ABSTRACT

A fault occurs when two or more conductors come in contact with each other or ground.
Ground faults are considered as one of the main problems in power systems and account for
more than 80% of all faults. This paper focuses on detecting faults on electric power
transmission lines. Fault detection have been achieved by using artificial neural networks.
Electrical power systems suffer from unexpected failures due to various random causes.
Unpredicted faults that occur in power systems are required to prevent from propagation to other
area in the protective system. The functions of the protective systems are to detect, then classify
and finally determine the location of the faulty line of voltage and/or current line magnitudes.
Then at last, for isolation of the faulty line the protective relay have to send a signal to the circuit
breaker. The ability to learn, generalize and parallel processing, pattern classifiers is powerful
applications of NN used as an intelligent means for detection. This paper presents neural network
NN architecture for fault detection in a transmission line power system. It aims to implement
complete scheme for distance protection that subdivided into different neural networks zones.
Single phase to ground, double phase and double phase to ground faults are considered. As a
result a protection relaying system for the power transmission line systems can be done using the
NNBP architecture.

1
 Chapter-1

INTRODUCTION

Overhead transmission lines are parts of the main components in an electric power
system and, because transmission lines are exposed to the nature, the possibility of
experiencing faults on transmission lines is generally higher than that on other main
components. Line faults are the most common faults because lines are exposed to the elements
and there are many causes of faults. Lines faults may be triggered by lightning strokes, trees
may fall across lines, fog and salt spray on dirty insulators may cause the insulator strings to
flash over, and ice and snow loading may cause insulator strings to fail mechanically. When a
fault occurs on an electrical transmission line, it is very important to detect it and find its type
and location in order to make necessary repairs and to restore power as soon as possible, the
time needed to determine the fault point along the line will affect the quality of power
delivery. Therefore, an accurate fault type classification and location on the line is an
important requirement for a permanent fault.
In transmission line protection, the analysis of voltages and currents at the relay
location is essential for the correct decision of the relay tripping action. During the fault,
transients arise in both voltage and current waveforms. The past techniques used to analyze
the symmetrical components of these waveforms. Some of them such as representing
transmission lines by either first or second order differential equations and traveling-wave
techniques have resulted in several commercial developments. However, both these
approaches are based on deterministic computations on a well-defined model of the system to
be protected. This results in difficulty in taking system variation into account as the rules are
fixed. They do not have the ability to adapt dynamically to the system operating conditions,
and to make correct decisions if the signals are uncertain.
With the advent of microprocessors and digital electronics, digital-based relaying has
been developed. Research activity has covered virtually every protection technique.
Furthermore, many novel algorithms and associated hardware implementations have emerged.
Recently artificial intelligent techniques are used for the waveforms analysis. The artificial

2
neural network in conjunction with wavelet transform is widely used in fault detection and
classification. Various applications of neural networks were used in the past to improve
recognition of the faults on transmission lines. The main aim of this paper is to design,
develop, test and implement a complete strategy for the fault detection and classification.
Power systems are prone to frequent faults, which may occur in any of its components,
such as generating units, transformers, transmission network and/or loads. It is well known that
faults can cause significant disruption of supply, destabilize the entire system and may also cause
injuries to personnel. Detection of faults is therefore of a paramount importance from economic
and operational viewpoints. In addition faults should be detected as quickly as possible, in real
time if possible, so that an appropriate remedial action can be promptly taken before major of the
disruptions to the power supply can occur. Neural networks are based on neurophysical models
of human brain cells and their interconnection. Such networks are characterized by exceptional
pattern recognition and learning capabilities. The major advantage of the neural networks is its
self-learning capability. First, the network is presented with a set of correct input and output
values. Then it adjusts the connection strength among the internal network nodes until proper
transformation is learned. Second the network is presented with only the input data, and then it
produces a set of output values
This paper presents a method for detection and identification of the fault type and its zone
in the line. Back propagation neural network approach is studied and implemented. Voltages and
currents signals of the line are observed to perform these three tasks. The detailed coefficients of
all phase current signals that are collected only at the sending end of a transmission line are
selected as parameters for fault classification.

