University of Nigeria, Nsukka
OU = Innovation Centre
SEPTEMBER 2012
POWER SYSTEM RESTORATION USING ARTIFICIAL NEURAL NETWORKS
BY
WOKORO WOYENGIPREYE GEORGE
(PG/M.ENG/10/52889)
A PROJECT SUBMITTED TO THE DEPARTMENT OF ELECTRICAL
ENGINEERING, FACULTY OF ENGINEERING,
UNIVERSITY OF NIGERIA, NSUKKA
NIGERIA
IN PARTIAL FULFILLMENT OF THE REQUIREMENT FOR THE AWARD OF
MASTERS DEGREE IN MASTERS OF ENGINEERING (M.ENG) IN ELECTRICAL
ENGINEERING (POWER SYSTEMS ENGINEERING)
SEPTEMBER 2012
APPROVAL PAGE
POWER SYSTEM RESTORATION USING ARTIFICIAL NEURAL NETWORKS
BY
WOKORO WOYENGIPREYE GEORGE
(PG/M.ENG/10/52889)
A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR
THE AWARD OF THE MASTER DEGREE (M.ENG) IN ELECTRICAL
ENGINEERING, UNIVERSITY OF NIGERIA, NSUKKA
SEPTEMBER 2012
WOKORO WOYENGIPREYE GEORGE Signature________________ Date_____
(Student)
VEN. PROF. T. C. MADUEME Signature_________________ Date______
(Project Supervisor)
Engr. Dr B.O. ANYAKA Signature_________________ Date_____
(Head of Department)
Engr. Prof. A. O. IBE Signature_________________ Date______
(External Examiner)
CERTIFICATION PAGE
WOKORO WOYENGIPREYE GEORGE, a postgraduate student in the Department of
Electrical Engineering and with Registration number PG/M.ENG/10/52889 has satisfactorily
completed the requirements for course and research work for the award of the Degree of Master
in Electrical Engineering.
The work embodied in dissertation is original and has been submitted either in part or in full for
any diploma or Degree of this University or another to the best of our knowledge.
_________________________ __________________________________
VEN. PROF. T. C. MADUEME WOKORO WOYENGIPREYE GEORGE
(Project Supervisor) (Student)
____________________
Engr. Dr. B.O. ANYAKA
(Head of Department)
DEDICATION
I, WOKORO WOYENGIPREYE GEORGE, dedicate this thesis first of all to the Almighty
God for abundant love and grace showered on me to see this time and work on this thesis for the
fulfillment of my program.
Also dedicate this thesis to my lovely parents Mr. & Mrs. Wokoro for their constant support in
prayers to enable me complete this program.
ACKNOWLEDGEMENT
God in his infinite protection, guidance and inspiration has made this work a reality and a
success. May his mercy endure forever! Amen
It is a great delight and passion that I wish to acknowledge the priceless assistance and
constructive criticisms given by my supervisor Ven. Prof. T. C. Madueme. He has not only
been my supervisor but also an outstanding and understanding father, uncle, brother and friend.
My profound appreciation goes to my wonderful mother Mrs. Yoreigha Charity Wokoro for
her moral and spiritual support; her prayers have really kept me going on the correct track. And
also my amiable and lovely father Mr. Konsin Wokoro for his love, words of encouragement
and prayers that kept me going on the right track as well. I am in short of words to express my
appreciation to my wonderful sisters, (elder and younger sisters), Mrs. Tombara Wokoro
Abalaba, Miss Layefa Wokoro, Miss Doubara Wokoro, Miss Tari Wokoro and my only
younger brother, Master Kuro Wokoro for their love and understanding praying for me through
school.
With great delight I appreciate the wonderful support of my friends Miss Oluchi Egbule, Miss
Uzoma Luisa Nnabuike, Miss Eloho Ireye, Mr. Andrew Ayabina, Mr. David Kiridi and
Kingsley Ondukakpor Oyakemeagbegha who are brothers from another mother to me and
their fairness in prayers about making my thesis a reality and my program at large.
Finally, to my colleagues, Engr. Gerald Diyoke, Engr. Bola Akuru, Engr. Bala Shuaibu
Ayegba, Engr. Nelson Mbah, and other too numerous to mention. I Wokoro Woyengipreye
George, remain sincerely grateful to you all for supporting me through this program and making
my stay in University of Nigeria Nsukka a memorable one.
ABSTRACT
This thesis focuses on power system restoration using artificial neural networks stream-lined to
just transmission lines by detecting, classifying and locating faults on electric power
transmission lines. Feed-forward networks have been employed along with back-propagation
algorithm for each of the three phases in the Fault location process. Analysis on neural networks
with varying number of hidden layers and neurons per hidden layer has been provided to validate
the choice of the neural networks in each step. Simulations were done in MATLAB 7.5 software
and results have been provided to demonstrate that artificial neural network based methods are
efficient in locating faults on transmission lines and achieve satisfactory performances compared
to other types of techniques used in detecting faults on electric power transmission lines. The
simulated result for fault detection found the best artificial neural network configuration of (6-
10-5-3-1) which gave 98% result after training the network and the result for fault classification
found the best artificial neural network configuration of (6-35-4) which gave 97% result after
training the network. While the result for fault location dealt with the design, development and
the implementation of the neural network based fault locators for each of the various types of
faults which are line to ground, double line to ground, line to line and three phase faults (L-G,
LL-G, LL and 3phase) and in these different types of faults its results are based on the
appropriate mean square error for each simulation. The result obtained for single line to ground
fault location which is seen to be satisfactory at the end of the training and testing process of the
artificial neural network is said to have the network configuration of 6-input neurons, 7- hidden
neurons, 1-output neuron (6-7-1), with an average error of 0.89% and can be used for the
purpose of single line to ground fault location. For line to line fault, it is seen that the network
configuration of 6-input neurons,10-hidden neurons,5-hidden neurons,1-output neuron (6-10-5-
1) with an average percentage error of 0.966% is said to be satisfactory and can be used for the
purpose of line to line fault location. In the case of double line to ground fault, it is seen that the
network configuration of 6-input neurons, 21- hidden neurons, 11- hidden neurons, 1-output (6-
21-11-1), with an average percentage error of 1.122% to be satisfactory and can be used for the
purpose of double line to ground fault location. While for the 3-phase fault, it is seen that the
network configuration of 6-input neurons, 6-hidden neurons, 21-hidden neurons, 16-hidden
neurons, 1-output neuron (6-6-21-16-1), with an average percentage error of 0.836% to be
satisfactory and can be used for the purpose of 3-phase fault location on transmission lines.
TABLE OF CONTENTS PAGE
COVER PAGE - - - - - - - - - i
TITLE PAGE - - - - - - - - - ii
CERTIFICATION PAGE - - - - - - - - iii
DEDICATION - - - - - - - - - iv
ACKNOWLEDGEMENT - - - - - - - - v
ABTRACT- - - - - - - - - - - vi
TABLE OF CONTENTS- - - - - - - - - vii
LIST OF TABLES - - - - - - - - - ix
LIST OF FIGURES - - - - - - - - x
CHAPTER ONE: INTRODUCTION
1.1 Background of the Study - - - - - - - - 1
1.2 Statement of Problem - - - - - - - - 2
1.3 Objective of the Study - - - - - - - - 2
1.4 Significance of the Study - - - - - - - - 2
1.5 Scope of the Study - - - - - - - - 2
1.6 Definition of Terms - - - - - - - - 3
CHAPTER TWO: LITERATURE REVIEW
2.1POWER SYSTEM OVERVIEW - - - - - - - 5
2.2REPRESENTATION OF POWER SYSTEMS - - - - - 6
2.2.1 Single Line Diagram of a Power System - - - - 6
2.2.2 Impedance Diagram Representation of a Power System - - 6
2.3 POWER SYSTEM RESTORATION - - - - - - 7
2.4 GOALS AND STEPS IN RESTORATION - - - - - 8
2.5 PROBLEMS IN RESTORATIONS - - - - - - 9
2.6 CONVENTIONAL RESTORATION TECHNIQUES - - - - 10
2.7 ARTIFICIAL NEURAL NETWORKS - - - - - - 11
2.7.1 Neural Network Design - - - - - - - 12
2.7.2 Neural Network Architecture - - - - - - 12
2.7.2.1 Determination of Network Topology - - - - 12
2.7.2.2 Determination of System Dynamics - - - - 12
2.7.3 Features of ANNs over Other Techniques - - - - 12
2.7.4 Advantages of ANNs - - - - - - - 13
2.7.5 Disadvantages of ANNs - - - - - - - 13
2.7.6 Learning Paradigms - - - - - - - 13
2.7.6.1 Supervised Learning - - - - - - 13
2.7.6.2 Unsupervised Learning - - - - - 13
2.7.6.3 Reinforcement Learning - - - - - 14
2.8 APPLICATION OF ANNs TO POWER SYSTEM RESTORATION - - 14
2.9ANN BASED RESTORATION SCHEME CASE STUDY OF ISLAND RESTORATION
SCHEMES (IRS). - - - - - - - - - 15
2.10 RESTORATION CONSTRAINTS - - - - - - 16
2.11 POWER SYSTEM RESTORATION CASE STUDIES - - - 17
2.12 POWER SYSTEM PROTECTION - - - - - - 17
2.13TRANSMISSION LINE FAULT LOCATION TECHNIQUES - - 18
2.13.1. Impedance Based Methods - - - - - - 18
2.13.2. Simple Reactance Method - - - - - - 18
2.13.3. Takagi Method - - - - - - - 19
2.13.4. Modified Takagi Method - - - - - - 20
2.13.5. Travelling Wave Based Methods - - - - - 20
2.13.6. Neural Networks Based Methods - - - - - 21
CHAPTER THREE: METHODOLOGY FOR RESEARCH
3.1 FAULTS IN POWER SYSTEM - - - - - - - 26
3.1.1 Single Line-to-Ground Faults - - - - - - 27
3.1.2 Line-To-Line Fault - - - - - - - 29
3.1.3 Double Line-To-Ground Fault - - - - - 30
3.1.4 Three Phase Fault - - - - - - - 31
3.2 MODELLING THE POWER TRANSMISSION LINE SYSTEM - - 32
3.3 OUTLINE OF THE PROPOSED SCHEME - - - - - 33
3.4 DATA PRE-PROCESSING - - - - - - - 34
3.5 OVERVIEW OF THE TRAINING PROCESS - - - - - 36
3.6 OVERVIEW OF THE TESTING PROCESS - - - - - 36
CHAPTER FOUR: EXPERIMENTAL RESULTS AND DISCUSSIONS
4.1 FAULT DETECTION - - - - - - - - 39
4.1.1 Training the Fault Detection Neural Network - - - - 39
4.1.2 Testing the Fault Detection Neural Network - - - - 41
4.2 FAULT CLASSIFICATION - - - - - - - 43
4.2.1 Training the Fault Classifier Neural Network - - - - 43
4.2.2 Testing the Fault Classifier Neural Network - - - - 48
4.3 FAULT LOCATION - - - - - - - - 51
4.3.1 Single Line – Ground Faults - - - - - - 51
4.3.1.1training the Neural Network for Single Line-Ground Fault Location 51
4.3.1.2 Testing the Neural Network for Single Line-Ground Fault Location 57
4.3.2 LINE-LINE FAULTS - - - - - - - - 59
4.3.2.1 Training the Neural Network for Line-Line Fault Location - - 59
4.3.2.2 Testing the Neural Network for Line-Line Fault Location - - 63
4.3.3 DOUBLE- LINE-GROUND FAULTS - - - - - - 66
4.3.3.1 Training the Neural Network for Double Line-Ground Fault Location- 66
4.3.3.2 Testing the Neural Network for Double Line-Ground Fault Location- 71
4.3.4 THREE PHASE FAULTS - - - - - - - 73
4.3.4.1 Training the Neural Network for Three Phase Fault Location - 73
4.3.4.2 Testing the Neural Network for Three Phase Fault Location - - 74
CHAPTER FIVE: CONCLUSIONS AND RECOMMENDATIONS
5.1 CONCLUSIONS - - - - - - - - 75
5.2 RECOMMENDATIONS - - - - - - - 76
REFERENCES
LIST OF TABLES
Table 3.1: Sample of Inputs to the neural network for various fault cases - - 35
Table 4.1: Fault classifier ANN outputs for various faults- - - - - 44
Table 4.2: Percentage errors as a function of fault distance and fault resistance for the ANN
chosen for single line - ground fault location- - - - - - 59
Table 4.3: Percentage errors as a function of fault distance and fault resistance for the ANN
chosen for line - line fault location- - - - - - - - 65
Table 4.4 Percentage errors as a function of fault distance and fault resistance for the ANN
chosen for double line - ground fault location- - - - - - 73
Table 4.5 Percentage errors as a function of fault distance and fault resistance for the
ANN chosen for three phase fault location- - - - - - - 80
LIST OF FIGURES
Fig 2.1: single line diagram of a power system - - - - - 6
Fig 2.2: impedance diagram representation of a power system - - - 6
Fig 2.3 : power system restoration goals - - - - - - 8
Fig.2.4: Island restoration scheme as a case study - - - - - 16
Fig 2.5: Faulted Transmission Line illustrating simple-reactance method - - 19
Fig 2.6: A single-phase circuit illustrating Takagi method - - - - 19
Fig 2.7: Illustration of travelling wave based Fault Location - - - - 20
Fig 3.1: single line to ground fault on an unloaded generator - - - 27
Fig 3.2: sequence network representing single line to ground fault on phase of an unloaded
generator - - - - - - - - - - 28
Fig 3.3: line-to-line fault on an unloaded generator - - - - - 29
Fig 3.4: sequence network representing Line-to-line fault on phases and of an unloaded
generator - - - - - - - - - - 30
Fig 3.5: double line to ground fault on an unloaded generator - - - 30
Fig 3.6: sequence network representing double line to ground fault on phases of an
unloaded generator - - - - - - - - - 31
Fig 3.7: 3-phase fault - - - - - - - - - 31
Fig 3.8: three phase positive sequence network of an unloaded generator - - 32
Fig 3.9: online diagram of the studied power system - - - - - 32
Fig 3.10: a snap shot of the studied power system in simpowersystems - - 32
Fig 3.11: flow chart showing the outline of the proposed scheme - - - 33
Fig 3.12: Data pre-processing illustration - - - - - - 34
Fig 4.1: Mean-square error performance of the network (6-10-1) - - - 39
Fig 4.2: Mean-square error performance of the network (6-10-5-1)- - - - 40
Fig 4.3: Mean-square error performance of the network (6-10-5-3-1)- - - 40
Fig 4.4: Regression fit of the outputs vs. targets for the network (6-10-5-3-1)- - 41
Fig 4.5: Confusion matrices for Training, Testing and Validation Phases - - 42
Fig 4.6: Overview of the ANN (6-10-5-3-1) chosen for fault detection - - 42
Fig 4.7: Chosen ANN for Fault Detection (6 – 10 – 5 – 3 – 1)- - - - 43
Fig 4.8: Mean-square error performance of the network with configuration (6-5-5-31-4) 45
Fig 4.9: Mean-square error performance of the network with configuration (6-5-31-4) 45
Fig 4.10: Mean-square error performance of the network with configuration (6-5-4)- 46
Fig 4.11: Mean-square error performance of the network with configuration (6-10-4)- 46
Fig 4.12: Mean-square error performance of the network with configuration (6-20-4) 47
Fig 4.13: Mean-square error performance of the network with configuration (6-35-4) 47
Fig 4.14: Regression fit of the Outputs vs. Targets of ANN with configuration (6-35-4) 48
Fig 4.15: Gradient and Validation performance of the ANN with configuration (6-35-4) 49
Fig 4.16: Overview of the ANN with configuration (6-35-4), chosen as fault classifier 50
Fig 4.17: Chosen ANN for Fault Classification (6 – 35 – 4) - - - - 50
Fig 4.18: Regression fit of the Outputs vs. Targets with configuration (6-5-5-1) - 52
Fig 4.19: Test Phase performance of the Neural Network with configuration (6-5-5-1) 52
Fig 4.20: Regression fit of the outputs versus targets with configuration (6-25-1) - 53
Fig 4.21: Test phase performance of the ANN with configuration (6-25-1) - - 53
Fig 4.22: Regression fit of the outputs versus targets with configuration (6-16-1) - 54
Fig 4.23: Test phase performance of the neural network with configuration (6-16-1) 54
Fig 4.24: Regression fit of the outputs versus targets with configuration (6-7-1) - 55
Fig 4.25: Test phase performance of the ANN with configuration (6-7-1) - - 55
Fig 4.26: Overview of the chosen ANN with configuration (6-7-1) - - - 56
Fig 4.27: Mean-square error performance of the network with configuration (6-7-1) 56
Fig 4.28: Gradient and validation performance of the network with configuration (6-7-1) 57
Fig 4.29: Regression plots of various phases of learning ANN with configuration (6-7-1) 58
Fig 4.30: Structure of the chosen ANN with configuration (6-7-1) - - - 58
Fig 4.31: Mean Square Error performance plot with configuration (6-10-20-5-1) - 60
Fig 4.32: Test Phase performance of the ANN with configuration (6-10-20-5-1) - 60
Fig 4.33: Mean Square Error performance plot with configuration (6-10-1) - - 61
Figure 4.34: Test Phase performance of the ANN with configuration (6-10-1) - 61
Fig 4.35: Mean Square Error performance of the ANN with configuration (6-10-5-1) 62
Fig 4.36: Test phase performance of the neural network with configuration (6-10-5-1) 62
Fig 4.37: Overview of the chosen ANN for Line-Line Faults (6-10-5-1) - - 63
Fig 4.38: Regression fit of the outputs versus targets with configuration (6-10-5-1)- 63
Figure 4.39: Gradient and validation performance plot of the ANN (6-10-5-1) - 64
Fig 4.40 Regression plots of the various phases of learning of the chosen ANN (6-10-5-1) 64
Fig 4.41: Structure of the chosen Neural Network (6 – 10 – 5 – 1) - - - 65
Fig 4.42: Mean Square Error performance of the ANN with configuration (6-10-1) 66
Fig 4.43: Test Phase performance of the ANN with configuration (6-10-1) - - 67
Fig 4.44: Mean Square Error performance of the ANN with configuration (6-20-1) 67
Fig 4.45: Test Phase performance of the ANN with configuration (6-20-1) - - 68
Fig 4.46: MSE performance of neural network with configuration (6-10-5-1) - 68
Fig 4.47: Test Phase performance of the ANN (6-10-5-1) - - - - 69
Fig 4.48: MSE performance of the neural network with configuration (6-21-11-1) - 69
Fig 4.49: Test phase performance of the ANN (6-21-11-1) - - - - 70
Fig 4.50: Overview of the chosen ANN (6-21-11-1) for Double Line-Ground Faults 70
Fig 4.51: Regression fit of the outputs versus targets with configuration (6-21-11-1) 71
Fig 4.52: Gradient&validation performance plot of ANN with configuration (6-21-11-1) 71
Fig 4.53: Regression plots of the various stages of learning of ANN (6-21-11-1) - 72
Fig 4.54: Structure of the chosen ANN (6 – 21 – 11 – 1) - - - - 72
Fig 4.55: Regression fit of the outputs vs targets of ANN with configuration (6-21-10-1) 74
Fig 4.56: MSE performance of the neural network with configuration (6-21-10-1) - 75
Fig 4.57: Test Phase performance of the ANN with configuration (6-21-10-1) - 75
Fig 4.58: MSE performance of the neural network with configuration (6-21-1) - 76
Fig 4.59: Regression fit for the outputs vs targets of ANN with configuration (6-21-1) 76
Fig 4.60: Test Phase performance of the ANN with configuration (6-21-1) - - 77
Fig 4.61: Regression fit of the outputs versus targets of ANN (6-6-21-16-1) - 77
Fig 4.62: Test Phase performance of the ANN (6-6-21-16-1) - - - 78
Fig 4.63: Overview of the chosen neural network for three phase fault location - 78
Fig 4.64: Mean Square Error performance of the neural network (6-6-21-16-1) - 79
Fig 4.65: Gradient and validation performance plots of the ANN (6-6-21-16-1) - 79
Fig 4.66: Regression plots of the various phases of learning of the ANN (6-6-21-16-1) 79
Fig 4.67: Structure of the chosen ANN (6 – 6 – 21 – 16 – 1) - - - - 80
CHAPTER ONE
INTRODUCTION
1.1 BACKGROUND OF THE STUDY
In the past several decades, there has been a rapid growth in the power grid all over the world
which eventually led to the installation of a huge number of new transmission and distribution
lines. Moreover, the introduction of new marketing concepts such as deregulation has increased
the need for reliable and uninterrupted supply of electric power to the end users who are very
sensitive to power outages [1].