3
 Chapter-2

REVIEW OF LITERATURE

Inception of neural network concept took place in 1943 when Warren McCulloch and
Walter Pitts proposed a model of computing element called McCulloch Pitts neuron. In 1949
Donald Hebb proposed learning scheme for connection weights based on pre- and post-synaptic
values of variable. In 1958 Rosenblatt proposed perceptron rule. After these developments, LMS
algorithm, an important training rule of adaptive elements, which was published in the yearl960
by Widrow and his students, gained much popularity. In 1970. Grossberg developed Adaptive
Resonance Theory. In 1980, Hopfield and others introduced outer product rules for training a
class of recurrent networks, which arc now called Hopfield model. Parallel to these
developments Widrow and his students devised Madaline-I rule and applied this to speech and
pattern recognition, weather forecasting and adaptive controls. After this, their work switched to
adaptive signal processing. Till these developments in neural networks research, attempt to
develop learning rule for network with multiple adaptive layer were unsuccessful.
The first major extension of feedforward neural network took place in 1971. when
Werbos developed backpropagation learning algorithms. Unfortunately his work remained
unnoticed. In 1985-86 Rumelhart, Hinton and Williams rediscovered this technique. The
elements used by Rumelhart in the backpropagation network differ from Madaline architectures.
Madaline structure uses hard limiting quantizer, while the elements in the backpropagation
network use only differentiable nonlinearities.
After the development of BP algorithm research in neural network gained momentum and
opened the possibility of their application to many diverse Fields. Power System is one such
area where processing of voltage and current is needed. A. K. Gosh et. al. classified pow'er
system disturbances with ANN approach. Jiansheng Huang applied neural fuzzy techniques for
assessment of power quality disturbances. An expert system using two stage artificial neural
network for real time control of capacitor in distribution system is proposed by N. Iwan Santoso ,
A. U. Narendranath et. al. explore the application of ANN for voltage stability assessment Off-
line simulation and on-line calculation of demand & energy losses has been done by T. S. Sidhu ,

4
To achieve multilayer feedforward neural network A has been used by them.
ANN application in power system is not limited to areas discussed above. With the
inception of multilayer training rules and many new architectures the signal processing capability
of ANN has increased many folds. Therefore it is proving more useful in protection as compared
to oilier areas. High impedance fault detection in distribution feeder is always difficult. This can
be detected with ANN approach. N. Kandil et. al. proposed identification of fault in AC-DC
system with feedforward network. In this the faults have been detected with rms and
instantaneous values of voltage signals, Automated fault disturbance and detection is reported in,
Use of multilayer feedforward network is shown for direction discrimination by T. S. Sidhu.
Fault direction discrimination is attempted with feedback network by M. Sanaye-Pasand, Fault
generated high frequency noise signals have been used with preprocessing to detect the faulty
phase Jyh-Shing Roger Jang developed adaptive network based fuzzy inference system and
shown its application for time series prediction. Attempt totrain the neural network with fuzzy
input-output is shown in. These developments have attracted power engineers also to attempt
neuro-fuzzy approaches to address power system protection problems. Some of these attempts
are available.
Review of literature as above clearly suggests that power scientists are keen to explore
all the possibility of intelligent techniques such as fuzzy to be applied for protection needs.

5
 Chapter-3

ELECTRIC POWER TRANSMISSION

Electric power transmission is the bulk movement of electrical energy from a generating
site, such as a power plant, to an electrical substation. The interconnected lines which facilitate
this movement are known as a transmission network.
Most transmission lines are high-voltage three-phase alternating current (AC),
although single phase AC is sometimes used in railway electrification systems. High-voltage
direct-current (HVDC) technology is used for greater efficiency over very long distances
(typically hundreds of miles). HVDC technology is also used in submarine power
cables (typically longer than 30 miles (50 km)), and in the interchange of power between grids
that are not mutually synchronized. HVDC links are used to stabilize large power distribution
networks where sudden new loads, or blackouts, in one part of a network can result in
synchronization problems and cascading failures.