One of the most important factors that hinder the continuous supply of electricity and power is a
fault in the power system [2]. Any abnormal flow of current in a power system’s components can
lead to a fault in the power system. These faults cannot be completely avoided since a portion of
these faults also occur due to natural reasons which are always beyond the control of mankind.
Hence, it is very important to have a well-coordinated protection system that detects any kind of
abnormal flow of current in the power system, identifies the type of fault and then accurately
locates the position of the fault in the power system. The faults are usually taken care of by
devices that detect the occurrence of a fault and eventually isolate the faulted section from the
rest of the power system.
Hence some of the important challenges for the incessant supply of power are detection,
classification and location of faults [3]. Faults can be of various types namely transient,
persistent, symmetric or asymmetric faults and the fault detection process for each of these faults
is distinctly unique in the sense, there is no one universal fault location technique for all these
kinds of faults.
The High Voltage Transmission Lines (that transmit the power generated at the generating plant
to the high voltage substations) are more prone to the occurrence of a fault than the local
distribution lines (that transmit the power from the substation to the commercial and residential
customers) because there is no insulation around the transmission line cables unlike the
distribution lines. The reason for the occurrence of a fault on a transmission line can be due to
several reasons such as a momentary tree contact, a bird or an animal contact or due to other
natural reasons such as thunderstorms or lightning. Most of the research done in the field of
protective relaying of power systems concentrates on transmission line fault protection due to the
fact that transmission lines are relatively very long and can run through various geographical
terrain and hence it can take anything from a few minutes to several hours to physically check
the line for faults [4].
The automatic location of faults can greatly enhance the systems reliability because the faster we
restore power, the more money and valuable time we save. Hence, many utilities are
implementing fault locating devices in their power quality monitoring systems that are equipped
with Global Information Systems for easy location of these faults. Fault location techniques can
be broadly classified into the following categories
[5]:
Impedance measurement based methods
Travelling-wave phenomenon based methods
High-frequency components of currents and voltages generated by faults based methods
Intelligence based methods
From quite a few years, intelligent based methods are being used in the process of fault detection
and location. Three major artificial intelligence based techniques that have been widely used in
the power and automation industry are [6]:
Expert System Techniques
Artificial Neural Networks
Fuzzy Logic Systems
Among these available techniques, Artificial Neural Networks (ANN) has been used extensively
in this thesis for fault location on electric power transmission lines. These ANN based methods
do not require a knowledge base for the location of faults unlike the other artificial intelligence
based methods [7].
Therefore the application of artificial neural network to power system restoration is to make sure
of a steady supply of electric power and fault diagnosis in power systems because the importance
of electricity in our day to day life has reached such a stage that it is very necessary to protect the
power system equipment from damage and to ensure maximum continuity of power supply.
1.2 STATEMENT OF PROBLEM
Power system blackout is a major problem we face in the country. When they occur, the effects
on commerce, industry and everyday life of the general population can be quite severe. Since it is
a major part of any successful economic system and development at large, it is important to
reduce the economic and social cost of any power system blackout.
1.3 OBJECTIVE OF THE STUDY
The goal of this thesis is to propose an integrated method to perform each of these tasks using
artificial neural networks. A back-propagation based neural network has been used for the
purpose of fault detection and another similar one for the purpose of fault classification in
transmission lines. To achieve this, we need to design, develop, test and implement a complete
strategy for the fault diagnosis in order to restore transmission lines back to service. The first
step in the process is fault detection. Once we know that a fault has occurred on the transmission
line, the next step is to classify the fault into one of the different categories based on the phases
that are faulted. Then, the third step is to pin-point the position of the fault on the transmission
line.
1.4 SIGNIFICANCE OF THE STUDY
With respect to the objectives of this thesis, it will benefit the Power Holding Company of
Nigeria in aspect of effective fault location and restoration of power system transmission lines.
1.5 SCOPE OF THE STUDY
The bulk of this work is centered on transmission lines of the Nigerian power system.
1.6 DEFINITION OF TERMS:
NEURONS: These are large number of simple processing units.
DISTURBANCE: A disturbance is a change in the operating condition or an operating
parameter of the system.
WEIGHTS: these are the values which are multiplied to each neuron through the process of
giving a desired output
CAPACITY: It is related to the amount of information that can be stored in the network and to
the notion of complexity.
TRANSMISSION LINES: a transmission line is the material medium or structure that forms all
or part of a path from one place to another for directing the transmission of energy, such as
electromagnetic waves or acoustic waves, as well as electric power transmission. And its
components are wires, coaxial cables, dielectric slabs, optical fibers, electric power lines.
STEADY STATE STABILITY: this is the ability of systems to maintain synchronism among
its various generating stations during the desired range of system loading when there is no
periodic disturbance in the system.
TRANSIENT STABILITY: the transient stability limit is always below the steady state
stability limit because it may have many different values depending upon the nature and the
magnitude of the disturbance.
LEARNING: is using a set of observations to find a function which solves the task in some optimal sense. Or Learning involves adjustment of synaptic connections that exist between
neurons.
COST: It is frequently defined as a statistic to which only approximations can be made.
SYNAPSE: is a connection between two nerve cells.
RESTORATION: to return a thing back to its normal state or condition. In this context of
power system restoration, is returning power back to the system after a faulty disturbance.
CHAPTER TWO
LITERATURE REVIEW
This chapter gives a literature survey that provides an overview of the relevant areas in
restoration and artificial neural networks. It is subdivided into thirteen different topics. The first
topic provides a general overview of the power system and representation of power system. It is
followed by the general overview of restoration process, goals and steps in restoration. Then
followed by problems in restoration and conventional restoration techniques. This is then
followed by artificial neural networks, the application of artificial neural networks to power
system restoration, artificial neural network based restoration scheme using a case study of
Island restoration scheme, restoration constraints, power system restoration case studies. Then
followed by power system protection stream lining it to transmission lines by learning several
techniques used to locate faults and restoring lines back to service.
2.1 POWER SYSTEM OVERVIEW
Electric power systems may be of great complexity and spread over large geographical area. An
electric power system consists of generators, transformers, transmission lines and consumer
equipment (loads). The majority of these systems rely upon three-phase AC power - the standard
for large-scale power transmission and distribution across the modern world. Specialized power
systems that do not always rely upon three-phase AC power are found in aircraft, electric rail
systems, ocean liners and automobiles. The system must be protected against flow of heavy
short-circuit currents which can cause permanent damage to major equipment by disconnecting
the faulty section of system by means of circuit breakers and protective relaying.
It is necessary to know the maximum short-circuit currents that can occur at the different points
of a system in order that circuit breakers may be selected that are adequate to withstand the
currents and operate successfully to cut out the faulty section, and also in order that the
protective relays may be selected for correct operation. The design of machines, bus-bars,
isolators, circuit breakers etc, is based on the consideration of normal and short-circuits currents.
It is also necessary to be able to calculate, approximately at least, the size of the protective
reactors which must be inserted in the system to limit the short-circuit currents to a value which
is not beyond that capable of being withstood by the circuit breakers.
The short-circuit currents in an AC system are determined mainly by the reactance of the
alternators, transformers and lines up to the point of the fault in the case of phase to phase faults.
When the fault is between phase and earth, the resistance of the earth path plays an important
role in limiting the currents.
In the case of circuit breakers, their rupturing capacities are based on the symmetrical short-
circuit current which is the most simple calculation among all types of short-circuits. However,
for determination of settings of relays it is absolutely necessary to know fault current due to
unsymmetrical fault condition too for which knowledge of symmetrical components etc. is
required [8].
2.2 REPRESENTATION OF POWER SYSTEMS
A complete diagram of power system representing all the three phases becomes too complicated
and cumbersome for a system of practical size, so much so that it may no longer convey the
information it is intended to convey. It is much more practical to represent a power system by
means of simple symbols for each component resulting in what is called a single line diagram.
2.2.1 Single Line Diagram. The single line diagram of a power system network shows the
main connections and arrangements of the system components along with their data such as
output rating, voltage, resistance and reactance etc. in case of transmission lines sometimes the
conductor size and spacing are given. It is not necessary to show all the components of the
system on a single line diagram, e.g., circuit breakers need to be shown in a load flow study but
are must for a protection study. In a single line diagram, the system components are usually
drawn in the form of their symbols. Generators and transformer connections-star, delta and
neutral earthing are indicated by symbols drawn by the side of the representation of these
elements. Circuit breakers are represented by rectangular blocks. Fig 2.1 represents the single
line diagram of a typical power system. The ratings of generator, motor and transformers are
given below the diagram [8].
Fig 2.1: single line diagram of a power system.
2.2.2 Impedance Diagram Representation of a Power System. A further simplification from
the single line diagram with its symbols for the various components is to draw the diagram with
impedances only. The impedance diagram of the power system of fig 2.1 is shown in fig 2.2. In
impedance diagram, each component is represented by its equivalent circuit, e.g., the
synchronous generator at the generating station by a voltage source in series with a resistance
and reactance, the transformer by its equivalent circuit and the transmission line by nominal -
equivalent circuit. Loads are assumed to be passive not involving rotating machines and are
represented by resistance and inductive reactance in series. Neutral earthing impedances do not
appear in the diagram as balanced conditions are assumed [8].
Fig 2.2 impedance diagram representation of a power system
The impedance diagram shown in fig 2.2 is known as positive sequence diagram since it is
drawn for a balanced 3-phase system.
2.3 POWER SYSTEM RESTORATION
Since late 1960s, power system industries have undertaken considerable effort to develop and
implement preventive and corrective measures to reduce the possibility and extent
of system outage. However, relatively little effort has gone into planning system restoration for
minimizing the duration of an outage, should one occur [9]. It is always felt that system
restoration could significantly be improved by combined effort of system analysts, operating
personal and the concurrent use of on-line and off-line computer facilities at the operating
center [10]. Widespread black-outs are rare but when they occur, they pose severe and unusual
problems. Sectionalization of power system into islands are proposed to arrest
total system collapse [11], [12] and [13]. Creation of islands using strategically placed
under/over frequency relays, allows simultaneous restoration of the islands, resulting
speedy restoration of the system [14]. Immediate re-starting of generators is always given
propriety during restoration of a power system [15] and [16]. A delay in re-starting of
thermal power plant increases re-starts time rapidly [17]. Therefore, supply of cranking power to
non-black start thermal power plant should always be taken as a primary task during power
system restoration. Utilization of the line power has also been proposed to crank non-black start
thermal units for quick re-energization [18].
Restoration steps must provide continuity between the target system state and the present state of
the system. Also, restoration strategies must allow for change in events
during restoration process. The topology and system operating condition keep changing
frequently due to energization of line(s), energization of generator(s) and picking-up of load(s)
during the course of power system restoration. Thus, preparation of a successful and
effective power system restoration plan depends upon the collective use of appropriate power
system mathematical models and algorithm which provide fast and accurate solution to the
problems encountered at different stages of restoration process.
The restoration of a power system can be simplified significantly, if it is viewed as ‘event-
oriented’, i.e. target event that is to be achieved. Therefore, general restoration strategies can be
used in a co-ordinated manner to develop restoration plan with respect to a
target restoration event. Each planned restoration event requires validation before qualification
for implementation. The post-restoration system is not necessarily identical to the pre-
disturbance system as some equipment may not be available for the timely restoration (because
of fault) and/or change in load demand.
Each restoration case is highly complex and unique, which makes it difficult to develop
systematic restoration procedures, and to find methods that can be generalized in order to be
applied to other systems and scenarios. However, restoration cases can be considered as a
succession of simple restoration steps that are common in all restoration procedures, as described
in more detail in [19].
The same principle is true when the feasibility of restoration steps needs to be assessed: the
investigation of an aggregation of complex phenomena can be simplified by defining different
problem areas that are common in all restoration scenarios and by developing rules that address
specific problems during restoration. In this thesis project we focus mainly on problems that are
related to the initial phases of a restoration process. However, the same principles can be
extended to include other problem areas, such as the analysis of the reintegration of subsystems.
2.4 GOALS AND STEPS IN RESTORATION Even though each power blackout and restoration scenario is a unique event, there are certain
goals and steps that are common in all restoration procedures. They involve almost all aspects of
power system operation and planning. Each restoration procedure that follows a complete or
partial blackout of a power system can be subdivided into the following steps [20];
1. Determination of System Status: In this stage, the boundaries of energized areas are
identified, and frequencies and voltages within these areas are assessed. Furthermore, in
cases where no connections to neighboring systems exist, black start (or cranking)
sources are identified in each subsystem and critical loads are located.
System restoration
Tie line
utilization
Path
management
Stability
Inspection
MVAR
management
MW
managementLoad
management
Switching sequence
Inspection
Static MVAR
InspectionNon-Black start
Unit operation
Black start
Unit operation
Man-power
dispatchFuel inspection
Mechanical
inspection
Capacity
inspection
Fault
inspection
Transient
Voltage
inspection
Fig 2.3 power system restoration goals
2. Black Start of Large Thermal Power Plants: Large thermal power plants have to be
restarted within a certain period of time. For example, hot restart of drum type boilers is
only possible within thirty minutes. If it cannot be accomplished and the boiler is not
available for four to six hours, a cold restart has then to be performed. Thermal power
plants can be restarted by means of smaller units with black start capability, i. e. power
plants that can be started and brought online without external help and within a short
period of time. Power plants with black start capability are hydro, gas, or diesel power
plants. After such a power plant has been brought to full operation, a high voltage path to
a large thermal power station is built and the thermal unit's auxiliaries which are driven
by large induction motors are started. Along the path, “balanced" loads have to be
supplied to maintain the voltage profile within acceptable limits and to prepare a load
base for the thermal units. Additional smaller units can also be brought online through the
path to improve system stability.
3. Energization of Subsystems: In case of a large power blackout, it is advantageous in most
cases to section the power system into subsystems in order to allow parallel restoration of
islands, and to reduce the overall restoration time. Within each subsystem, starting from a
large thermal power station, the skeleton of the bulk power system is energized. Paths to
other power plants and to the major load centers are built, and loads are energized to form
up the transmission system. At the end of this step, the network has sufficient power and
stability to withstand transients as a result of further load pick-up and addition of large
generating units.
4. Interconnection of Subsystems: In this stage of the power system restoration process, the
subsystems are interconnected. Eventually, remaining loads are picked up and the system
performs its transition to the alarm or normal state.
Among the above four general steps of the restoration process, the second and third step are the
most critical ones. Mistakes in these stages can lead to unwanted tripping of generators and load
shedding due to extensive frequency and voltage deviations, and consequently to a recurrence of
the system outage. Because of time-critical boiler-turbine start-up characteristics and possible
further equipment damage, extensively prolonged restoration times may occur, resulting in a
much higher impact on the public and industry, and an increased damage to the economy [20].
2.5 PROBLEMS IN RESTORATION
During power system restoration, a multitude of different phenomena and abnormal conditions
may occur. The problems encountered during restoration can be subdivided into three general
areas;
1. Active Power Balance and Frequency Response: During the restoration process, two
different aspects of this type of problem can be identified. The first one is the black start
of large thermal power plants, where large auxiliary motor loads are picked up, using
relatively small hydro generators, diesel, or gas turbines. This can result in large
frequency excursions and consequently in an activation of under-frequency load shedding
relays and, in the worst case, in the loss of already restored load and a recurrence of the
blackout. Due to the importance of this problem, it is defined as a problem area in its
own, as discussed further below. The second aspect is the pick-up of cold loads. When
the network is extended, power plants are added to the generation, and loads are picked
up, it is necessary to preserve a balance between active load and generation. In the case
this balance is disturbed, frequency deviations result. If these are extensive, an unwanted
activation of load shedding schemes can occur, and newly connected loads can be lost
again. In the worst case, the frequency decline may reach levels that can lead to the
tripping of steam turbine generating units as a consequence of the operation of under
frequency protective relays. This is due to the fact that the operation of steam turbines
below a frequency of 58.8 Hz is severely restricted as a result of vibratory stress on the
long low-pressure turbine blades. Thus, in order to keep the frequency deviations within
allowable limits; the load increments should not exceed a certain level. However, if the
load increments are too small, the overall restoration time will be unnecessarily
prolonged.
2. Reactive Power Balance and Voltage Response: Analogous to the active power balance it
is necessary to maintain a balance in reactive power. High charging currents, originating
from lightly loaded transmission lines, can lead to the violation of generator reactive
capability limits and to the occurrence of sustained (power frequency) over voltages.
These may cause under excitation, self-excitation, and instability. Sustained over voltages
can also cause the over excitation of transformers and the generation of harmonic
distortions. Transient over voltages are a consequence of switching operations on long
transmission lines, or of switching of capacitive devices, and may result in arrester
failures. Harmonic resonance over voltages is a result of system resonance frequencies
close to multiples of the fundamental frequency in combination with the injection of
harmonics, mainly caused by transformer switching. They may lead to long-lasting over
voltages, resulting in arrester failures and system faults. Due to the small amount of load
connected to the system, especially at the beginning of a restoration process, the voltage
oscillations are lightly damped and can last for a long time, reaching very high
amplitudes. This effect can be aggravated by transformer over excitation as a result of
sustained over voltages, and power electronics.
3. Auxiliary Systems. The auxiliaries of large thermal power plants are essentially large
induction motors driving pumps and fans. When they are energized during black or
emergency starts, several problems can occur. The high reactive currents that are drawn
during start-up may lead to voltage depressions which can result in overheating and
permanent damage of machine windings [20].
2.6 CONVENTIONAL RESTORATION TECHNIQUES
In recent years, there have been a number of approaches that propose new restoration techniques
as alternatives to these commonly used restoration procedures. While these techniques may vary
in implementation detail, it can be said that three main principles for power system restoration
(PSR) have been proposed [21].
a) Automated Restoration: In this restoration technique, computer programs are
responsible for the PSR plan development and implementation. The PSR techniques
based on this principle acquire system data from the supervisory control and data
acquisition system (SCADA) and the energy management system (EMS). Under a wide
area disturbance, a PSR program installed in the EMS system will use the acquired
system data to develop a restoration plan for the transmission system. After developing
the restoration plan, a switching sequence program, which is also a part of the EMS, will
be responsible for the transmission of control signals through SCADA to circuit breakers
and switches to implement the plan. In this technique, the system operator plays the role
of a supervisor.
b) Computer Aided Restoration: In this technique, the PSR plan development and
implementation is performed by the system operator. The PSR techniques that use this
principle also acquire system data from the local SCADA/EMS. Following a wide area
disturbance, the system operator uses power system data provided by the SCADA/EMS
to develop a PSR plan. The system operator can use the PSR procedures and power
system analysis programs as aids to develop the restoration plan. The system operator
will also use the local SCADA/EMS to transmit control commands to circuit breakers
and switches in order to implement the chosen PSR plan.
c) Cooperative Restoration: In this technique, a computer program installed at the EMS
will propose a PSR plan after the occurrence of a blackout. The system operator is
responsible for the implementation of the PSR plan. The PSR systems that apply this
technique also use power system data obtained from local SCADA/EMS. When the
power system is undergoing a wide area disturbance, the PSR program installed in the
EMS will use the system data to develop a restoration plan. With this restoration plan, the
system operator can send controlling signals through local SCADA/EMS to circuit
breakers and switches to implement the plan [21].
2.7 ARTIFICIAL NEURAL NETWORK
An artificial neural network (ANN), usually called neural network (NN) is a mathematical model
or computational model that is inspired by the structure and/or functional aspects of biological
neural networks. A neural network consists of an interconnected group of artificial neurons and it
processes information using a connectionist approach to computation. It is an adaptive system in
most cases which changes its structure based on external or internal information that flows
through the network during the learning phase [22]. It is a system closely modeled on human
brain and tries to obtain a performance similar to that of human performance while solving
problems.
Breaking down what it is made up of, it is seen as a computational system made of large number
of simple and highly connected processing elements which process information by its dynamic
state response to external inputs. Computational elements in ANN are non-linear and so the
result that comes out through non-linearity can be more accurate than other methods of
computation. These non-linear computational elements will be working in unison to solve
specific problems. It is actually configured for specific applications such as data classification or
pattern recognition through a learning process.
In order words, one type of network sees the nodes as “artificial neurons” and this is called
artificial neural network (ANN). An artificial neuron is an emulation of the biological nervous
system. It is inspired in the natural neurons which receive signals through the synapse located on
the dendrites or membrane of the neuron. When signals received are strong enough to surpass a
certain threshold, the neuron is activated and emits a signal through the axon. This signal might
be sent to another synapse and might activate other neurons.
The complexity of real neurons is highly abstracted when modeling artificial neurons. These
basically consist of inputs which are like synapses, which are multiplied by weights which are
the strength of the respective signals and then computed by a mathematical function which
determines the activation of the neuron. Another function which may be the identity computes
the output of the artificial neuron, sometimes in dependence of a certain threshold. ANNs
combine artificial neurons in order to process information.
The higher the weight of an artificial neuron is, the stronger the input which is multiplied by it
will be. But note that weights can also be negative, so we can say the signal is inhibited by the
negative weight. And by adjusting the weights of an artificial neuron we can obtain the desired
output of a specific input. This adjustment is done by algorithms designed to handle large
number of inputs in the network because it will be difficult to calculate weights of hundreds of
thousands by hand in a particular network to get a desired output. This process of adjusting the
weight is called learning or training.
Note that the various input to the network are represented by the mathematical symbol, X(n).
Each of their inputs are multiplied by a connection weight, these weights are represented by
W(n). In the simplest case, these products are simply summed, fed through a transfer function to
generate a result and then the output. Even though all artificial neural networks are constructed
from these basic building blocks, the fundamentals may vary in these building blocks and there
are differences [23].
2.7.1 Neural Network Design
A neural network element is the smallest processing unit of the whole network essentially
forming a weighted sum and transforming it by the activation function to obtain the output. In
order to gain sufficient computing power, several neurons are interconnected together. The
manner in which actually the neurons are connected together depends on the different classes of
the neural networks. Basically neurons are arranged in layers. ANNs have parallel distributed
architecture with a large number of nodes and connections [24].
2.7.2 ANN Architecture
Construction of neural Network involves the following tasks.
1. Determination of network topology
2. Determination of system (activation & synaptic) dynamics
2.7.2.1 Determination of the Network Topology
The topology of the neural network refers to its framework as well as its interconnection scheme.
The number of layers and the number of nodes per layer often specify the framework. The types
of layer include:
Input Layer, where the nodes are called input units, which do not process information but
distribute information to other units.
Hidden Layer(s), where the nodes are called hidden units, which are not directly observable.
They provide into the networks the capability to map or classify nonlinear problems.
The Output Layer, where the nodes are called output units, which encode possible concepts (or
values) to be assigned to the instance under consideration. For example each output unit
represents a class of objects. Other main important concept is the weightage for the connected
unit. It can be real or integer numbers. They can be confined to a range and are adjustable during
network training. When training is completed, all of them attain fixed values.
2.7.2.2 Determination of Systems (Activation & Synaptic) Dynamics
The dynamics of the network determines its operation. ANN’s can be trainable nonlinear
dynamical systems. Neural dynamics consists of two parts one which corresponding to the
dynamics of activation states and the other corresponding to the dynamics of synaptic weights.
The activation dynamics determines the time evolution of the neural activation’s. Synaptic
activation determines the change in the synaptic weights. The synaptic weights form Long Term
Memory (LTM) whereas the activation's state forms Short Term Memory (STM) of the network.
Synaptic weights change gradually, whereas the neuron's activation's fluctuate rapidly.
Therefore, while computing the activation dynamics, the system weights are assumed to be
constant. The synaptic dynamics dictates the learning process [24].
2.7.3 Features of Artificial Neural Network over Other Techniques There are several attractive features of artificial neural network over other techniques and they
are mentioned below.
Their ability to represent non-linear relations makes them well suited for non-linear
modeling in control systems.
The adaptation ability and learning of artificial neural network in uncertain system
through off-line and on-line weight adaptation is highly remarkable.
Parallel processing architecture allows fast processing for large scale dynamic systems.
Neural network can handle large number of inputs and can have many outputs
Artificial neural networks can store knowledge in a distributed fashion and consequently
have a high fault tolerance.
2.7.4 Advantages of Artificial Neural Networks
A neural network can perform task that a linear program cannot.
When an element of the neural network fails, it can continue without any problem by
their parallel nature.
A neural network learns and does not need to be reprogrammed
It can be implemented in any application
It can be implemented without any problem.
2.7.5 Disadvantages of Artificial Neural Networks
The neural network needs training to operate.
The architecture of a neural network is different from the architecture of microprocessors
therefore needs to be emulated.
Requires high processing time for large neural networks.
Another aspect of artificial neural network is that, there are different architectures, which
consequently requires different types of algorithms but despite the apparent complex system, a
neural network is relatively simple. They are among the newest technologies nowadays. The
field of work is very interdisciplinary, but the explanation here is restricted to engineering only.
In the world of engineering, neural networks have two main functions which are pattern
classifiers and as non-linear adaptive filters. As its biological predecessor, an artificial neural
network is an adaptive system. By adaptive, it means that each parameter is changed during its
operation and it’s deployed for solving the problem in matter [25].
2.7.6 LEARNING PARADIGMS
There are three major learning paradigms, each corresponding to a particular abstract learning
task. These are supervised learning, unsupervised learning and reinforcement learning.
2.7.6.1 Supervised Learning
Supervised learning is the process that incorporates an external guidance. In the supervised
learning, a training pair consists of an input vector and a desired target vector. The difference
constitutes an error that is used to modify network weights in a manner that reduces the error in
subsequent training cycles. These techniques include deciding, when to turn off the learning,
how long and how often to present each association for training and supplying performance error
information. Supervised learning is further classified as Structural learning / temporal learning.
Structural learning encodes the proper auto associate (single pattern vector) or heteroassociate
vector of patterns pair mapping into weight matrix W. Temporal learning encodes a sequence of
patterns necessary to achieve final outcome.
2.7.6.2 Unsupervised Learning
In the Unsupervised learning no target vector exists. The input vector is applied to the network
and the system “self organizes” so that a consistent output (possibly unpredicted before training)
is produced. During the training phase the weights of ANN stabilize and while testing for an
unknown pattern gives the output without a time-delay of learning phase. The recall or testing
depends on the interconnection of the network. In feedforward network, the network provides
output in just one pass and allows flow of signal in only one direction from input to hidden and
to output layers. In feedback network, signals can flow amongst neurons in either direction and
/or recursively. Some of the most popularly used rules for learning includes Hebb's rule and
Delta rule for single layer (perception) ANN, Backpropagation algorithm for multilayer
(perception) ANN.
Thus its architecture, its processing algorithm and its learning algorithm characterize a neural
network. The architecture specifies the way the neurons are connected. The processing algorithm
specifies how the neural network with a given set of weights calculates the output vector for any
input vector. The learning algorithm specifies how the network adapts its weights for all given
vectors.
2.7.6.3 Reinforcement Learning
In reinforcement learning, data are usually not given, but generated by an agent's interactions
with the environment. At each point in time, , the agent performs an action and the
environment generates an observation and an instantaneous cost , according to some (usually
unknown) dynamics. The aim is to discover a policy for selecting actions that minimizes some
measure of a long-term cost; i.e., the expected cumulative cost. The environment's dynamics and
the long-term cost for each policy are usually unknown, but can be estimated.
More formally, the environment is modeled as a Markov decision process (MDP) with states and
actions with the following probability distributions: the instantaneous cost distribution ,
the observation distribution and the transition , while a policy is defined as
conditional distribution over actions given the observations. Taken together, the two define a
Markov chain (MC). The aim is to discover the policy that minimizes the cost; i.e., the MC for
which the cost is minimal. ANNs are frequently used in reinforcement learning as part of the
overall algorithm.
2.8 APPLICATION OF ANNs TO POWER SYSTEM RESTORATION
In order to reduce the economic and social costs of a blackout, the majority of electric utility
companies have pre-established guidelines and operating procedures to restore the power system.
These guidelines and operating procedures contain sequential restoration steps that an operator
should follow in order to restore the system. However, the highly stressful conditions
encountered in the aftermath of a blackout together with the fact that these guidelines are based
on assumed system conditions which may not be present, diminishes the success rate (defined as
that fraction of restoration attempts which does not result in unacceptable loading and voltage
profiles, or in breaker operations by the protection systems) of the technique. The main reason
for unsuccessful restoration attempts based on this technique is that the prevailing conditions of
the power system can differ significantly from the assumed conditions when the restoration plan
was developed.
Recent publications indicate high interest on the part of power utilities in the Cooperative
restoration principle. The most commonly used techniques proposed in the above-mentioned
publications are the rule-based expert systems and the mathematical programming approach.
Both of these techniques have produced very good results; however, few implementations of
these techniques exist at this time. One interesting limitation of these techniques is the time
required to find the restoration plan. The rule-based techniques can take several minutes to find
the plan in large transmission systems, mainly because the number of rules is proportional to the
size of the system. The mathematical programming approach has similar performance
characteristics. This technique considers the system to be in a state space where a search must be
conducted in order to find the final configuration of the restored system.
The system configuration is represented by a vector containing the breaker status. Several search
algorithms can be implemented to find this final system configuration. Breadth-first and depth-
first algorithms have been implemented and heuristic search algorithms have also been tried.
After finding a possible configuration of the system, a load flow program is necessary to check
the operating feasibility of the final restoration configuration. This process can be very time
consuming when applied to a large transmission system.
ANNs have attracted much attention due to their computational speed and robustness. They have
become an alternative to modeling of physical systems such as synchronous machine and
transmission line. Absence of full information is not a big as a problem in ANNs as it is in the
other methodologies. A major advantage of the ANN approach is that the domain knowledge is
distributed in manner. Therefore they reach the desired solution efficiently. Most of the
applications make use of the conventional multilayer Perception (MLP) model based on back
propagation algorithm. However, multilayer perception model suffers from slow learning rate
and the need to guess the number of hidden layers and neurons in each hidden layer. Many
improvements are suggested over the conventional MLP to overcome these advantages.
The field of ANNs has a history of nearly five decades but has found solid application only in the
past ten years, and the field is still developing rapidly. In recent years, many interesting
applications of ANNs have been reported in the power system areas like load forecasting, power
system stabilizer design, unit commitment, and security assessment, Economic load Dispatch and
fault analysis [26].
2.9 ANN BASED RESTORATION SCHEME CASE STUDY OF ISLAND
RESTORATION SCHEMES(IRS):
The proposed restoration scheme is composed of several Island Restoration Schemes (IRS). Each
IRS is responsible for the development of an island restoration plan when the power system is
recovering from a wide-area disturbance. The number of IRSs will be defined by off-line studies
and will depend on regional load-generation balance. The division of the system into islands is a
common action in large transmission systems where parallel restoration is more efficient and
desired. The parallel restoration technique is commonly used in the restoration schemes applied
to large transmission systems. This technique is also used in the proposed restoration scheme.
The all-open switching strategy where all circuit breakers of the system are open will be used to
create the islands. In order to restore a power system following a wide-area disturbance, each
IRS of restoration scheme will generate local restoration plans composed of switching sequences
of local circuit breakers and a forecast restoration load [26].
Each IRS is composed of two ANNs and a switching sequence program (SSP). The first ANN of
each IRS is responsible for an island restoration load forecast. The input of this ANN will be a
normalized vector composed of the pre-disturbance load. The second ANN of each IRS is
responsible for the determination of the final island configuration and the associated forecast
restoration load pick up percentage that will generate a feasible operational condition. The input
of this ANN will be a normalized vector composed of the forecast island restoration load
provided by the first ANN of the respective IRS, three elements describing possible unavailable
transmission paths(because of outages) for use in the restoration plan. The final element of each
IRS is the SSP. The SSP will determine the energizing sequence of transmission paths that will
lead to the final configuration chosen by the second ANN. The SSP input vector is composed of
the final restoration island configuration generated by the second ANN of the IRS and an
energizing sequence database. The energizing sequence database of each IRS is composed of
transmission path sequences connecting island generators to island loads. The following figure
illustrates the functional block diagram of an IRS.
The proposed restoration scheme will present a restoration plan to the EMS operator following
the occurrence of a wide area disturbance. The power system operator must apply the all open
switch strategy through the EMS/SCADA or through regional control centers before the plan is
implemented. The restoration plan provided by the proposed scheme will be composed of
energizing sequences and restoration load percentage pick up values for all islands. As the final
step of the total restoration, the closing of the tie-lines will be the responsibility of the system
operator. The tie-lines should be closed when all the islands are restored and are in steady state.
Fig 2.4: Island restoration scheme as a case study.
The transmission system chosen for the application of the technique is the IEEE 162-bus 17-
generator system. The operating conditions of a transmission system can vary significantly
during a year and even during a week. Therefore, more than one operating condition was
simulated to train the IRSs. Due to the size of the transmission system, it was assumed that up to
ten islands could be formed following the occurrence of a wide area disturbance. Each island was
predetermined based upon offline studies satisfying the requirement of approximate load
generation balance within each island. The all open switching strategy was assumed to be in
place after the occurrence of the wide area disturbance. The restoration load considered in this
study was equal to the island pre-disturbance load [26].
2.10 RESTORATION CONSTRAINTS: In order to generate a feasible restoration plan to be used as a training pattern by the IRSs,
certain operational constraints must be considered. The various constraints considered
are:
1. Thermal limits of transmission lines (The maximum amount of power a transmission
line can carry without suffering heat-related deterioration of line equipment).
2. Stability limits
3. Number of lines used in the restoration plan
4. Allowable over and under voltage
5. Recognition of locked cutout circuit breakers
The thermal rating of the normally designed transmission lines depends mainly on the voltage
level at which they operate, the line length and reactance. Power system stability is a subject of
major concern in PSR. The restored system generated by the PSR scheme has to be able to allow
for sufficiently large load and generation variations without encountering undesirable and
uncontrollable behavior that could lead to instability and a recurrence of the system blackout. In
order to check the stability of the restored power system, transient stability studies must be
conducted. The number of transmission lines used in the restoration plan also needs some
consideration. The number of transmission lines used in the PSR plan is very important.
Transmissions play a critical role in reactive power balance and over voltage control during the
restoration implementation. In order to maintain a normal voltage profile and avoid the
generation of excessive reactive power, it is advisable to energize the smallest possible number
of transmission lines in a proper sequence during the restoration process.