Fig 3.1-Power Generation & Transmission

Electricity is transmitted at high voltages (115 kV or above) to reduce the energy loss
which occurs in long-distance transmission. Power is usually transmitted through overhead
power lines. Underground power transmission has a significantly higher installation cost and
greater operational limitations, but reduced maintenance costs. Underground transmission is
sometimes used in urban areas or environmentally sensitive locations.
A lack of electrical energy storage facilities in transmission systems leads to a key
limitation. Electrical energy must be generated at the same rate at which it is consumed. A
sophisticated control system is required to ensure that the power generation very closely matches

6
the demand. If the demand for power exceeds supply, the imbalance can cause generation
plant(s) and transmission equipment to automatically disconnect or shut down to prevent
damage. In the worst case, this may lead to a cascading series of shut downs and a major
regional blackout. Examples include the US Northeast blackouts of 1965, 1977, 2003, and major
blackouts in other US regions in 1996 and 2011. Electric transmission networks are
interconnected into regional, national, and even continent wide networks to reduce the risk of
such a failure by providing multiple redundant, alternative routes for power to flow should such
shut downs occur. Transmission companies determine the maximum reliable capacity of each
line (ordinarily less than its physical or thermal limit) to ensure that spare capacity is available in
the event of a failure in another part of the network.
The modern electrical power system is divided into three subsystems namely, generation,
transmission, and distribution subsystem. In Nigeria, the generation and distribution subsystems
are linked via the transmission subsystem by means of overhead transmission lines. This part of
the power system network is characterized by very lengthy transmission lines which often pass
through different environmental topography. Therefore, the possibility of experiencing a fault is
very high in overhead transmission lines compared to generation and distribution subsystems in
the power system network.
One of the key factors that hamper the steady supply of electricity is a fault. A fault is an
abnormal flow of current in the transmission lines in the power system networks. These faults are
broadly classified into series faults and shunt faults. This work focuses on shunt faults which
may be caused by lightning strokes, storms, trees falling across lines, trees growing up to the
lines, bird shorting the lines, vandalism, etc. Therefore, for continuity of power supply, the
transmission lines must be protected against any form of disturbances by not only disconnecting
the faulty section of the system through the protective relays and circuit breakers but also by
detecting and classifying the faults. Therefore, in no small measure, this necessitates the need for
an intelligent system that is capable of detecting and classifying faults on the transmission lines.
To this end, one of the main challenges facing the Nigeria 33-kV transmission system network
today is lack of intelligent fault detectors and classifiers.

7
 Chapter-4

TRANSMISSION LINE PROTECTION

Fault is an unwanted short circuit condition that occurs either between two phases of
wires or between a phase of wire and ground. Short circuit is the most risky fault type as low
of heavy currents can cause overheating or create mechanical forces which may damage
equipment and other elements of power system. Faults also can be classified into three types,
such as symmetrical faults, unsymmetrical faults, and open circuit faults.
The fault that results in symmetrical fault currents (i.e., equal currents with 120
displacements) is known as a symmetrical fault. Three-phase fault is an example of
symmetrical fault where all three phases are short circuited with or without involving the
ground. Examples of different unsymmetrical faults are single phase to ground, two phases to
ground, and phase to phase short circuits. The details of these shunt fault types that can occur
in transmission line are described as follows.

 Single Phase to Ground (L-G) Fault. L-G is a short circuit between any one of phase
conductors and earth (prevalence is 70%–80%). It may be caused either by insulation
failure between a phase conductor and earth or breaking and falling of phase conductor
to the ground.

 Two Phases to Ground (L-L-G) Fault. L-L-G is a short circuit between any two phases
and earth (prevalence is 10%–17%).

 Phase to Phase (L-L) Fault. L-L is a short circuit between any two phases of the
system (prevalence is 8%–10%).

 Three-Phase (L-L-L) Fault. L-L-L is a short circuit between any two phases of the
system (prevalence is 2%-3%).

When a fault occurs, the power transmission line can be protected by fuses, over
current relays, distance relays, pilot protection schemes or by a combination of these relays.
Fuses and over current relays are generally used for the protection of distribution lines
because these devices are simple and inexpensive. Over current relays are also applicable,

8
where a large impedance component such as a transformer, is involved. Currents due to a fault
on the load end of a transformer are considerably less than the currents due to a fault on the
source end. Therefore it is possible to current grade the relays. With the passage of time and
increase in demand of electric energy, power systems have grown to large areas. The HV and
EHV transmission lines connect the generating sources and load points located at long
distance. Due to the long lengths of transmission lines and networking of the transmission
systems, over current relays cannot be used to protect. Distance relays are the most often used
to protect transmission lines. Distance relays take voltages and currents as inputs from the
power system and calculate impedance. If the calculated impedance lies in a pre-defined
operating region on the impedance plane, distance relays operate to isolate the faulted .

Digital relays, used in the transmission line protection, are based on the same
protection logic as is used in analog relays but, the physical structure and operation of the
digital relays are different. Most digital relays, presently in practice, use the fundamental
frequency components of the voltages and currents to protect the transmission line. Voltages
and currents on a transmission line contain high frequency components at the occurrence of a
fault. These components, generated by faults in the currents are shown in Figure 1. The high
frequency components contain traveling waves, which originate at the fault and travel away
from it. The information contained in the waves can be extracted and used to detect and locate
faults.

Fig 4.1-Three phase current waveforms in steady state, during a fault and after a fault on phase-a.

9
4.1 Wavelet transform :
To improve the reliability of the protection algorithms, signal processing was
introduced into transmission line protection. For instance, Wavelet Transform (WT) has been
applied to extract transient characteristics of different fault conditions for analyzing various
transient voltage traveling waves. WT has the function of filtering, which can get rid of noise
disturbance. A technique similar to Discrete Fourier Transform (DFT) is presented for
decomposing the transient components of current signals. Due to the effectiveness of WT in
transient analysis, this technique has been combined with other techniques with the aim of
improving the reliability of protection schemes. In other words, wavelet entropy can identify
the power signals with different complexity. Therefore, the wavelet energy is applied to
describe the signals with different complexity which can distinguish internal fault from
external fault. Moreover, wavelet energy entropy has a unique sensibility to slight change of
signals and it can reflect the energy distribution information in both time and frequency
domain. The relative energy entropy is considered to be a measure of the normal signals and
the faulty signals, which can discriminate between the faulty groups and the normal groups
The application of wavelet transform in engineering areas usually requires discrete wavelet
transform (DWT).The representation of DWT can be written as:

(1)

where a0, is the scale factor and b0 is the translation factor,.the functions of integers ‘m’, ‘n’.
‘k’ is an integer variable and it refers to a sample number in an input signal. A very useful
implementation of DWT, called multi-resolution analysis, is demonstrated in Figure 4.2.

10
 Fig.4.2-Wavelet multi-resolution analysis.

The original sampled signal x (n) is passed through a high pass filter h (n) and a low
pass filter l (n). Then the outputs from both filters are decimated by 2 to obtain the detail
coefficients and the approximation coefficients at level 1 (D1 and A1). The approximation
coefficients are then sent to the second stage to repeat the procedure. Finally, the signal is
decomposed at the expected level. In the case shown in Figure 2, if the original sampling
frequency is F, the signal information captured by D1 is between F/4 and F/2 of the frequency
band. D2 captures the information between F/8 and F/4. D3 captures the information between
F/16 and F/8, and A3 retains the rest of the information of original signal between0 and F/16.
By such means, useful information can be easily extracted from the original signal into
different frequency bands and at the same time the information is matched to the related time
period. The original signal is one cycle of a post-fault current signal, as shown in Figure 3.
Db4 wavelet is used to make a 5 level decomposition. The reconstructed versions of each
detail and the approximation are shown in Figure 3. The information of original signal is
clearly represented at each frequency band. The original signal can be reconstructed by adding
up those wavelet signals at the same sample point. The wavelet toolbox in MATLAB provides
a lot of useful techniques for wavelet analysis.

11
 Fig.4.3-An example of wavelet multi-resolution analysis

Signal processing was introduced into transmission line protection to improve the
reliability of the protection algorithms. Wavelets are a class of functions used to localize a given
function in both scaling and position. They are used in applications such as time series analysis
and signal processing. Wavelets form the basis of the wavelet transform which “cuts up data of
operators or functions into different components of frequency, and then studies each component
with a resolution matched to its scale” (Dr I. Daubechies [16]). In the context of signal
processing, the wavelet transform depends upon two variables: time and scale (or frequency).

12
 There are two main types of wavelet transforms: discrete (DWT) and continuous
(CWT).The first deals with functions that are defined over a range of integers (usually t = 0, 1,2.
. . N-1, where N denotes the number of values in the time series).The second is designed to work
with functions defined over the whole real axis.