Circuit breakers have the capability to go through a certain number of open-close sequences
when automatic enclosing is enabled. Once the available number of open-close sequences is
exhausted, the circuit breaker goes into a lock-out state. Permanent non recoverable equipment
faults may also lead to circuit breaker lock-outs. A locked out circuit breaker will normally
require manual resetting before it can be made available for normal operations. Clearly, the
locked-out circuit breakers cannot be used for automatic restoration and should be taken into
account by the PSR scheme. Power System Restoration (PSR) has been a subject of study for
many years. In recent years many techniques were proposed to solve the limitations of
predetermined restoration guidelines and procedures used by a majority of system operators to
restore a system following the occurrence of a wide area disturbance. This paper discusses
limitations encountered in some currently used PSR techniques and a proposed improvement
based on Artificial Neural Networks (ANNs). This proposed scheme has been tested on a 162-
bus transmission system and compared with a breadth search transmission system. The results
indicate that, this is a feasible option that should be considered for real time applications.
Artificial Neural Networks (ANNs) are computational techniques that try to obtain a
performance similar to that of human performance when solving problems. The building block of
ANN is Artificial Neuron, which has got structural & functional similarities with biological
neurons. ANN is also an efficient alternative for problem solutions where it is possible to obtain
data describing the problem behavior, but a mathematical description of the process is
impossible. The proposed restoration scheme is composed of several Island Restoration Schemes
(IRS). Each IRS is responsible for the development of an Island Restoration Plan when the
power system is recovering from a wide area disturbance [26].
2.11 POWER SYSTEM RESTORATION CASE STUDIES
Most of the published case studies come from North America. A number of black start studies
can be found in [27-31]. The restoration of large power systems in North America for Pacific
Northwest, Ontario-Hydro and Hydro-Quebec is covered in [32-37], and a report dealing with
the restoration of a metropolitan electrical system in [38]. A Mexican study of restoration
policies and their application is treated in [38].
European case studies describe restoration experiences in French, Greek, and Swedish Systems
[39-42], Italy [43, 44], Slovenia [45], and Germany [46].
2.12 POWER SYSTEM PROTECTION
All electrical machines, apparatus and other forms of electrical equipment must satisfy two main
requirements which are; they must be able to operate continuously under normal service
conditions and must be able to withstand short-time over-currents and over-voltages such as may
arise during emergent conditions.
Now, one of the most important components of a power protection system is the relay which is a
device that trips the circuit breakers when the input voltage and current signals correspond to the
fault conditions designed for the relay operation. Relays in general can be classified into the
following categories [47-51]:
Directional Relays: These relays respond to the phase angle difference between two
inputs to the relay.
Differential Relays: These relays respond to the magnitude of the algebraic sum of two or
more of its inputs.
Magnitude Relays: These relays respond to the magnitude of the input quantity.
Pilot Relays: These relays respond to the input signals that are communicated to the relay
from a remote location.
Distance Relays: These relays respond to the ratio of two input phasor signals.
Among the various relays that are used for the protection of power lines distance relays are the
most relevant to fault locators. Usually a pair of these distance relays is used for the protection of
a two-terminal transmission line [52].
2.13 TRANSMISSION LINE FAULT LOCATION TECHNIQUES
The transmission line fault location process has been researched for a while and several
innovative and efficient techniques have been proposed and analyzed by several authors [53-61].
These techniques can be broadly classified as Impedance based methods, Travelling wave based
methods and Artificial Intelligence based methods. Each of these methods is discussed briefly in
the following subsections.
2.13.1 Impedance Based Methods
In the case of Impedance based methods, the operation of the distance relay greatly relies on the
fault resistance and is not successful in cases with very high fault resistance [62]. Impedance
based methods can be classified into single-ended methods and two-ended methods depending
upon the number of terminals at which the voltage and current data are collected.
The basic logic behind a single-ended impedance based fault locator is to calculate the location
of the fault from the apparent impedance seen looking into the line from one end. The various
impedance based methods available in literature are discussed in the upcoming subsections.
2.13.2 Simple Reactance Method
The measured voltage and current values at the terminal are used to calculate the impedance of
the line to the fault position as shown in equation (2.1). Once the line impedance per unit length
has been determined, the fault distance can be calculated accordingly as illustrated by equations
(2.2) and (2.3) [24].
. . . . . . . . . .
(2.1)
Where,
VA is the voltage at terminal A,
= the distance to the fault from the terminal A,
IA = the current flowing out of the terminal A,
Vf = the fault voltage and
ZL = the line impedance.
. . . . . . . . . .
(2.2)
Where,
= the fault current, = the fault resistance as shown in Fig 2.5.
. . . . . . . . . .
(2.3)
Figure 2.5: Faulted Transmission Line illustrating simple-reactance method.
2.13.3 Takagi Method
The Takagi method [63] is very simple yet innovative single-ended impedance based Fault
location technique and is illustrated by Fig 2.6. It requires both the pre-fault and fault data and
enhances the simple reactance method by minimizing the effect of fault resistance and reducing
the effect of load flow.
Figure 2.6: A single-phase circuit illustrating Takagi method.
The Fault Resistance is given by,
. . . . . . . . . .
(2.4)
Where VA is voltage measured at terminal A,
IA is the flowing out of terminal A,
γ is the propagation constant,
ZC is the characteristic impedance,
ZL is the line impedance,
IA is the superposition current which is the difference between the fault current and the pre-fault
current.
And
is the distance to the fault from terminal A. . . . .
(2.5)
Where,
. . . . . . . . . . . .
(2.6)
2.13.4 Modified Takagi Method
The modified Takagi method also called the Zero Sequence current method does not require pre-
fault data because it uses zero-sequence current instead of the superposition current for ground
faults [64]. The location of the fault in this method is given by x in equation (2.7).
. . . . . . . . . .
(2.7)
Where IR is the zero-sequence current
β is the zero-sequence current angle.
The position of the fault ‘x’ is given by equation (2.7);
VA is voltage measured at terminal A,
IA is the flowing out of terminal A and
Z1L is the positive sequence line impedance.
2.13.5 Travelling Wave Based Methods
Travelling wave based methods have been widely used [65-68] for the purpose of fault location
and are usually based on the correlation between the forward and backward waves travelling
along the transmission line as shown in Fig 2.7. The basic idea is to successively identify the
fault initiated by high-frequency travelling waves at the fault locator [69].
Figure 2.7: Illustration of travelling wave based Fault Location.
The time taken by the high frequency components for propagation is used for the location of
fault. In Fig 2.3, a single phase lossless transmission line of length ‘l’ is considered with a
travelling wave velocity of υ, capacitance and inductance per unit length L’ and C’ and a
characteristic impedance of Zc. Assuming the occurrence of a fault at a distance of ‘x’ from the
terminal A, the voltage and current values are given by (2.8) and (2.9).
. . . . . . . . . . .
(2.8)
. . . . . . . . . . .
(2.9)
Whose solutions are given by (2.10) and (2.11)
. . . . . . . .
(2.10)
. . . . . . .
(2.11)
The times taken for the waves to travel from the fault to the discontinuity and are to be
determined using GPS technology. Once this is done, the fault location (x) can be readily
determined by the following equation (2.12)
. . . . . . . . . . .
(2.12)
Where c is the wave propagation speed of 299.79 m/sec.
2.13.6 NEURAL NETWORKS BASED METHODS
Neural networks have been put in use for fault location quite recently [70] and have gained
significant importance since Sobajic and Pao used neural networks for the prediction of critical
clearing time [71]. Wide usage of neural networks started by late 80s and during early 90s.
Neural networks are usually used to achieve greater efficiency in fault detection, classification
and location. A lot of research has been done and abundant literature has been published in the
field of fault location using neural networks. Certain significant techniques and results that have
been published are briefly discussed here. A majority of the work mentioned here made use of
feed-forward multilayer perceptron technique. Kulicke and Dalstein [72] used neural networks
for the detection of faults on transmission lines and also differentiated between arcing and
nonarcing faults. A new technique for the detection and location of high speed faults using neural
networks has been proposed by Rikalo, Sobajic and Kezunovic [73]. Neural network based
single ended fault location techniques have been widely researched by Chen and Maun while
Song used neural networks for fault location on series compensated lines. Other relevant work in
the field of fault location using artificial neural networks can be found in these references [74-
78]
CHAPTER THREE
METHODOLOGY FOR RESEARCH
As discussed in the previous chapter about artificial neural networks been used for the protection
of power transmission lines. The excellent pattern recognition and classification abilities of
neural networks have been cleverly utilized in this thesis to address the issue of transmission line
fault location and restoration of lines after fault.
In this chapter, a complete neural-network based approach has been outlined in detail for the
location of faults and restoration of power transmission lines in a power system. To achieve the
same, the original problem has been dealt with in three different stages namely fault detection,
fault classification and fault location.
3.1 FAULTS IN POWER SYSTEM
A fault in an electrical equipment/apparatus is defined as a defect in the electrical circuit due to
which current is diverted from the intended path.
The nature of a fault simply implies any abnormal condition which causes a reduction in the
basic insulation strength between phase conductors or between phase conductors and earth or
any earthed screen surrounding the conductors. Actually the reduction of insulation strength is
not considered as a fault until it creates some effect on the system, i.e, until it results either in
excessive current or in the reduction of the impedance between conductors or between
conductors and earth to a value below that of the lowest load impedance normal to the circuit
[79].
In an electrical power system comprising of generators, switchgears, transformers, power
receivers and transmission and distribution circuits, it is inevitable that sooner or later some
failure will occur somewhere in the system. The probability of the failure or the occurrence of
abnormal condition is more on power lines-about one-half of the faults occur on the power lines.
This can be explained that the power lines are widely branched, have greater length, operate
under variable weather conditions and are subject to the action of atmospheric disturbances of
electrical nature.
According to the causes of incidence, causes of failures may be classified, as mentioned below,
i. Breakdown may occur at normal voltage due to the deterioration of ageing of insulation
and the damages caused by the unpredictable happenings such as blowing of heavy
winds, trees falling across lines, vehicles colliding with towers or poles, birds shorting
out lines, aircraft colliding with lines, line breaks etc.
ii. Breakdown may occur due to abnormal voltages caused by switching surges or lightning
strokes which may be either direct or induced.
The current practice is of providing a high insulation level of the order of 3 to 5 times the
nominal values of the voltage but still the insulation strength is reduced because of pollution on
an insulator string, commonly caused by deposited soot or cement dust in industrial areas and by
wind borne sea-spray in coastal areas. Initially the insulation resistance is lowered and small
leakage currents are diverted and thus the deterioration is hastened. Even in enclosed
installations such as sheathed and armored cables and metal-clad switchgear, insulation gets
deteriorated because of ageing. Void formation in the insulating compound of underground
cables due to unequal expansions and contractions caused by the increase and decrease in
temperature is another cause of insulation failure.
The line and apparatus insulation may be subjected transient over-voltages because of the
switching operations. The voltage which rises at a rapid rate may achieve a peak value
approaching three times phase-to-neutral voltage. This is the reason that a higher insulation level
is provided initially. In case the insulation levels have been correctly chosen and they have not
been impaired in a way described under (i) above, the system will withstand these routine over-
voltages. But if the insulation gets deteriorated due to one or other reason, it is at the time of
switching that failure may occur [80].
Faults can be classified broadly into four different categories namely:
1. Single line to ground faults
2. line to line faults
3. double-line to ground faults
4. three-phase faults
3.1.1 Single Line-to-Ground Faults
An unloaded generator with its phase grounded is shown in fig 3.1. The generator neutral is
assumed to be grounded through impedance . Let the fault impedance be .
Fig 3.1: single line to ground fault on an unloaded generator
Under fault condition, the currents and voltages are given as
. . . . . . . . .
(3.1)
The symmetrical components of the fault current are given by substituting above values of
currents in equation (3.2) given below;
. . . . . . . . . .
(3.2)
And we get,
. . . . . . . . . .
(3.3)
From the above equation (3.3) we have,
. . . . . . . . . .
(3.4)
From equations (3.1) and (3.4) we have,
. . . . . . . .
(3.5)
Substituting above values in equations (3.6), (3.7) and (3.8)
. . . . . . . . .
(3.6)
Where
. . . . . . . . . . .
(3.7)
. . . . . . . . . . .
(3.8)
We have,
and
. . . . . . . . . . . .
(3.9)
From equations (3.5) and (3.9) we have,
Or . . .
(3.10)
From the above equation (3.10) we have,
. . . . . . . . . .
(3.11)
Therefore the fault current,
. . . . .
(3.12)
It then may be recalled that
The equations (3.4) and (3.10) show that the positive, negative and zero sequence networks are
to be connected in series for the solution of currents under fault conditions. Sequence network
representing the single line to ground on phase is given in fig 3.2.
In case of direct short circuit (i.e, when ) we get fault current from equation (3.12). Thus
in this case fault current is given as ,
. . . . . . . . . .
(3.13)
In fig 3.2, the fault impedance equals zero. In case of generator with isolated neutral, the
zero-sequence impedance becomes infinite and therefore, from equation (3.11)
. . . . . . . . .
(3.14)
Fig 3.2: sequence network representing single line to ground fault on phase of an unloaded
generator.
The same result can be envisaged by looking at the system when the neutral is isolated: there is
no return path for the current and therefore, . This means that for such a
system the fault current is zero.
If the generator was supplying balanced load during pre-fault condition, the load current was of
positive-sequence only. A generator does not generate negative-sequence or zero –sequence
voltage and therefore during pre-fault condition, there was no negative or zero-sequence
component of current to the load. Under the fault condition, the positive-sequence current flows
into the fault provided the pre-fault current into the fault is zero.
3.1.2 Line-To-Line Fault
An unloaded generator with line-to-line fault on its phases and is shown in fig 3.3. The
generator neutral is assumed to be grounded through an impedance . The phase is open. Let
the fault impedance be .
Fig 3.3: line-to-line fault on an unloaded generator.
Under fault condition, the currents and voltages are given as,
and or . . . . . .
(3.15)
The symmetrical components of currents and voltages are given by substituting ;
and in the general expressions for symmetrical components of currents and
voltages at the point of fault.
from which we have,
and
or and . .
(3.16)
. . . . . . . . .
(3.17)
From which we have,
And
Solving the above equations we have,
. . . . . . . .
(3.18)
Now, . . . . . . . . .
(3.19)
But and
Substituting from above equation (3.19) in equation (3.18) we have,
. . . . . . . .
(3.20)
Equations (3.16) and (3.20) suggest parallel connection of positive and negative-sequence
networks through a series impedance , as shown in fig 3.4. since , the zero sequence-
network is unconnected.
Fig 3.4: sequence network representing Line-to-line fault on phases and of an unloaded
generator.
In terms of the thevenin equivalents, from fig 3.4 we have,
. . . . . . . . . . .
(3.21)
From equation (3.19) we have,
. . . . . . . . . .
(3.22)
Knowing , we can calculate from which voltages at the fault point can be
determined.
In case of direct short-circuit (i.e., when ) we get and fault current from equations
(3.21) and (3.22) respectively. Thus we have,
. . . . . . . . . . .
(3.23)
. . . . . . . . . .
(3.24)
3.1.3 Double Line-To-Ground Fault
The circuit diagram for double line to ground fault through an impedance on an unloaded
synchronous generator is shown in fig 3.5. the generator neutral is assumed to be grounded
through an impedance .
Fig 3.5: double line to ground fault on an unloaded generator.
Under fault conditions, currents and voltages are given as,
. . . . . . . . . . .
(3.25)
. . . . . . . .
(3.26)
The symmetrical components of voltages are given as,
or
. . .
(3.27)
But and
. . . . . . . . .
(3.28)
Subtracting equation (3.27) from (3.28) we have
or . . . . .
(3.29)
Equations (3.25), (3.27) and (3.29) suggest the sequence network, as shown in fig 3.6.
Fig 3.6: sequence network representing double line to ground fault on phases of an
unloaded generator.
In terms of the thevenin equivalents from fig 3.6 we have,
. . . . . .
(3.30)
In case of direct short-circuit (i.e, when ) we have,
. . . . . . . . .
(3.31)
For an ungrounded generator, , therefore and we have,
. . . . . . . . . . .
(3.32)
3.1.4 Three Phase Fault
The simplest case of a three phase symmetrical fault is considered here. An unloaded star
connected 3-phase alternator is shorted by a 3-phase fault. The generator neutral is earthed
through an impedance . The connections are shown in fig 3.7 below. The generator terminals
being shorted , and are equal to zero and the sum of the currents , , and is zero.
Fig 3.7: 3-phase fault.
As the generator develops only positive sequence voltage hence we have,
In matrix form
. . . . . . . .
(3.33)
Solving above equations we have,
as is finite; and or
. . . . .
(3.34)
This means only positive-sequence network is present for the solution of three phase fault current
the sequence network is shown in fig 3.8 below.
Z1
Ea
Va1
Ia1
REFERENCE BUS
Fig 3.8: three phase positive sequence network of an unloaded generator.
3.2 MODELLING THE POWER TRANSMISSION LINE SYSTEM
A 132 kV transmission line system has been used to develop and implement the proposed
strategy using ANNs. Fig 4.1 shows a one-line diagram of the system that has been used
throughout the research. The system consists of two generators of 132 kV each located on either
ends of the transmission line along with a three phase fault simulator used to simulate faults at
various positions on the transmission line. The line has been modeled using distributed
parameters so that makes it more accurately which describes a very long transmission line.
A B
Relay
G1 G2
300km
Fig 3.9: online diagram of the studied power system
This power system was simulated using the SimPowerSystems toolbox in Simulink by The
MathWorks. A snapshot of the model used for obtaining the training and test data sets is shown
in Fig 3.10. In Fig 3.10, ZP and ZQ are the source impedances of the generators on either side.
The three phase V-I measurement block is used to measure the voltage and current samples at the
terminal A. The transmission line (line 1 and line 2 together) is 300 km long and the three-phase
fault simulator is used to simulate various types of faults at varying locations along the
transmission line with different fault resistances.
Fig 3.10: a snap shot of the studied power system in simpowersystems.
The values of the three-phase voltages and currents are measured and modified accordingly and
are ultimately fed into the neural network as inputs. The SimPowerSystems toolbox has been
used to generate the entire set of training data for the neural network in both fault and non-fault
cases.
There have been 1100 different fault cases simulated for the purpose of fault detection, 1100
different fault cases simulated for fault classification and varying number of fault cases (based on
the type of fault) for the purpose of fault location.
3.3 OUTLINE OF THE PROPOSED SCHEME
Although the basic concept behind relays remains the same, the digital technology has had a
significant influence on the way relays operate and have offered several improvements over
traditional electromechanical relays.
The main goal of this chapter is to design, develop, test and implement a complete strategy for
the fault diagnosis as shown in Fig 3.11. Initially, the entire data that is collected is subdivided
into two sets namely the training and the testing data sets. The first step in the process is fault
detection. Once we know that a fault has occurred on the transmission line, the next step is to
classify the fault into the different categories based on the phases that are faulted.
Fig 3.11: flow chart showing the outline of the proposed scheme
Then, the third step is to pin-point the position of the fault on the transmission line. The goal of
this thesis is to propose an integrated method to perform each of these tasks using artificial
neural networks. A back-propagation based neural network has been used for the purpose of fault
detection and another similar one for the purpose of fault classification. For each of the different
kinds of faults, separate neural networks have been employed for the purpose of fault location.
Each of these steps has been depicted in the flowchart shown in Fig 3.11.
3.4 DATA PRE-PROCESSING
A reduction in the size of the neural network improves the performance of the same and this can
be achieved by performing feature extraction. By doing this, all of the important and relevant
information present in the waveforms of the voltage and current signals can be used effectively.
Voltage and current waveforms have been generated and were sampled at a frequency of 720
Hertz. The voltage and current samples of all the three phases are noted along with the
corresponding pre-fault values.
Fig 3.12: Data pre-processing illustration.
Fig 3.12 shows the current waveform of a Phase B – ground fault at a distance of 60km from
terminal A on a 300 km transmission line. The waveform is the plot of the samples sampled at a
frequency of 720 Hz. Hence there are 12 samples per each cycle.
Now, the 50th sample (12th sample after the occurrence of the fault) on phase B is noted along
with the 26th sample (12th sample before the occurrence of the fault, corresponding to the post-
fault sample considered). Once this is done, the inputs to the neural network are the ratios of the
voltages and currents in each of the phases before and after the occurrence of fault as shown in
Table 3.1. The inputs in matrix format are shown below:
Where n=38 is the sample at which fault occurred.
Hence, there is a set of six inputs each time (3 for the phase voltages and 3 for the phase
currents) to all the neural networks discussed in this work [80]. Care has been taken each time to
make sure the denominator of each of the inputs is non-zero. If it is zero, the value of n is
incremented by 1 and the next sample is taken into consideration for the entire process. The
advantage of performing this scaling is to reduce the training computation time. For the sake of
illustration, the Table 4.1 shows the voltage and current values that are scaled with respect to
their pre-fault values and used as a part of the training set. In Table 4.1, Va, Vb and Vc are the
post fault voltage and current sample values and Va(pf), Vb(pf) and Vc(pf) are the corresponding
pre-fault values as illustrated earlier. The given table depicts the values for all the various types
of faults and also during the no fault case. The fault has been simulated on a 300 km long
transmission line at a distance of 100 km from the terminal A.
Table 3.1: Sample of Inputs to the neural network for various fault cases.
3.5 OVERVIEW OF THE TRAINING PROCESS
Two important steps in the application of neural networks for any purpose are training and
testing. The first of the two steps namely training the neural network is discussed in this section.
Training is the process by which the neural network learns from the inputs and updates its
weights accordingly. In order to train the neural network we need a set of data called the training
data set which is a set of input output pairs fed into the neural network. Thereby, we teach the
neural network what the output should be, when that particular input is fed into it. The ANN
slowly learns the training set and slowly develops an ability to generalize upon this data and will
eventually be able to produce an output when a new data is provided to it. During the training
process, the neural network’s weights are updated with the prime goal of minimizing the
performance function. This performance function can be user defined, but usually feed-forward
networks employ Mean Square Error as the performance function and the same is adopted
throughout this work.
As already mentioned in the previous chapter, all the voltages and currents fed into the neural
network are scaled with respect to the corresponding voltage and current values before the
occurrence of the fault. The outputs, depending upon the purpose of the neural network might be
the fault condition, the type of fault or the location of the fault on the transmission line.
For the task of training the neural networks for different stages, sequential feeding of input and
output pair has been adopted. In order to obtain a large training set for efficient performance,
each of the ten kinds of faults has been simulated at different locations along the considered
transmission line. In view of all these issues, about 100 different fault cases for each of the 10
kinds of faults have been simulated.
Apart from the type of fault, the phases that are faulted and the distance of the fault along the
transmission line, the fault resistance also has been varied to include several possible real-time
fault scenarios.
The fault resistance has been varied as follows: 0.25 ohm, 0.5 ohm, 0.75 ohm, 1 ohm, 5
ohm, 10 ohm, 25 ohm, 50 ohm.
Fault distance has been varied at an incremental factor of every 3 km on a 300 km
transmission line.
3.6 OVERVIEW OF THE TESTING PROCESS
As already mentioned in the previous section, the next important step to be performed before the
application of neural networks is to test the trained neural network. Testing the artificial neural
network is very important in order to make sure the trained network can generalize well and
produce desired outputs when new data is presented to it.
There are several techniques used to test the performance of a trained network, a few of which
are discussed in this section. One such technique is to plot the best linear regression fit between
the actual neural network’s outputs and the desired targets [81].
Analyzing the slope of this line gives us an idea on the training process. Ideally the slope should
be 1. Also, the correlation coefficient (r), of the outputs and the targets measures how well the
ANN’s outputs track the desired targets. The closer the value of ‘r’ is, to 1, the better the
performance of the neural network. Another technique employed to test the neural network is to
plot the confusion matrix and look at the actual number of cases that have been classified
positively by the neural network [81]. Ideally this percentage is a 100 which means there has
been no confusion in the classification process. Hence if the confusion matrix indicates very low
positive classification rates, it indicates that the neural network might not perform well. The last
and a very obvious means of testing the neural network is to present it with a whole new set of
data with known inputs and targets and calculate the percentage error in the neural networks
output. If the average percentage error in the ANN’s output is acceptable, the neural network has
passed the test and can be readily applied for future use.
The Neural Network toolbox in Simulink by The MathWorks divides the entire set of data
provided to it into three different sets namely the training set, validation set and the testing set.
The training data set as indicated above is used to train the network by computing the gradient
and updating the network weights. The validation set is provided during to the network during
the training process (just the inputs without the outputs) and the error in validation data set is
monitored throughout the training process. When the network starts over fitting the data, the
validation errors increase and when the number of validation fails increase beyond a particular
value, the training process stops to avoid further over fitting the data and the network is returned
at the minimum number of validation errors [81]. The test set is not used during the training
process but is used to test the performance of the trained network. If the test set reaches the
minimum value of MSE at a significantly different iteration than the validation set, then the
neural network will not be able to provide satisfactory performance.
CHAPTER FOUR
EXPERIMENTAL RESULTS AND DISCUSSIONS
4.1 FAULT DETECTION
For the purpose of fault detection, various topologies of Multi-Layer Perceptron have been
studied. The various factors that play a role in deciding the ideal topology are the network size,
the learning strategy employed and the training data set size.
After an exhaustive study, the back-propagation algorithm has been decided as the ideal
topology. Even though the basic back-propagation algorithm is relatively slow due to the small
learning rates employed, few techniques can significantly enhance the performance of the
algorithm. One such strategy is to use the Levenberg-Marquardt optimization technique. The
selection of the apt network size is very vital because this not only reduces the training time but
also greatly enhance the ability of the neural network to represent the problem in hand.
Unfortunately there is no thumb rule that can dictate the number of hidden layers and the number
of neurons per hidden layer in a given problem.
4.1.1 TRAINING THE FAULT DETECTION NEURAL NETWORK
In the first stage which is the fault detection phase, the network takes in six inputs at a time,
which are the voltages and currents for all the three phases (scaled with respect to the pre-fault
values) for ten different faults and also no-fault case. Hence the training set consisted of about
1100 input output sets (100 for each of the ten faults and 100 for the no fault case) with a set of
six inputs and one output in each input-output pair. The output of the neural network is just a yes
or a no (1 or 0) depending on whether or not a fault has been detected. After extensive
simulations it has been decided that the desired network has one hidden layer with 10 neurons in
the hidden layer. For illustration purposes, several neural networks (with varying number of
hidden layers and neurons per hidden layer) that achieved satisfactory performance are shown
and the best neural network has been described further in detail. Figures 4.1 – 4.2 show the error
performance plots of neural networks with 1 and 2 hidden layers respectively. The chosen
network has been depicted in Fig 4.7 and the various error performance plots have been shown in
Figures 4.2 – 4.7.
Fig 4.1 shows the training performance plot of the neural network 6-10-1 (6 neurons in the input
layer, 1 hidden layer with ten neurons in it and one neuron in the output layer). It can be seen that
the network did not achieve the desired Mean Square Error (MSE) goal by the end of the training
process.
Fig 4.1 Mean-square error performance of the network (6-10-1).
Fig 4.2 shows the training performance plot of the neural network with 6-10-5-1 configuration (6
neurons in the input layer, two hidden layers with 10 and 5 neurons respectively and one neuron
in the output layer). It is to be noted that the neural network could not achieve the MSE goal of
0.0001 by the end of the training process.
Fig 4.2: Mean-square error performance of the network (6-10-5-1).
Fig 4.3 shows the training process of the neural network with 6-10-5-3-1 configuration (6
neurons in the input layer, 3 hidden layers with 10, 5 and 3 neurons in them respectively and one
neuron in the output layer).
Figure 4.3 Mean-square error performance of the network (6-10-5-3-1).
From the above training performance plots, it is to be noted that very satisfactory training
performance has been achieved by the neural network with the 6-10-5-3-1 configuration (6
neurons in the input layer, 3 hidden layers with 10, 5 and 3 neurons in them respectively and one
neuron in the output layer). The overall MSE of the trained neural network is way below the
value of 0.0001 and is actually 6.9776 e-5 by the end of the training process. Hence this has been
chosen as the ideal ANN for the purpose of fault detection.
4.1.2 TESTING THE FAULT DETECTION NEURAL NETWORK
Once the neural network has been trained, its performance has been tested by three different
factors. The first of these is by plotting the best linear regression that relates the targets to the
outputs as shown in Fig 4.4.
Fig 4.4: Regression fit of the outputs vs. targets for the network (6-10-5-3-1).
The correlation coefficient (r) is a measure of how well the neural network’s targets can track the
variations in the outputs (0 being no correlation at all and 1 being complete correlation). The
correlation coefficient in this case has been found to be 0.99967 in this case which indicates
excellent correlation.
The second means of testing the performance of the neural network is to plot the confusion
matrices for the various types of errors that occurred for the trained neural network. Fig 4.5 plots
the confusion matrix for the three phases of training, testing and validation. The diagonal cells in
green indicate the number of cases that have been classified correctly by the neural network and
the off diagonal cells which are in red indicate the number of cases that have been wrongly
classified by the ANN. The last cell in blue in each of the matrices indicates the total percentage
of cases that have been classified correctly in green and the vice-versa in red. It can be seen that
the chosen neural network has a percentage of high accuracy in fault detection.
Figure 4.5: Confusion matrices for Training, Testing and Validation Phases.
The third step in the testing process is to create a separate set of data called the test set to analyze
the performance of the trained neural network. A total of 300 different test cases have been
simulated with 200 cases corresponding to different types of faults (about 20 cases for each of
the ten faults where the fault resistance and the fault location have been varied in each case). The
rest of the 100 cases correspond to the no-fault situation.
After the test set has been fed into the neural network and the results obtained, it was noted that
the efficiency of the neural network in terms of its ability to detect the occurrence of a fault has a
percentage of high accuracy. Hence the neural network can, with utmost accuracy, differentiate a
normal situation from a fault condition on a transmission line.
Fig 4.6: Overview of the ANN (6-10-5-3-1) chosen for fault detection.
Figure 4.6 presents a snapshot of the trained ANN with the 6 – 10 – 5 – 3 – 1 configuration and it
is to be noted that the number of iterations required for the training process were 55. It can be
seen that the mean square error in fault detection achieved by the end of the training process was
9.43e-5 and that the number of validation check fails were zero by the end of the training
process.
The structure of the chosen neural network for fault detection is shown in Fig 4.7 with the input
layer, hidden layers and the output layer labeled. It is to be noted that there are 6 neurons in the
input layer, 3 hidden layers with 10, 5 and 3 neurons in them respectively and one neuron in the
output layer.
Fig 4.7: Chosen ANN for Fault Detection (6 – 10 – 5 – 3 – 1)
4.2 FAULT CLASSIFICATION
Once a fault has been detected on the power line, the next step is to identify the type of fault.
This section presents an analysis on the fault classification phase using neural networks. A
review of the different neural networks that were analyzed is provided which is followed by the
chosen network.
Fault classifiers based on neural networks have been extensively proposed and used in the past
and almost all of these classifiers made use of multilayer perceptron neural network and
employed the back-propagation learning strategy. Although back-propagation learning strategy is
inherently slow in learning and poses difficulty in choosing the optimal size of the network, it is
undoubtedly the ideal strategy to be employed when there is a large training set available
because back-propagation algorithm can provide a very compact distributed representation of
complex data sets.
4.2.1 TRAINING THE FAULT CLASSIFIER NEURAL NETWORK
The same process that was employed in the previous section (section 3.3) is also followed in this
section in terms of the design and development of the classifier neural network. The designed
network takes in sets of six inputs (the three phase voltage and current values scaled with respect
to their corresponding pre-fault values). The neural network has four outputs, each of them
corresponding to the fault condition of each of the three phases and one output for the ground
line. Hence the outputs are either a 0 or 1 denoting the absence or presence of a fault on the
corresponding line (A, B, C or G where A, B and C denote the three phases of the transmission
line and G denotes the ground).
Hence the various possible permutations can represent each of the various faults accordingly.
The proposed neural network should be able to accurately distinguish between the ten possible
categories of faults. The truth table representing the faults and the ideal output for each of the
faults is illustrated in Table 4.1.
Table 4.1: Fault classifier ANN outputs for various faults.
TYPES OF FAULTS NETWORK OUTPUTS
A B
C
D
A-G Fault 1 0 0 1
B-G Fault 0 1 0 1
C-G Fault 0 0 1 1
A-B Fault 1 1 0 0
B-C Fault 0 1 1 0
C-A Fault 1 0 1 0
A-B-G Fault 1 1 0 1
B-C-G Fault 0 1 1 1
C-A-G Fault 1 0 1 1
A-B-C Fault 1 1 1 0
Hence the training set consisted of about 1100 input output sets (100 for each of the ten faults
and 100 for the no fault case) with a set of six inputs and one output in each input-output pair.
Back-propagation networks with a variety of combinations of hidden layers and the number of
neurons per hidden layer have been analyzed. Of these, the ones that achieved satisfactory
performance are shown followed by the best neural network which has been described further in
detail. Figures 4.8 – 4.12 show the error performance plots of neural networks with 1 and 2
hidden layers respectively. The chosen network has been depicted in Fig 4.17 and the various
error performance plots have been shown in Figures 4.13 – 4.18.
Fig 4.8 shows the training performance plot of the neural network 6-5-5-31-4 (6 neurons in the
input layer, 3 hidden layers with 5, 5 and 31 neurons in them respectively and 4 neurons in the
output layer). It can be seen that the best validation performance in terms of the Mean Square
Error (MSE) by the end of the training process is 0.01289.
Fig 4.8: Mean-square error performance of the network with configuration (6-5-5-31-4).
Figure 4.9: Mean-square error performance of the network with configuration (6-5-31-4).
Fig 4.9 shows the training performance plot of the neural network 6-5-31-4 (6 neurons in the
input layer, 2 hidden layers with 5 and 31 neurons in them respectively and 4 neurons in the
output layer). It can be seen that the best validation performance in terms of the Mean Square
Error (MSE) by the end of the training process is 0.019773.
Fig 4.10 shows the training performance plot of the neural network 6-5-4 (6 neurons in the input
layer, 1 hidden layer with 5 neurons in it and 4 neurons in the output layer). It can be seen that
the best validation performance in terms of the Mean Square Error (MSE) by the end of the
training process in this case is 0.029578.
Fig 4.10: Mean-square error performance of the network with configuration (6-5-4).
Fig 4.11 shows the training performance plot of the neural network 6-10-4 (6 neurons in the
input layer, 1 hidden layer with 10 neurons in it and 4 neurons in the output layer). It can be seen
that the best validation performance in terms of the Mean Square Error (MSE) by the end of the
training process in this case is 0.0077.
Fig 4.11: Mean-square error performance of the network with configuration (6-10-4).
Fig 4.12 shows the training performance plot of the neural network 6-20-4 (6 neurons in the
input layer, 1 hidden layer with 20 neurons in it and 4 neurons in the output layer). It can be seen
that the best validation performance in terms of the Mean Square Error (MSE) by the end of the
training process in this case is 0.0093975.
Fig 4.12: Mean-square error performance of the network with configuration (6-20-4).
Fig 4.13 shows the training performance plot of the neural network 6-35-4 (6 neurons in the
input layer, 1 hidden layer with 35 neurons in it and 4 neurons in the output layer). It can be seen
that the best validation performance in terms of the Mean Square Error (MSE) by the end of the
training process in this case is 0.00359.
Fig 4.13: Mean-square error performance of the network with configuration (6-35-4).
From the above training performance plots, it is to be noted that satisfactory training
performance has been achieved by the neural network with the 6-35-4 configuration (6 neurons
in the input layer, 35 neurons in the hidden layer and one neuron in the output layer). The overall
MSE of the trained neural network is 0.0035986 and it can be seen from Fig 4.13 that the testing
and the validation curves have similar characteristics which is an indication of efficient training.
Hence this has been chosen as the ideal ANN for the purpose of fault classification.
4.2.2 TESTING THE FAULT CLASSIFIER NEURAL NETWORK
Once the neural network has been trained, its performance has been tested by taking three
different factors into consideration. The first of these is by plotting the best linear regression that
relates the targets to the outputs as shown in Fig 4.14. The correlation coefficient in this case was
found to be 0.98108 which indicates satisfactory correlation between the targets and the outputs.
The dotted line in the figure indicates the ideal regression fit and the red solid line indicates the
actual fit of the neural network. It can be seen that both these lines track each other very closely
which is an indication of very good performance by the neural network.
Fig 4.14: Regression fit of the Outputs vs. Targets of ANN with configuration (6-35-4).