4.2 spectral energy from wavelet transform :

It can be seen from Figure 3 that the wavelet coefficients d1 to d5 around the fault
inception point fluctuate between positive and negative values. Although, they are quite
distinctive features for fault identification, they are not discernable enough to be directly
employed in the decision making process about a particular fault; hence the spectral energy is
firstly extracted from these signals and it can be calculated using the following formula:

………(2)

where E is the spectral energy, ‘n’ is the coefficient number, ‘k’ is the window’s
number, ‘j’ is the wavelet decomposition level and CD is the magnitude of the coefficient for
the details from WT. Each spectral energy data contains the energy during a certain window
length. In this paper, the number of one coefficient is selected. The reason is that too less
window length will impact on the accuracy. Based on the coefficients generated from the
MATLAB wavelet toolbox, the spectral energy can be calculated and utilized using equation
no.2.

13
 Chapter-5

ARTIFICIAL NEURAL NETWORK

Artificial neural network (ANN) has been equipped with distinctiveness of parallel
processing, nonlinear mapping, associative memory, and offline and online learning abilities.
The wide uses of ANN with its conquering outcomes make it an effective diagnostic mean in
electric power systems. Its versatility with multitude applicability can be seen in other areas
of science and engineering research.

ANN is trained by using prior chosen fault samples as input and set of fault
information as output for fault diagnosis application, and reinforced learning.. This is applied
when the target is having identified value and is associated with each input in the training set.
Figure 4 represents the supervised architecture of ANN.

Fig.5.1-Supervised architecture of ANN.

Error back propagation (BP) neural network was applied for diagnosis of fault in
power system. However slow speed training and the shortcomings of local optimal lead to the
introduction of additional momentum factor for problem solving. Radial basis function (RBF)
neural network has a faster learning speed and the ability of arbitrary function approximation.
In this paper, (RBF) is employed for fault detection and classification for transmission line
protection system.

14
 Artificial neural network (ANN) can be applied to fault detection and classification
effectively because it is a programming technique, capable to solve the nonlinear problems
easily. The ANNs are able to learn with experiences. They are widely accepted and used in
the problem of fault detection and fault classification because of the following features:

a) Number of transmission line configuration are possible as there can be any possibility
from short length, long length, single circuit transmission line to double-circuit
transmission lines, etc.
b) There are several methods to simulate the network with different power system
conditions in a fast and reliable manner.
c) The conditions of the electrical power system change after each and every disturbance.
Hence a neural network is capable to incorporate the dynamic changes in the power
systems.
d) The ANN output is very fast, reliable and accurate depending on the training, because
its working depends upon a series of very simple operations.

The algorithm which employed ANNs programming offers many advantages, but it
also suffers with many disadvantages, which are very complex in nature. Some of the
important factors are the selection of type of network, architecture of the network (which
includes the selection of number of layers, number of neurons in each layer, selection of
activation functions, learning algorithms parameters etc.), termination criteria etc. There are
various parameters like values of the pre fault and post fault voltages and currents of the
respective three phases in steady state required for precise fault detection and classification.
The values of the pre fault and post fault voltage and current of respective three phases
are very different and are governed by the type of fault. Thus, the fault classification method
required a neural network that allows it to determine the type of fault from the patterns of pre
fault and post fault voltages and currents, which are generated from the values measured from
a three phase transmission line of an electrical power system at one terminal. The neural
network is based upon the total three numbers of inputs, i.e. the currents of respective three
phases. The neural network is trained by using these six inputs. The total number of outputs of
the neural network is four in numbers, i.e. three phases A, B, C and fourth is ground of three

15
phase transmission line
Artificial neural network (ANN) is made up of many computational processing elements
called neurons or nodes These nodes operate in parallel and are connected together in topologies
that are loosely modeled after biological neural system The training of ANN is carried out to
associate correct output responses to particular input pattern. The artificial neural network is
trained so that application of a set of input produces the desired set of outputs. Training is
accomplished by sequentially applying input vectors, while adjusting network weights according
to pre-detained procedures. During the training process, the network weights gradually converge
to values such that each input vector produces the desired output vector.
Transfer function in the ANN is an important key element to invoke the nonlinear
relationships that maps the input(s) to the output(s). Without the transfer function the whole
operation is linear and could be solved using linear algebra or similar methods. We consider the
transfer function for the weighted sum S (lumped input) of the inputs for a successful network
design. In the process of learning the neural network presented with pairs of input and output
data then teaches how to produce the output when the corresponding input is presented. When
learning is complete, the trained neural network, with the updated optimal weights, should be
able to produce the output within desired accuracy corresponding to an input pattern.
An artificial Neural Network (ANN) is modeled after the way and manner the biological
neural systems work. They are parallel computational systems and are made of several
processing elements connected together in a particular way in order to perform a particular task .
ANNs are massively paralleled and have the ability to learn from training data and generalize to
new situations. This makes them efficient and robust for real world applications, hence, they are
configured to perform tasks analogous to biological brains. According to, the uniqueness of
ANN that gave it an edge in the artificial intelligent world is based on the fact that, the
information processing of ANN can be carried out in a parallel distributed manner because it is
made of massive interconnection of elementary processing units, ANN can be used to solve
problems that are inherently nonlinear, it requires no prior knowledge functions relating the
problem variables and it is capable of handling situations of incomplete information and corrupt
data. Moreover, ANN learns to produce an output based on a given input data. The training of