The second factor in the testing process is to plot the Receiver Operating Characteristics curve
(ROC). The ROC curves for each of the training, testing and validation phases have been shown
in Fig 4.15 along with the overall ROC curve. The ROC curves are actually plots between the
true positive rates (rate of positive classification) and the false positive rates (rate of incorrect
classification) of the neural network classifier. Hence, an ideal ROC curve would show points
only in the upper-left corner because that is an indication of high percentage true positivity and
low percentage false positivity in the classification. It is to be noted that the ROC curves plotted
in Fig 4.15 are almost perfect since they all have the lines in the upper-left corner.
Fig 4.15: Gradient and Validation performance of the ANN with configuration (6-35-4).
The third step in the testing process is to create a separate set of data called the test set to analyze
the performance of the trained neural network. A total of 300 different test cases have been
simulated with 550 cases corresponding to different types of faults (about 50 cases for each of
the ten faults where the fault resistance and the fault location have been varied in each case). The
rest of the 50 cases correspond to the no-fault situation.
After the test set has been fed into the neural network and the results obtained, it was noted that
the efficiency of the neural network in terms of its ability to identify the type of the fault has a
percentage of accuracy. Hence the neural network can, with utmost accuracy, differentiate
between the ten possible types of faults on a transmission line. Fig 4.16 provides an overview on
the neural network and is a screen shot of the training window simulated using the Artificial
Neural Network Toolbox in Simulink.
Important things to be noted are that the training process converged in about 144 iterations and
that the performance in terms of mean square error achieved by the end of the training process
was 6.26e-3.
Fig 4.16: Overview of the ANN with configuration (6-35-4), chosen as fault classifier.
Fig 4.17 shows the structure of the chosen ANN for the purpose of fault classification and the
neural network has 6 neurons in the input layer, 35 neurons in the hidden layer and four neurons
in the output layer as shown. Each of the neurons in the output layer would indicate the fault
condition on each of the three phases (A, B and C) and the fourth neuron is to identify if the fault
is a ground fault. An output of 0 corresponds to no fault while an output of 1 indicates that the
phase is faulted.
Fig 4.17: Chosen ANN for Fault Classification (6 – 35 – 4).
4.3 FAULT LOCATION
This section talks about the design, development and the implementation of the neural network
based fault locators for each of the various types of faults. This forms the third step in the entire
process of fault location after the inception of the fault. The following subsections deal with the
various kinds of faults and their error performances individually.
4.3.1 Single Line – Ground Faults
Now that we can detect the occurrence of a fault on a transmission line and also classify the fault
into the various fault categories, the next step is to pin-point the location of the fault from either
ends of the transmission line. Three possible single line – ground faults exist (A-G, B-G, C-G),
corresponding to each of the three phases (A, B or C) being faulted.
4.3.1.1 Training the Neural Network for Single Line – Ground Fault Location
Feed forward back – propagation neural networks have been surveyed for the purpose of single
line – ground fault location, mainly because of the availability of sufficient relevant data for
training. In order to train the neural network, several single phase faults have been simulated on
the transmission line model. For each of the three phases, faults have been simulated at every
3Km on a 300Km long transmission line. Along with the fault distance, the fault resistance has
been varied as mentioned earlier in section 3.4.
Hence, a total of 2400 cases have been simulated (100 for each of the three phases with each of
the eight different fault resistances as 0.25, 0.5, 0.75, 1, 5, 10, 25 and 50 ohms respectively). In
each of these cases, the voltage and current samples for all three phases (scaled with respect to
their pre-fault values) are given as inputs to the neural network. The output of the neural network
is the distance to the fault from terminal A. Firstly, a few of the various neural networks (with
varying combination of hidden layers and number of neurons per hidden layer) that performed
reasonably well are presented along with their respective error performances and then the chosen
neural network is shown with all its characteristics depicted in detail. Efficiency of each of the
trained networks is analyzed based on their regression performance and their performance in the
testing phase. The test performance plots are obtained by simulating various faults on different
phases at varying locations and calculating the error in the output produced by the Neural
Network. Figures 4.18 – 4.25 show the error performance and regression plots of neural
networks with 1 and 2 hidden layers. The chosen network has been depicted in Fig 4.30 and its
various error performance plots have been shown in Figures 4.26 – 4.31.
Fig 4.18 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 2 hidden layers with 5 and 5 neuron in them
respectively and 1 neuron in the output layer (6-5-5-1). The correlation coefficient (r) as
mentioned earlier is a measure of how well the neural network relates the outputs and the targets.
The closer the value of r is to 1, the better the performance of the neural network. The value of r
in this case is found to be 0.99799. In order to test the performance of this network, 12 different
single phase faults have been simulated on different phases with the fault distance being
incremented by 25 Km in each case and the percentage error in calculated output has been
calculated. Fig 4.19 shows the results of this test conducted on the neural network (6-5-5-1). It
can be seen that the maximum error is almost 4.5%.
Fig 4.18: Regression fit of the Outputs vs. Targets with configuration (6-5-5-1).
Fig 4.19: Test Phase performance of the Neural Network with configuration (6-5-5-1).
Fig 4.20 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 25 neurons in the hidden layer and 1neuron in the
output layer (6-25 1). The value of the correlation coefficient r in this case is found to be 0.9959.
In order to test the performance of this network, 12 different single phase faults have been
simulated on different phases with the fault distance being incremented by 25 Km in each case
and the percentage error in calculated output has been calculated. Fig 4.21 shows the results of
this test conducted on the neural network
(6-25-1). It can be seen that the maximum error is around 7% which is not very satisfactory.
Fig 4.20: Regression fit of the outputs versus targets with configuration (6-25-1).
Fig 4.21: Test phase performance of the ANN with configuration (6-25-1)
Fig 4.22 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 16 neurons in the hidden layer and 1 neuron in the
output layer (6 16-1). The value of the correlation coefficient r in this case is found to be
0.99906.
Fig 4.22: Regression fit of the outputs versus targets with configuration (6-16-1).
Fig 4.23: Test phase performance of the neural network with configuration (6-16-1).
In order to test the performance of this network, 12 different single phase faults have been
simulated on different phases with the fault distance being incremented by 25Km in each case
and the percentage error in calculated output has been calculated. Fig 4.23 shows the results of
this test conducted on the neural network (6-16-1). It can be seen that the maximum error is
around 4.75%.
Fig 4.24 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 7 neurons in the hidden layer and 1 neuron in the
output layer (6-7-1). The value of the correlation coefficient r in this case is found to be 0.99924
which is by far the best and the closest to one.
Fig 4.24: Regression fit of the outputs versus targets with configuration (6-7-1).
Fig 4.25: Test phase performance of the ANN with configuration (6-7-1).
In order to test the performance of this network, 100 different single phase faults have been
simulated on different phases with the fault distance being incremented by 10Km in each case
and the percentage error in calculated output has been calculated. Fig 4.25 shows the results of
this test conducted on the neural network (6-7-1). It can be seen that the maximum error is
around 1.65 percent which is very satisfactory. It is to be noted that the average error in fault
location is just 0.89%.
Fig 4.26 shows an overview of the chosen ANN and it can be seen that the training algorithm
used is Levenberg - Marquardt algorithm. The performance function chosen for the training
process is mean square error. Fig 4.27 plots the mean-square error as a function of time during
the learning process and it can be seen that the achieved MSE is about 0.0005056 which is way
below the MSE goal of 0.01.
Fig 4.26: Overview of the chosen ANN with configuration (6-7-1).
Fig 4.27: Mean-square error performance of the network with configuration (6-7-1).
4.3.1.2 Testing the Neural Network for Single Line – Ground Fault Location
Several factors have been considered while testing the performance of the neural networks. One
prime factor that evaluates the efficiency of the ANN is the test phase performance already
illustrated in Fig 4.27. As already mentioned, the average and the maximum error percentages
are in tolerable ranges and hence the networks performance is considered satisfactory. Another
form of analysis is provided by Fig 4.30, which is the gradient and validation performance plot.
It can be seen that there is a steady decrease in the gradient and also that the number of
validation fails are 0 during the entire process which indicates smooth and efficient training.
Fig 4.28: Gradient and validation performance of the network with configuration (6-7-1).
The third factor that is considered while evaluating the performance of the network is the
correlation coefficient of each of the various phases of training, validation and testing. Fig 4.29
shows the regression plots of the various phases such as training, testing and validation. It can be
seen that the best linear fit very closely matches the ideal case with an overall correlation
coefficient of 0.99924.
Fig 4.29: Regression plots of various phases of learning of the ANN with configuration (6-7-1).
Fig 4.30 shows the structure of the chosen ANN for single line – ground faults with 6 neurons in
the input layer, 7 neurons in the hidden layer and 1 neuron in the output layer (6-7-1).
Fig 4.30: Structure of the chosen ANN with configuration (6-7-1).
Table 4.2 illustrates the percentage errors in Fault location as a function of Fault Distance and
Fault Resistance. Two different cases have been considered (shown in adjacent columns), one
with a fault resistance of 20ohms and another with a fault resistance of 60ohms. It is to be noted
that the resistance of 20ohms was used as a part of training data set and hence the average
percentage error in fault location in this case is just 0.1646%. The second case illustrates the
same with a different fault resistance of 60ohms which is relatively very high and is not a part of
the training set. Hence, the performance of the neural network in this case illustrates its ability to
generalize and react upon new data. It is to be noted that the average error in this case is just
0.878% which is very satisfactory. Thus the neural networks performance is considered
satisfactory and can be used for the purpose of single line – ground fault location.
Table 4.2: Percentage errors as a function of fault distance and fault resistance for the
ANN chosen for single line - ground fault location.
Serial No. % Error vs. fault distance
(Fault Resistance=20Ω)
% Error vs. fault distance
(Fault Resistance=60Ω)
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
1 25 25.49 0.163 50 51.56 0.52
2 75 75.58 0.287 100 101.02 0.34
3 125 125.12 0.04 150 153.03 1.01
4 175 175.09 0.03 200 202.67 0.89
5 225 225.91 0.303 250 254.89 1.63
4.3.2 LINE – LINE FAULTS
The design, development and performance of neural networks for the purpose of Line – Line
fault location are discussed in this section. Now that we can detect the occurrence of a fault on a
transmission line and also classify the fault into the various fault categories, the next step is to
pin-point the location of the fault from either ends of the transmission line. Three possible line –
line faults exist (A-B, B-C, C-A), corresponding to each of the three phases (A, B or C) being
faulted.
4.3.2.1 Training the Neural Network for Line – Line Fault Location
Feed forward back – propagation neural networks have been surveyed for the purpose of line –
line fault location, mainly because of the availability of sufficient data to train the network. In
order to train the neural network, several line – line faults have been simulated on the
transmission line model. For each pair formed by the three phases, faults have been simulated at
every 3 Km on a 300 Km long transmission line. Along with the fault distance, the fault
resistance has been varied as 0.25, 0.5, 0.75, 1, 5, 10, 25 and 50 ohms respectively. Hence, a total
of 2400 cases have been simulated (100 for each of the three phases with each of the eight
different fault resistances). In each of these cases, the voltage and current samples for all three
phases (scaled with respect to their pre-fault values) are given as inputs to the neural network.
The output of the neural network is the distance to the fault from terminal A. Hence, each input
output pair consists of six inputs and one output. An exhaustive survey on various neural
networks has been performed by varying the number of hidden layers and the number of neurons
per hidden layer. Certain neural networks that achieved satisfactory performance are presented
first along with their error performance plots. Of these ANNs, the most appropriate ANN is
chosen based on its Mean Square Error performance and the Regression coefficient of the
Outputs versus Targets. Figures 4.31 – 4.32 show the MSE and the Test phase performance
plots of the neural networks 6 – 10 – 20 – 5 – 1 with 3 hidden layers. Figures 4.33 – 4.34 show
the MSE and the Test phase performance plots of the neural network 6 – 10 – 1 with 1 hidden
layer.
Fig 4.31 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 3 hidden layers with 10, 20 and 5 neurons in them
respectively and 1 neuron in the output layer (6 – 10 – 20 – 5 – 1). It can be seen that the best
MSE performance of this neural network is 0.0073438 which is below the MSE goal of 0.01. It
was found that the correlation coefficient between the outputs and the targets was 0.98469 in this
case.
Fig 4.31: Mean Square Error performance plot with configuration (6-10-20-5-1).
In order to test the performance of this network, 12 different line – line faults have been
simulated on different phases with the fault distance being incremented by 25 Km in each case
and the percentage error in calculated output has been calculated. Fig 5.32 shows the results of
this test conducted on the neural network (6-10-20-5-1). It can be seen that the maximum error is
around 2.75 percent.
Fig 4.32: Test Phase performance of the ANN with configuration (6-10-20-5-1).
Fig 4.33 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 10 neurons in the hidden layer and 1 neuron in the
output layer (6 – 10 – 1). It can be seen that the best MSE performance of this neural network is
0.0045535 which is below the MSE goal of 0.01. It was found that the correlation coefficient
between the outputs and the targets was 0.9825 for this neural network.
Fig 4.33: Mean Square Error performance plot with configuration (6-10-1).
Fig 4.34: Test Phase performance of the ANN with configuration (6-10-1).
In order to test the performance of this network, 12 different line – line faults have been
simulated on different phases with the fault distance being incremented by 25 Km in each case
and the percentage error in calculated output has been calculated. Fig 4.34 shows the results of
this test conducted on the neural network (6-10-1). It can be seen that the maximum error is
around 4.65 percent which is unacceptable.
Fig 4.35 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 2 hidden layers with 10 and 5 neurons in them
respectively and 1 neuron in the output layer (6 – 10 – 5 – 1). It can be seen that the best MSE
performance of this neural network is 0.002089 which is below the MSE goal of 0.01. It was
found that the correlation coefficient between the outputs and the targets was 0.98648 for this
neural network.
Fig 4.35: Mean Square Error performance of the ANN with configuration (6-10-5-1).
Fig 4.36: Test phase performance of the neural network with configuration (6-10-5-1).
In order to test the performance of this network, 100 different phase to phase faults have been
simulated on different phases with the fault distance being incremented by 10 Km in each case
and the percentage error in calculated output has been calculated. Fig 4.36 shows the results of
this test conducted on the neural network (6-10-5-1). It can be seen that the maximum error is
around 1.7 percent which is very satisfactory. It is to be noted that the average error in fault
location is just 0.97 percent. Hence, this neural network has been chosen as the ideal network for
the purpose of line – line fault location on transmission lines.
Fig 4.37 shows an overview of the chosen ANN and it can be seen that the training algorithm
used is Levenberg - Marquardt algorithm. The performance function chosen for the training
process is mean square error. Fig 4.38 plots the best linear regression fit between the outputs and
the targets and the correlation coefficient for the same has been found to be 0.98648 which is a
decently good regression fit.
Fig 4.37: Overview of the chosen ANN for Line-Line Faults (6-10-5-1).
Fig 4 .38: Regression fit of the outputs versus targets with configuration (6-10-5-1).
4.3.2.2 TESTING THE NEURAL NETWORK FOR LINE – LINE FAULT LOCATION
Several factors have been considered while testing the performance of the chosen neural
network. One prime factor that evaluates the efficiency of the ANN is the test phase performance
plot which is already illustrated in Fig 4.39. As already mentioned, the average and the
maximum error percentages are in tolerable ranges and hence the network’s performance is
considered satisfactory. Another means of evaluating the ANN is provided by Fig 4.41, which is
the gradient and validation performance plot.
It can be seen that there is a steady decrease in the gradient and also that the number of
validation fails did not exceed 1 during the entire process which indicates smooth and efficient
training because the validation and the test phases reached the MSE goal at the same time
approximately.
Figure 4.39: Gradient and validation performance plot of the ANN (6-10-5-1).
The third factor that is considered while evaluating the performance of the network is the
correlation coefficient of each of the various phases of training, validation and testing. Fig 4.40
shows the regression plots of the various phases such as training, testing and validation. It can be
seen that the best linear fit very closely matches the ideal case with an overall correlation
coefficient of 0.98648.
Fig 4.40: Regression plots of the various phases of learning of the chosen ANN (6-10-5-1).
Fig 4.41 shows the structure of the chosen ANN for line – line faults with 6 neurons in the input
layer, 2 hidden layers with 10 and 5 neurons in them respectively and 1 neuron in the output
layer (6 – 10 – 5 – 1).
Fig 4.41: Structure of the chosen Neural Network (6 – 10 – 5 – 1).
Table 4.3 illustrates the percentage errors in Fault location as a function of Fault Distance and
Fault Resistance. Two different cases have been considered (shown in adjacent columns), one
with a fault resistance of 20 ohms and another with a fault resistance of 60 ohms. It is to be noted
that the resistance of 20 ohms was used as a part of training data set and hence the average
percentage error in fault location in this case is just 0.1386 %. The second case illustrates the
same with a different fault resistance of 60 ohms which is relatively very high and is not a part of
the training set. Hence, the performance of the neural network in this case illustrates its ability to
generalize and react upon new data. It is to be noted that the average error in this case is just
0.966 % which is still very satisfactory. Thus the neural networks performance is considered
satisfactory and can be used for the purpose of line – line fault location.
Table 4.3 Percentage errors as a function of fault distance and fault resistance for the ANN
chosen for line - line fault location.
Serial No. % Error vs. fault distance
(Fault Resistance=20Ω)
% Error vs. fault distance
(Fault Resistance=60Ω)
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
1 25 25.03 0.01 50 51.17 0.39
2 75 75.39 0.13 100 102.52 0.84
3 125 125.67 0.223 150 153.63 1.21
4 175 175.14 0.047 200 201.98 0.66
5 225 225.85 0.283 250 255.19 1.73
4.3.3 DOUBLE- LINE – GROUND FAULTS
The design, development and performance of neural networks for the purpose of Double Line -
Ground fault location are discussed in this section. The third category of faults is the double line
–ground faults. Three possible double line – ground faults exist which are denoted as ABG, BCG
and ACG (based on which two of the three phases A, B and C are faulted).
4.3.3.1 Training the Neural Network for Double Line –Ground Fault Location
Feed forward back – propagation algorithm was once again used for the purpose of double line –
ground fault location on transmission lines. The reason for doing so, as already mentioned is that
these networks perform very efficiently when there is availability of a sufficiently large training
data set. For the purpose of training the neural network, several double line – ground faults have
been simulated on the modeled transmission line on each of the three phases. The various factors
that were varied were the fault distance (incremented by 3 km each time), the fault resistance
(one of the chosen eight different fault resistances) and the phases that were faulted. About 100
fault cases were simulated for each phase with each of the eight different resistances as 0.25, 0.5,
0.75, 1, 5, 10, 25 and 50 ohms respectively. Hence a total of 2400 fault cases were simulated on
the transmission line. In each of these cases, the voltage and current samples on all three phases
(scaled with respect to their pre-fault values) are fed as inputs to the neural network. The neural
network’s output is the distance to the fault from terminal A.
Thus each input output pair fed into the neural network has a set of six inputs and one output.
An exhaustive survey on various neural networks has been performed by varying the number of
hidden layers and the number of neurons per hidden layer. A few neural networks that achieved
satisfactory performance are presented first along with their error performance plots. Of these
ANNs, the most appropriate ANN is chosen based on its Mean Square Error performance and the
Regression coefficient of the Outputs vs. Targets. Figures 4.42 – 4.45 show the MSE and the
Test phase performance plots of the neural networks 6 – 10 – 1 and 6 – 20 – 1 with 1 hidden
layer. Figures 4.46 – 4.49 show the MSE and the Test phase performance plots of the neural
network 6 – 10 – 5 – 1 and 6
– 21 – 11 – 1 with 2 hidden layers.
Fig 4.42 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 10 neurons in the hidden layer and 1 neuron in the
output layer (6 – 10 – 1). It can be seen that the best MSE performance of this neural network is
0.0047967 which is below the MSE goal of 0.01 (denoted by the black dotted line). It was found
that the correlation coefficient between the outputs and the targets was 0.98193 in this case.
Fig 4.42: Mean Square Error performance of the ANN with configuration (6-10-1).
In order to test the performance of this network, 12 different double line – ground faults have
been simulated on different phases with the fault distance being incremented by 25 Km in each
case and the percentage error in ANN’s output has been calculated. Fig 4.43 shows the results of
this test conducted on the neural network (6-10-1). It can be seen that the maximum error is
higher than 5 percent which is exorbitantly high.
Fig 4.43: Test Phase performance of the ANN with configuration (6-10-1).
Fig 4.44 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 20 neurons in the hidden layer and 1 neuron in the
output layer (6 – 20 – 1). It can be seen that the best MSE performance of this neural network is
0.0013561 which is below the MSE goal of 0.01 (denoted by the black dotted line in the figure).
It was found that the correlation coefficient between the outputs and the targets was 0.98804 for
this neural network.
Fig 4.44: Mean Square Error performance of the ANN with configuration (6-20-1).
Fig 4.45: Test Phase performance of the ANN with configuration (6-20-1).
In order to test the performance of this network the same method adopted for the earlier case is
followed. 12 different double line – ground faults have been simulated on different phases with
the fault distance being incremented by 25 Km in each case and the percentage error in ANN’s
output has been calculated. Fig 4.45 shows the results of this test conducted on the neural
network (6-20-1). It is to be noted that the maximum error is higher than 4.75 percent which is
too high for this purpose.
Fig 4.46 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 2 hidden layers with 10 and 5 neurons in them
respectively and 1 neuron in the output layer (6 – 10 – 5 – 1). It can be seen that the best MSE
performance of this neural network is 0.00338785 which is below the MSE goal of 0.01 (denoted
by the black dotted line in the figure). It was found that the correlation coefficient between the
outputs and the targets was 0.98913 for this neural network.
Fig 4.46: Mean Square Error performance of the neural network with configuration (6-10-5-1).
In order to test the performance of this network the same method adopted for the earlier case is
followed. 12 different double line – ground faults have been simulated on different phases with
the fault distance being incremented by 25 Km in each case and the percentage error in ANN’s
output has been calculated. Fig 4.47 shows the results of this test conducted on the neural
network (6-10-5-1). It is to be noted that the maximum error is higher than 3.5 percent which is
still not satisfactory for this purpose.
Fig 4.47: Test Phase performance of the ANN (6-10-5-1).
Fig 4.48 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 2 hidden layers with 21 and 11 neurons in them
respectively and 1 neuron in the output layer (6 – 21 – 11 – 1). It can be seen that the best MSE
performance of this neural network is 0.00159395 which is below the MSE goal of 0.01 (denoted
by the black dotted line in the figure). It was found that the correlation coefficient between the
outputs and the targets was 0.99329 for this neural network which indicates very good regression
fit.
Fig 4.48: Mean Square Error performance of the neural network with configuration (6-21-11-1).
Fig 4.49: Test phase performance of the ANN (6-21-11-1).
In order to test the performance of this network, 100 different double line – ground faults have
been simulated on different phases with the fault distance being incremented by 10 Km in each
case and the percentage error in calculated output has been calculated. Fig 4.49 shows the results
of this test conducted on the neural network (6-21-11-1). It can be seen that the maximum error
is around 1.71 percent which is very satisfactory. It is to be noted that the average error in fault
location is just 0.863 percent. Hence, this neural network has been chosen as the ideal network
for the purpose of double line – ground fault location on transmission lines.
Fig 4.50 shows an overview of the chosen ANN and it can be seen that the training algorithm
used is Levenberg - Marquardt algorithm. The performance function chosen for the training
process is mean square error. Fig 4.51 plots the best linear regression fit between the outputs and
the targets. As already mentioned, the correlation coefficient in this case is found to be 0.99329
which is very good.
Fig 4.50: Overview of the chosen ANN (6-21-11-1) for Double Line-Ground Faults.
Fig 4.51: Regression fit of the outputs versus targets with configuration (6-21-11-1).
4.3.3.2 Testing the Neural Network for Double Line – Ground Fault Location
Now that the neural network has been trained, the next important step is to analyze the
performance of this network which is called testing. The methods and means by which this
neural network has been tested are discussed here under. One important factor that helps test the
network is the test phase performance plot as shown in Fig 4.51.
It is to be noted that both the average as well as the maximum error percentages are in acceptable
levels and hence the networks performance is satisfactory. Another means of determining the
efficiency of a trained neural network is to check the gradient and validation performance plot as
shown in Fig 4.54. It can be seen that there is a steady decrease in the gradient and also that the
maximum number of validation fails is 3 during the training process. This indicates efficient
training because the validation phase follows the test phase closely if the number of validation
fails is low. This further implies that the neural network can generalize new data fed into it more
effectively.
Fig 4.52: Gradient and validation performance plot of ANN with configuration (6-21-11-1).
Fig 4.53: Regression plots of the various stages of learning of ANN (6-21-11-1).
The third factor that is considered while evaluating the performance of the network is the
correlation coefficient of each of the various phases of training, validation and testing. Fig 4.53
shows the regression plots of the various phases such as training, testing and validation. It can be
seen that the best linear fit very closely matches the ideal case with an overall correlation
coefficient of 0.99329.
Fig 4.54 shows the structure of the chosen ANN for double line - ground fault location with 6
neurons in the input layer, 2 hidden layers with 21 and 11 neurons in them respectively and 1
neuron in the output layer (6 – 21 – 11 – 1).
Fig 4.54: Structure of the chosen ANN (6 – 21 – 11 – 1).
Table 5.4 illustrates the percentage errors in Fault location as a function of Fault Distance and
Fault Resistance. Two different cases have been considered (shown in adjacent columns), one
with a fault resistance of 20 ohms and another with a fault resistance of 60 ohms. It is to be noted
that the resistance of 20 ohms was used as a part of training data set and hence the average
percentage error in fault location in this case is just 0.091 %. The second case illustrates the same
with a different fault resistance of 60ohms which is relatively very high and is not a part of the
training set. Hence, the performance of the neural network in this case illustrates its ability to
generalize and react upon new data. It is to be noted that the average error in this case is just
1.122 % which is still acceptable. Thus the neural networks performance is considered
satisfactory and can be used for the purpose of double line – ground fault location.
Table 4.4 Percentage errors as a function of fault distance and fault resistance for the ANN
chosen for double line - ground fault location.
Serial No. % Error vs. fault distance
(Fault Resistance=20Ω)
% Error vs. fault distance
(Fault Resistance=60Ω)
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
1 25 25.53 0.177 50 53.81 1.27
2 75 75.18 0.06 100 103.12 1.04
3 125 125.11 0.037 150 152.13 0.71
4 175 175.16 0.053 200 202.88 0.96
5 225 225.39 0.13 250 254.89 1.63
4.3.4 THREE PHASE FAULTS
The design, development and performance of neural networks for the purpose of three-phase
fault location are discussed in this section. The fourth and the final category of faults are the
three phase faults. There exists only one kind of three phase faults which is denoted as ABC fault
where in all the three phases A, B and C are faulted.
4.3.4.1 Training the Neural Network for Three Phase Fault Location
Feed forward back – propagation algorithm was once again used for the purpose of three phase
fault location on transmission lines. The reason for doing so, as already mentioned is that these
networks perform very efficiently when there is availability of a sufficiently large training data
set. For the purpose of training the neural network, several three phase faults have been
simulated on the modeled transmission line. The various factors that were varied were the fault
distance (incremented by 3km each time) and the fault resistance (one of the chosen eight
different fault resistances). About 100 fault cases were simulated with each of the eight different
resistances as 0.25, 0.5, 0.75, 1, 5, 10, 25 and 50 ohms respectively. Hence a total of 800 fault
cases were simulated on the transmission line. In each of these cases, the voltage and current
samples on all three phases (scaled with respect to their pre-fault values) are fed as inputs to the
neural network. The neural network’s output is the distance to the fault from terminal A. Thus
each input output pair fed into the neural network has a set of six inputs and one output. An
exhaustive survey on various neural networks has been performed by varying the number of
hidden layers and the number of neurons per hidden layer. A few neural networks that achieved
satisfactory performance are presented first along with their error performance plots. Of these
ANNs, the most appropriate ANN is chosen based on its Mean Square Error performance and the
Regression coefficient of the Outputs vs. Targets. Figures 4.55 – 4.57 show the MSE and the
Test phase performance plots of the neural network 6 – 21 – 10 – 1 with 2 hidden layers. Figures
4.58 – 4.60 show the MSE and the Test phase performance plots of the neural network 6 – 21 – 1
with 1 hidden layer.
Fig 4.55 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 2 hidden layers with 21 and 10 neurons in them
respectively and 1 neuron in the output layer (6 – 21 – 10 – 1). The correlation coefficient (r) as
mentioned earlier is a measure of how well the neural network relates the outputs and the targets.
The closer the value of r is, to 1, the better the performance of the neural network. The value of r
in this case is found to be 0.99706.
Fig 4.55: Regression fit of the outputs versus targets of ANN with configuration (6-21-10-1).
Fig 4.56: MSE performance of the neural network with configuration (6-21-10-1).
Fig 4.56 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 2 hidden layers with 21 and 10 neurons in them
respectively and 1 neuron in the output layer (6 – 21 – 10 – 1). It can be seen that the best MSE
performance of this neural network is 0.00067433 (denoted by the dotted green line) which is
below the MSE goal of 0.01 (denoted by the black line).
Fig 4.57: Test Phase performance of the ANN with configuration (6-21-10-1).
In order to test the performance of this network, 12 different three phase faults have been
simulated on the transmission line with the fault distance being incremented by 25Km in each
case and the percentage error in ANN’s output has been calculated. Fig 4.57 shows the results of
this test conducted on the neural network (6-21-10-1). It can be seen that the maximum error is
higher than 3 percent which is fairly satisfactory. However neural networks that can perform
better are more desirable.
Fig 4.58 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 1 hidden layer with 21 neurons in it and 1 neuron in
the output layer (6 – 21 – 1). It can be seen that the best MSE performance of this neural network
is 0.00076875 (denoted by the dotted green line) which is below the MSE goal of 0.01 (denoted
by the black dotted line).
Fig 4.58: MSE performance of the neural network with configuration (6-21-1).
Fig 4.59 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 1 hidden layer with 21 neurons in it and 1 neuron in
the output layer (6 – 21 – 1). The correlation coefficient (r) as mentioned earlier is a measure of
how well the neural network relates the outputs and the targets. The closer the value of r is, to 1,
the better the performance of the neural network. The value of r in this case is found to be
0.99804 which is an improvement from the previous case (6-21-10-1).
Fig 4.59: Regression fit for the outputs versus targets of ANN with configuration (6-21-1).
In order to test the performance of this network, 12 different three phase faults have been
simulated on the transmission line with the fault distance being incremented by 25Km in each
case and the percentage error in ANN’s output has been calculated. Fig 4.60 shows the results of
this test conducted on the neural network (6-21-1). It can be seen that the maximum error is just
lower than 3 percent which is a significant improvement from the previous case.
Fig 4.60: Test Phase performance of the ANN with configuration (6-21-1).
Fig 4.61 plots the best linear regression fit between the outputs and the targets of the neural
network with 6 neurons in the input layer, 3 hidden layers with 6, 21 and 16 neurons in them
respectively and 1 neuron in the output layer (6 – 6 – 21 – 16 – 1). The correlation coefficient (r)
as mentioned earlier is a measure of how well the neural network relates the outputs and the
targets. The closer the value of r is, to 1, the better the performance of the neural network. The
value of r in this case is found to be 0.99897 which is very close to 1.
Fig 4.61: Regression fit of the outputs versus targets of ANN (6-6-21-16-1).
Fig 4.62: Test Phase performance of the ANN (6-6-21-16-1).
In order to test the performance of this network, 100 different three phase faults have been
simulated on the transmission line with the fault distance being incremented by 10 Km in each
case and the percentage error in ANN’s output has been calculated. Fig 4.62 shows the results of
this test conducted on the neural network (6-6-21-16-1). It can be seen that the maximum error is
around 1.62 percent which is very satisfactory. It is to be noted that the average error in fault
location is just 0.677 percent. Hence, this neural network has been chosen as the ideal network
for the purpose of three phase fault location on transmission lines.
Fig 4.63 shows an overview of the chosen ANN and it can be seen that the training algorithm
used is Levenberg - Marquardt algorithm. The performance function chosen for the training
process is mean square error.
Fig 4.63: Overview of the chosen neural network for three phase fault location.
Fig 4.64 shows the performance of the neural network (in terms of training, testing and
validation) with 6 neurons in the input layer, 1 hidden layer with 21 neurons in it and 1 neuron in
the output layer (6 – 6 – 21 –16 – 1). It can be seen that the best MSE performance of this neural
network is 0.00060607 (denoted by the dotted green line) which is below the MSE goal of 0.01
(denoted by the black dotted line).
Fig 4.64: Mean Square Error performance of the neural network (6-6-21-16-1).
4.3.4.2 TESTING THE NEURAL NETWORK FOR THREE PHASE FAULT LOCATION
Now that the neural network has been trained, the next step is to analyze the performance of this
network which is called testing. The methods and means by which this neural network has been
tested are discussed here in this section. One important factor that helps test the network is the
test phase performance plot as shown in Fig 4.64. It is to be noted that both the average as well
as the maximum error percentages in accurately determining the location of the fault are in
acceptable levels and hence the network’s performance is satisfactory.
Fig 4.65: Gradient and validation performance plots of the ANN (6-6-21-16-1).
Fig 4.66: Regression plots of the various phases of learning of the ANN (6-6-21-16-1).
Another important means of determining the efficiency of a trained neural network is to check
the gradient and validation performance plot as shown in Fig 4.67. It can be seen that there is a
steady and smooth decrease in the gradient and also that the maximum number of validation fails
is 0 during the training process. This indicates efficient training because the validation phase
follows the test phase closely if the number of validation fails is low. This is further indicated by
the test and validation curves on Fig 4.66. This further implies that the neural network can
generalize new data fed into it more effectively.
The third factor that is considered while evaluating the performance of the network is the
correlation coefficient of each of the various phases of training, validation and testing. Fig 4.66
shows the regression plots of the various phases such as training, testing and validation. It can be
seen that the best linear fit very closely matches the ideal case with an overall correlation
coefficient of 0.99329.
Fig 4.67 shows the structure of the chosen ANN for three-phase faults with 6 neurons in the
input layer, 1 hidden layer with 21 neurons in it and 1 neuron in the output layer (6 – 6 – 21 –16
– 1).
Fig 4.67: Structure of the chosen ANN (6 – 6 – 21 – 16 – 1).
Table 4.5 illustrates the percentage errors in Fault location as a function of Fault Distance and
Fault Resistance. Two different cases have been considered (shown in adjacent columns), one
with a fault resistance of 20 ohms and another with a fault resistance of 60 ohms. It is to be noted
that the resistance of 20 ohms was used as a part of training data set and hence the average
percentage error in fault location in this case is just 0.178 %. The second case illustrates the same
with a different fault resistance of 60ohms which is relatively very high and is not a part of the
training set. Hence, the performance of the neural network in this case illustrates its ability to
generalize and react upon new data. It is to be noted that the average error in this case is just
0.836 % which is still acceptable. Thus the neural networks performance is considered
satisfactory and can be used for the purpose of three phase fault location.
Table 4.5 Percentage errors as a function of fault distance and fault resistance for the
ANN chosen for three phase fault location.
Serial No. % Error vs. fault distance
(Fault Resistance=20Ω)
% Error vs. fault distance
(Fault Resistance=60Ω)
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
Fault
Distance
(Km)
Measured
Fault
Location
Percentage
Error
1 25 25.51 0.17 50 53.41 0.47
2 75 75.17 0.057 100 103.03 1.01
3 125 125.52 0.28 150 152.37 0.79
4 175 175.69 0.23 200 201.99 0.63
5 225 225.46 0.1533 250 253.84 1.28
CHAPTER FIVE
CONCLUSIONS AND RECOMMENDATIONS
5.1 CONCLUSIONS
This thesis has studied the usage of neural networks as an alternative method for the detection,
classification and location of faults and restoration of power system transmission lines. The
methods employed make use of the phase voltages and phase currents (scaled with respect to
their pre-fault values) as inputs to the neural networks. Various possible kinds of faults namely
single line-ground, line-line, double line-ground and three phase faults have been taken into
consideration into this work and separate ANNs have been proposed for each of these faults.
All the neural networks investigated in this thesis belong to the back-propagation neural network
architecture. A fault location scheme for the transmission line system, right from the detection of
faults on the line to the fault location stage has been devised successfully by using artificial
neural networks.