16
the network is accomplished by sequentially applying input vectors while adjusting network
weights accordingly. It is in training that the neural network learns to map the input into the
output based on a given input data. The network weight converges gradually as the adjustment of
different weights progresses (training) to values that will enable each input vector to produce the
target. The supervised learning which is commonly used in electric power transmission lines
fault detection and classification is employed in this study. In supervised learning, the network
weights are modified iteratively with the aim of minimizing the error between a given set of
input data and their corresponding target values using backpropagation algorithm. Figure depicts
a supervised learning approach for a feedforward neural network.

Table 5.1- Sample of the scaled Voltage and Current values for Phase A

S/N Va Vb Vc Ia Ib Ic

1 0.3253 0.2693 0.3471 0.4805 0.4861 0.4835

2 0.3979 0.3383 0.3955 0.4336 0.4385 0.4362

3 0.4472 0.3923 0.4380 0.3868 0.3912 0.3891

4 0.5715 0.5604 0.5677 0.1104 0.1114 0.1109

5 0.5755 0.5720 0.5748 0.0604 0.0608 0.0606

6 0.5765 0.5759 0.5765 0.0340 0.0342 0.0341

7 0.5766 0.5767 0.5769 0.0252 0.0253 0.0252

8 0.5768 0.5771 0.5772 0.0183 0.0184 0.0183

17
 Chapter-6

FAULT CLASSIFICATION BY ANN

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. In ANN based method, the entire data that is
collected is subdivided into three sets namely the training, validation 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. The aim of this paper 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 fault classification. For each of the
different kinds of faults, separate neural networks have been employed for the purpose of fault
classification. Each of these steps has been depicted in the flowchart shown in Figure 5.
The flowchart explains the basic procedure which was used to implement the ANN
based fault detection and classification in HV transmission line. During Pre-processing, three
current input signals were sampled at a sampling frequency of 1.0 MH and further processed
by low-pass filter. Then one full cycle Discrete Fourier Transform is used to calculate the
energy output of three currents. The input signals were normalized in order to make the ANN
input level between +1 or 0. The most important aspects of the fault location systems are
network architecture, selection of learning rule, training method, and testing the fault locator
based on ANN.

18
 Fig.6.1-Flowchart depicting the outline of ANN Based.

6.1 Execution for ANN based fault classification:

For the ANN based fault detection and classification, the detailed study is executed at
"Mansan-Shwesaryan, Mandalay Region, Myanmar" 230 kV twin bundled double circuit
transmission line. The Simulink model for the system under study is "Mansan-Shwesaryan,
Mandalay Region, Myanmar" 230 kV twin bundled double circuit transmission line shown in
Figure 6. It consists of 230 kV transmission line 193.2 km in length. A source is located at

19
Mansan bus and a load is attached to Shwesaryan bus. The transmission lines are represented
by distributed parameter lines. The entire transmission line is divided into two and a fault is
located between them as shown in Figure.

Fig 6.2-Simulink Model for ANN Based Fault Detection and Classification.

The important parameters for the model are obtained from line characteristics and conductor
size. They are as follow:

1. Line Length = 120 mile (193.2 km)

2. Positive/negative Sequence Impedance =0.053+j0.142ohm/km
3. Zero Sequence Impedance = 0.2856+j0.7205ohm/km
4. Positive/ negative Sequence Susceptance= 7.92µS/km
5. Zero Sequence Susceptance = 3.84 µS/km
6. Fault Starting = 0.05 seconds
7. Duration in fault = 0.03 Seconds
8. Type Conductor = Peacock

Simulation time is set as 0.1 second with sampling time of 1 µsec. Thus sampling
frequency of the model is 1 MHz.
Current techniques for fault detection and diagnosis rely on experts and expert systems
modeling using Classical techniques in the time or frequency domain. Neural network classifiers
can learn and adapt themselves to different statistical distribution and non-linear mappings. The

20
parallel structure of neural net- works permits ‘INCIPIENT FAULT DETECTION’ which is an
indication of an increase in the lead time for detecting faults.
The design and development of the uses BP network as a fault classifier. The network
designed here has six inputs (the three phase voltages and currents) and four outputs associated
with the four fault categories. The outputs contain variables whose values are given as either 0 or
I corresponding to the three phases and the ground (that is, A, B, C and G) and can be
generalized to represent all the practical fault categories permutation involving combinations of
phases.
To obtain different spectral energy, ten types of faults are performed with different
fault resistance and location as shown in Table.

Table 6.1-Input Data for ANN Training.

Parameter Set Type

Fault Type AG, BG, CG, ABG, BCG, ACG, AB, BC, AC, ABC;

Fault Location in percent 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95;

Fault Resistance(Ω) 0.05, 0.1, 0.2, 0.5, 1, 2, 5, 10, 20, 50;

According to Table, 1900 input training data are obtained for ANN based fault
classification. There are four outputs from ANN as binary digit. Then (1900 × 3) input data
and (1900 × 4) target data are fed to ‘nprtool’ (Neural network pattern-recognition tool). From
400 input data, 70% are used for training, 20 % for validation and 10 percent for testing. In
the ANN training, 2 layers with 50 neurons are set as the basic parameters. The weight and
bias are set random according to default toolbox parameter. The sample input and target data
for AG fault with 20 Ω fault resistance with various fault distances are depicted in Table 2.
For the training case, the Neural Network structure is shown in Figure 7 and performance
curve is shown in Figure.

21
 Fig 6.3-Neural Network Structure for Fault Classification.

To obtain accurate outputs for fault classification, several trainings are carried out
during simulations. Once trained, the networks performance was tested using a validation data
set. The suitable network which showed satisfactory results was finally selected. According to
the ANN training, the resulting Simulink block for fault classification is shown in Figure

(a)

(b)

Fig 6.4-Simulink Block: (a) ANN Output and (b) Testing Arrangement.

22
6.2 Design of neural network for fault isolation/location:

The design and development of the detection neural network is followed in this section in
order to choose the most suitable BP network as a fault isolator. The network is expected to
identify the location of the fault by classifying the identified fault into one of the three fault
zones, namely Z1, Z2 and Z3. In fig 3 each Π Transmission lines placed with certain distances
and also particular per unit values. when fault occurred within the placed distance and with pu
values. Then the proposed neural networks here should isolate the specific zone involved in the
fault. The desired truth table for the network training is shown in Tabl
A large number of BP networks with different structures were studied and analyzed in
order to obtain the simplest structure. The training includes some of the selected networks,
namely structures, 6-5-5-3, 6-6-6-3, 6-7-6-3 and 6-5-4-3. It is found experimentally through trial
and error that a BP network with two hidden layers provides the best training performance. The
first hidden layer has 5 neurons and the second hidden layer has 4 neurons. The activation
function at input layer is linear (-1, 1) function while at hidden layer and output layer is logistic
function. This network is shown below

Fig.6.5-Neural network chosen for fault isolation

23
6.3 Analysis on the Simuation result:

The trained ANN based fault classifier modules were then extensively tested using
independent data sets consisting of fault scenarios never used previously in training. Fault
type, fault location and fault time were changed to investigate the effects of these factors on
the performance of the proposed algorithm. Table gives some examples for the test results.
The left four columns are the desired outputs and the right four columns are the actual outputs
of the ANNs.
Results for fault type classification shown According to the simulation results, the
Neural Network application for classification of faults is consistent. In case of fault
classification, ANN output display same result as desired output.

Fig 6.6-line diagram of transmission line

6.4 Design of ANN for transmission line:

Artificial neural network (ANN) is an interconnected group of artificial neurons that uses
mathematical model or computational model for information processing based on a connectionist
approach to the computation. Transfer function in the ANN is an important key element to
invoke the nonlinear relationships that maps the input(s) to the output(s). In the process of
learning the neural network presented with pairs of input and output data then teached how to
produce the output when the corresponding input is presented. Through iterative training

24
procedure the network’s weights are adjusted by the error signal in a way that the network output
tries to follow the desired output as close as possible.