The simulation results obtained prove that satisfactory performance has been achieved by all of
the proposed neural networks in general. As further illustrated, depending on the application of
the neural network and the size of the training data set, the size of the ANN (the number of
hidden layers and number of neurons per hidden layer) keeps varying. The importance of
choosing the most appropriate
ANN configuration, in order to get the best performance from the network, has been stressed
upon in this work. The sampling frequency adopted for sampling the voltage and current
waveforms in this thesis is just 720 Hz which is very low compared to higher frequencies used in
other works. This is of significant importance because, the lower the sampling frequency, the
lesser the computational burden on the industrial PC that uses the neural networks. This means a
lot of energy savings because a continuous online detection scheme of this kind consumes a large
amount of energy, a major portion of which is due to the continuous sampling of waveforms. The
above mentioned are some significant improvements that this thesis offers over existing neural
network based techniques for transmission line fault location.
To simulate the entire power transmission line model and to obtain the training data set,
MATLAB R2010a has been used along with the SimPowerSystems toolbox in Simulink. In
order to train and analyze the performance of the neural networks, the Artificial Neural Networks
Toolbox has been used extensively.
Some important conclusions that can be drawn from this thesis are:
Neural Networks are indeed a reliable and attractive scheme for an ideal transmission line
fault location scheme especially in view of the increasing complexity of the modern
power transmission systems.
It is very essential to investigate and analyze the advantages of a particular neural
network structure and learning algorithm before choosing it for an application because
there should be a trade-off between the training characteristics and the performance
factors of any neural network.
Back Propagation neural networks are very efficient when a sufficiently large training
data set is available and hence Back Propagation networks have been chosen for all the
three steps in the fault location process namely fault detection, classification, fault
location and restoration of transmission lines.
5.2 RECOMMENDATIONS
As a possible extension to this work, it would be quite useful to analyze all the possible
neural network architectures and to provide a comparative analysis on each of the
architectures and their performance characteristics. The possible neural network
architectures that can be analyzed apart from back propagation neural networks are radial
basis neural network (RBF) and support vector machines (SVM) networks.
For the implementation of this thesis to have effect in Nigerian power system, a total
overhauling of the system is required by changing all analogue system to automatic and
modern computer compliant systems.
Personals should be trained on and about artificial intelligence so as to make the system
work effectively.
REFERENCES
[1] R. Das, D. Novosel, “Review of fault location techniques for transmission and sub –
transmission lines”. Proceedings of 54th Annual Georgia Tech Protective Relaying Conference,
2000.
[2] “IEEE guide for determining fault location on AC transmission and distribution lines”. IEEE
Power Engineering Society Publ., New York, IEEE Std C37.114, 2005.
[3] M. M. Saha, R. Das, P. Verho, D. Novosel, “Review of fault location techniques for
distribution systems”, Proceedings of Power Systems and Communications Infrastructure for the
Future Conference, Beijing, 2002, 6p.
[4] L. Eriksson, M. M. Saha, G. D. Rockefeller, “An accurate fault locator with compensation for
apparent reactance in the fault resistance resulting from remote-end feed”, IEEE
Trans on PAS 104(2), 1985, pp. 424-436.
[5] M. M. Saha, J. Izykowski, E. Rosolowski, “Fault Location on Power Networks”, Springer
publications, 2010.
[6] F. H. Magnago, A. Abur, “Advanced techniques for transmission and distribution system
fault location”, Proceedings of CIGRE – Study committee 34 Colloquium and Meeting,
Florence, 1999, paper 215.
[7] Y. Tang, H. F. Wang, R. K. Aggarwal, “Fault indicators in transmission and distribution
systems”, Proceedings of International conference on Electric Utility Deregulation and
Restructuring and Power Technologies – DRPT, 2000, pp. 238-243.
[8]J. B. Gupta, A Course in Electrical Power, s.k. kataria & sons January 1, 2005
[9] R.J Kafka, D.R Penders, S.H Bomchey, M.M Adibi., “System restoration plan development
for a metropolitan electrical system”, IEEE Trans Power Apparatus Syst, PAS-100 (8) (1981),
pp. 3703–3713.
[10] T. Sakaguchi, K. Matsumoto., “Development of a knowledge based system for a power
system restoration”, IEEE Trans Power Apparatus Syst, PAS-102 (2) (1983), pp. 320–329.
[11] J.J Ancona., “A framework for power system restoration following a major power failure”,
IEEE Trans Power Syst, 10 (3) (1995), pp. 1480–1485.
[12] Y Kojima, S. Warashina, S. Nakamura, K. Matasumoto., “Development of a guidance
method for power system restoration” IEEE Trans Power Syst, 4 (3) (1989), pp. 1219–1227.
[13] F. Felix, A. Monticelli., “Analytical tools for power system restoration—conceptual
design”, IEEE Trans Power Syst, PAS-3 (1) (1988), pp. 10–16.
[14] M.E. Coop, M. M. Adibi., “Bulk power system restoration training technique”, IEEE
Trans Power Syst, 8 (1) (1993), pp. 191–197.
[15] M.M. Adibi, R.J. Kafka., “Power system restoration issues” IEEE Comput Appl Power, 4
(2) (1991), pp. 19–24.
[16] M.M. Adibi, L.H. Fink., “Power system restoration planning” IEEE Trans Power Syst, 9 (1)
(1994), pp. 22–28.
[17] K. Nara, S. Yamashiro, T. Koike., “Evaluation of loads in restoration state by the cost of
interruption of electric service”, Int J Electr Power Energy Syst, 7 (2) (1985), pp. 109–119
[18] K.L. Liou, C.C. Liu, R.F. Chu., “Tie line utilization during power system restoration” IEEE
Trans Power Syst, 10 (1) (1995), pp. 192–198.
[19] J. S. Wu, C. C. Liu, K. L. Liou, and R.F. Chu, “Petri Net Algorithm for Scheduling of
Generic Restoration Actions," IEEE Transactions on Power Systems, vol. 12, no. 1, pp. 69-76,
February 1997.
[20] D. Lindenmeyer, “A framework for power system restoration”, Diplom-ingenieur,
university at karlsruhe (TH), Germany, 1996.
[21] A. S. Bretas, A. G. Phadke ,”Artificial Neural Networks in Power System Restoration”,
IEEE Transactions on Power Delivery, vol. 18, no. 4, october 2003.
[22] J., Hertz. R.G.Palmer., A.S.Krogh. “Introduction to the theory of neural computation”,
Perseus Books. ISBN 0-201-51560-1, 1990
[23] Artificial neural network –‘wikipedia, the free encyclopedia
[24] S. Kamalasadan, “Application of Artificial Intelligence Techniques in Power Systems”,
Electric Power Systems Management, Energy Program, SERD, Asian Institute of Technology,
Bangkok, Thailand November 1998.
[25] I.J.Nagrath & M.Gopal “Neural networks and control systems engineering (3rd
edition)”
[26] A. S. Bretas, A. G. Phadke ,”Artificial Neural Networks in Power System Restoration”,
IEEE Transactions on Power Delivery, vol. 18, no. 4, october 2003
[27] L.N. Hannett, F.P. de Mello, G.H. Tylinski, and W.H. Becker, “Validation of Nuclear Plant
Auxiliary Power Plant Supply by Test," IEEE Transactions on Power Apparatus and Systems,
vol. PAS-101, no. 9, pp. 3068-3074, September 1982.
[28] I.D. Hassan, R. Weronick, and R.M. Bucci, “Evaluating the Transient Performance of
Standby Diesel-generator units by simulation," IEEE Transactions on Energy Conversion, vol. 7,
no. 3, pp. 470-477, June 1992.
[29] R.R. Lindstrom, “Simulation and Field Test of the Black Start of a Large Coal-Fired
Generating Station Utilizing Small Remote Hydro Generation," IEEE Transactions on Power
Systems, vol. 5, no. 1, pp. 162-168, February 1990.
[30] R.D. Shultz and G.A. Mason, “Blackstart Utilization of Remote Combustion Turbines:
analytical Analysis and Field Test," IEEE Transactions on Power Apparatus and Systems, vol.
PAS-103, no. 8, pp. 2186-2191, August 1984.
[31] K.E. Yeager and J.R. Willis, “Modeling of Emergency Diesel Generators in an
800Megawatt Nuclear Power Plant," IEEE Transactions on Energy Conversion, vol. 8, no. 3, pp.
433-441, September 1993.
[32] P.F. Arnold, “Summary of System Restoration Plan for the Paci_c Northwest Power
System," in Proceedings of the IEEE/PES Winter Meeting, New York, NY, USA, 1982.
[33] J. Huang, F.D. Galiana, and G.T. Vuong, “Power System Restoration incorporating
Interactive Graphics and Optimization," in Proceedings of the IEEE Power Industry Computer
Application Conference, Baltimore, MD, USA, November 1992, pp.216-222.
[34] J.A. Huang, L.Audette, and S. Harrison, “A Systematic Method for Power System
Restoration Planning," IEEE Transactions on Power Systems, vol. 10, no. 2, pp. 869-875, May
1995.
[35] G. Morin, “Service Restoration following a Major Failure on the Hydro-Quebec Power
System," IEEE Transactions on Power Delivery, vol. PWRD-2, no. 2, pp. 454-462, April 1987.
[36] D. Scheurer, “System Restoration at Philadelphia Electric Company," in IEEE 8th
Biennial
Workshop on Real-Time Monitoring and Control of Power Systems, Montreal, Canada, October
1985.
[37] E.J. Simburger and F.J. Hubert, “Low Voltage Bulk Power System Restoration Simulation,"
IEEE Transactions on Power Apparatus and Systems, vol. PAS-100, no. 11, pp. 4479-4484,
November 1981.
[38] R.J. Kafka, D.R. Penders, S.H. Bouchey, and M.M. Adibi, “System Restoration Plan
Development for a Metropolitan Electric System," IEEE Transactions on Power Systems, vol.
PAS-100, no. 8, August 1981.
[39] J. Guti_errez, M. Starpolsky, and M. Garc__a, “Policies for Restoration of a Power
System," IEEE Transactions on Power Systems, vol. PWRS-2, no. 2, pp. 436-442, May 1987.
[40] E. Agneholm, “Cold Load Pick-up”, Ph.D. thesis, Chalmers University of Technology,
Department of Electric Power Engineering, Gothenburg, Sweden, 1999.
[41] C. Counan, “Major Incidents on the French Electric System: Potentiality and Curative
Measures Studies," IEEE Transactions on Power Systems, vol. 8, no. 3, pp. 879-886, August
1993.
[42] R. Kearsley, “Restoration in Sweden and Experience gained from the Blackstart of 1983,"
IEEE Transactions on Power Systems, vol. PWRS-2, no. 2, pp. 422-428, May 1987.
[43] B. Del_no, G.B. Denegri, E. C. Bonini, R. Marconato, and P. Scarpellini, “Black Start and
Restoration of a Part of the Italian HV Network: Modelling and Simulation of a Field Test,"
IEEE Transactions on Power Systems, vol. 11, no. 3, pp. 1371-1379, July 1996.
[44] E. Mariani, F. Mastroianni, and V. Romano, “Field Experiences in Re-energization of
Electrical Networks from Thermal and Hydro Plants," IEEE Transactions on Power Apparatus
and Systems, vol. PAS-103, no. 7, pp. 1707-1713, July 1984.
[45] P. Omahen and F. Gubina, “Simulations and Field Tests of a Reactor Coolant Pump
Emergency Start-Up by Means of Remote Gas Units," IEEE Transactions on Energy
Conversion, vol. 7, no. 4, pp. 691-697, December 1992.
[46] A. Vorbach, H. Schwingshandl, D. Rumpel, and M Brockmann, “Power System
Restorations. Methods and Model-simulations," in Power Systems Modelling and Control
Applications - Selected Papers from the IFAC Symposium, Brussels, Belgium, 1989, pp. 281-
285.
[47] V. Cook, “Fundamental aspects of fault location algorithms used in distance protection”,
Proceedings of IEE Conference 133(6), 1986, pp. 359-368.
[48] V. Cook, Analysis of Distance Protection, Research Studies Press Ltd., John Wiley & Sons,
Inc., New York, 1985.
[49] Network Protection & Automation Guide, T&D Energy Automation & Information,
Alstom, France. 1986
[50] A. Wright, C. Christopoulos, Electrical Power System Protection, Chapman & Hall
publications, London, 1993.
[51] G. Ziegler, Numerical Distance Protection, Principles and Applications, Siemens AG,
Publicis MCD Verlag, Erlangen, 2006.
[52] A. Wright, C. Christopoulos, Electrical Power System Protection, Chapman & Hall
publications, London, 1993.
[53] M. B. Djuric, Z. M. Radojevic, V. V. Terzija, “Distance Protection and fault location
utilizing only phase current phasors”, IEEE Transactions of Power Delivery 13(4), 1998, pp.
1020-1026.
[54] L. Eriksson, M. M. Saha, G. D. Rockefeller, “An accurate fault locator with compensation
for apparent reactance in the fault resistance resulting from remote-end feed”, IEEE Trans on
PAS 104(2), 1985, pp. 424-436.
[55] B. Kasztenny, D. Sharples, V. Asaro, “Distance Relays and capacitive voltage transformers
– balancing speed and transient overreach”, Proceedings of 55th Annual Georgia Tech Protective
Relaying Conference, 2001.
[56] Y. Zhang, Q. Zhang, W. Song, “Transmission line fault location for double phase to- earth
fault on non-direct-ground neutral system”, IEEE Transactions on Power Delivery 15(2), 2000,
pp. 520-524.
[57] A. A. Girgis, D. G. Hart, W. L. Peterson, “A new fault location techniques for two and three
terminal lines”, IEEE Transactions on Power Delivery 7(1), 1992, pp. 98-107.
[58] M. M. Saha, J. IzykowskiJ, E. Rosolowski, “A method of fault location based on
measurements from impedance relays at the line ends”, Proceedings of the 8th
International
Conference on Developments in Power Systems Protection – DPSP, IEE CP500, 2004, pp. 176-
179.
[59] X. Wanjing, L. Yuan , “Accurate transmission line fault location considering shunt
capacitances without utilizing line parameters”, Electric Power components and Systems, 2012.
[60] L. Yuan , “Generalized fault location methods for overhead electric distribution systems”,
IEEE Transactions on Power Delivery, vol. 26, no. 1, pp. 53-64, Jan 2011.
[61] L. Yuan, K. Ning, “Fault Location algorithms without utilizing line parameters based on
distributed parameter line model”, IEEE Transactions on Power Delivery, vol. 24, no. 2, pp.
579-584, Apr 2009.
[62] L. Eriksson, M. M. Saha, G. D. Rockefeller, “An accurate fault locator with compensation
for apparent reactance in the fault resistance resulting from remote-end feed”, IEEE Trans on
PAS 104(2), 1985, pp. 424-436.
[63] K. Zimmerman, D. Costello, “Impedance-based fault location experience”, Schweitzer
Engineering Laboratories, Inc. Pullman, WA USA.
[64] T. Takagi, Y. Yamakoshi, M. Yamaura, R. Kondou, and T. Matsushima, “Development of a
New Type Fault Locator Using the One-Terminal Voltage and Current Data,” IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-101, No. 8, August 1982, pp. 2892-
2898.
[65] E. O. Schweitzer, III, “A Review of Impedance-Based Fault Locating experience,”
Proceedings of the 15th Annual Western Protective Relay Conference, Spokane, WA, October
24-27, 1988.
[66] M. Aurangzeb, P. A. Crossley, P. Gale, “Fault location using high frequency travelling
waves measured at a single location on transmission line”, Proceedings of 7th
International
conference on Developments in Power System Protection – DPSP, IEE CP479, 2001, pp. 403-
406.
[67] Z. Q. Bo, G. Weller, M. A. Redfern, “Accurate fault location technique for distribution
system using fault-generated high frequency transient voltage signals”, IEEE Proceedings of
Generation, Transmission and Distribution 146(1), 1999, pp. 73-79.
[68] M. Silva, M. Oleskovicz, D. V. Coury, “A fault locator for transmission lines using
travelling waves and wavelet transform theory”, Proceedings of 8th International conference on
Developments in Power System Protection – DPSP, IEE CP500, 2004, pp. 212-215.
[69] M. M. Saha, J. Izykowski, E. Rosolowski, Fault Location on Power Networks, Springer
publications, 2010.
[70] M. El-Sharkawi, D. Niebur, “A tutorial course on artificial neural networks with
applications to Power systems”, IEEE Publ. No. 96TP 112-0, 1996.
[71] Y. H. Pao, D. J. Sobajic, “Autonomous Feature Discovery of Clearing time assessment”,
Symposium of Expert System Applications to Power Systems, Stockholm – Helsinki, Aug 1988,
pp. 5.22-5.27.
[72] T. Dalstein, B. Kulicke, “Neural network approach to fault classification for highspeed
protective relaying”, IEEE Transactions on Power Delivery, vol. 4, 1995, pp. 1002 – 1009.
[73] M. Kezunovic, I. Rikalo, D. J. Sobajic, “Real-time and Off-line Transmission Line Fault
Classification Using Neural Networks”, Engineering Intelligent Systems, vol. 10, 1996, pp. 57-
63.
[74] T. Bouthiba, “Fault location in EHV transmission lines using artificial neural networks”, Int.
J. Appl. Math. Comput. Sci., 2004, Vol. 14, No. 1, pp. 69-78.
[75] M. Sanaye-Pasand, H. Kharashadi-Zadeh, “An extended ANN-based high speed accurate
distance protection algorithm”, Electric Power and Energy Systems, vol. 28, no. 6, 2006, pp. 387
-395.
[76] B. R. Bhalja, R. P. Maheshwari., “High resistance faults on two terminal parallel
transmission line: Analysis, simulation studies, and an adaptive distance relaying scheme, IEEE
Trans. Power Delivery, vol. 22, no. 2, 2007, pp. 801-812.
[77] R. Venkatesan, B. Balamurugan, “A real-time hardware fault detector using an artificial
neural network for distance protection”, IEEE Trans. on Power Delivery, vol. 16, no. 1, 2007,
pp. 75 – 82.
[78] U. Lahiri, A. K. Pradhan, S. Mukhopadhyaya, “Modular neural-network based directional
relay for transmission line protection”, IEEE Trans. on Power Delivery, vol. 20, no. 4, 2005, pp.
2154-2155.
[79] J. B. Gupta, A Course in Electrical Power, s.k. kataria & sons January 1, 2005
[80] M. Akke, J. T. Thorp, “Some improvements in the three-phase differential equation
algorithm for fast transmission line protection,” IEEE Transactions on Power Delivery, vol. 13,
1998, pp. 66-72.
[81] D. Howard, B. Mark, H. Martin, “The MathWorks user’s guide for MATLAB and
Simulink, Neural Networks Toolbox 6”.