The learning procedure continues until the error signal is close to zero or below a
predefined value. The sum of errors over all the training sample scan be considered as a kind of
network performance measure, which is a function of free parameters of the system. Such
function can be visualized a multidimensional error surface where network free parameters
serves as coordinates. During the course of learning the system gradually moves to a minimum
point along an error surface. The error surface is determined by the network architecture and the
cost function [17, 4]. Data generated from the transmission line system are collected, trained and
tested. The detection of a fault situation in the system is the first step. Following that is the
investigation of the fault class and finally location of the faulty zone to be isolated.
Artificial neural network (ANN) is made up of many computational processing elements
called neurons or nodes These nodes operate in parallel and are connected together in topologies
that are loosely modeled after biological neural system. The training of ANN is carried out to
associate correct output responses to particular input pattern. The artificial neural network is
trained so that application of a set of input produces the desired set of outputs. Training is
accomplished by sequentially applying input vectors, while adjusting network weights according
to pre-detained procedures. During the training process, the network weights gradually converge
to values such that each input vector produces the desired output vector.
Transfer function in the ANN is an important key element to invoke the nonlinear
relationships that maps the input(s) to the output(s). Without the transfer function the whole
operation is linear and could be solved using linear algebra or similar methods. We consider the
transfer function for the weighted sum S (lumped input) of the inputs for a successful network
design. In the process of learning the neural network presented with pairs of input and output
data then teaches how to produce the output when the corresponding input is presented. When
learning is complete, the trained neural network, with the updated optimal weights, should be
able to produce the output within desired accuracy corresponding to an input pattern.

25
 Chapter-7

ADVANTAGES , DISADVANTAGES AND APPLICATIONS

7.1 Advantages of Artificial Neural Networks (ANN):

a. Storing information on the entire network : Information such as in traditional
programming is stored on the entire network, not on a database. The disappearance of a
few pieces of information in one place does not prevent the network from functioning.

b. Ability to work with incomplete knowledge : After ANN training, the data may
produce output even with incomplete information. The loss of performance here depends
on the importance of the missing information.

c. Having fault tolerance: Corruption of one or more cells of ANN does not prevent it
from generating output. This feature makes the networks fault tolerant.

d. Having a distributed memory: In order for ANN to be able to learn, it is necessary to

determine the examples and to teach the network according to the desired output by
showing these examples to the network. The network's success is directly proportional to
the selected instances, and if the event cannot be shown to the network in all its aspects,
the network can produce false output

e. Gradual corruption: A network slows over time and undergoes relative

degradation. The network problem does not immediately corrode immediately.

f. Ability to make machine learning: Artificial neural networks learn events and make
decisions by commenting on similar events.

g. Parallel processing capability: Artificial neural networks have numerical strength that
can perform more than one job at the same time.

26
7.2 Disadvantages of Artificial Neural Networks (ANN):

a. Hardware dependence: Artificial neural networks require processors with parallel

processing power, in accordance with their structure. For this reason, the realization of
the equipment is dependent.

b. Unexplained behavior of the network: This is the most important problem of

ANN. When ANN produces a probing solution, it does not give a clue as to why and
how. This reduces trust in the network.

c. Determination of proper network structure: There is no specific rule for determining

the structure of artificial neural networks. Appropriate network structure is achieved
through experience and trial and error.

d. Difficulty of showing the problem to the network: ANNs can work with numerical
information. Problems have to be translated into numerical values before being
introduced to ANN. The display mechanism to be determined here will directly
influence the performance of the network . This depends on the user's ability.

e. The duration of the network is unknown: The network is reduced to a certain value of
the error on the sample means that the training has been completed. This value does not
give us optimum results.

7.3 Applications of Artificial Neural Network (ANN) :

Neural networks have broad applicability to real world business problems. In fact, they have
already been successfully applied in many industries. Since neural networks are best at identifying
patterns or trends in data, they are well suited for prediction or forecasting needs including:

a. sales forecasting
b. industrial process control
c. customer research
d. data validation

27
 ANN are also used in the following specific paradigms: recognition of speakers in
communications; diagnosis of hepatitis; undersea mine detection; texture analysis; three-dimensional
object recognition; hand-written word recognition; and facial recognition.

Typical applications of ANN are:

1. Signal classification with Perceptron

2. Weed identification
3. Agricultural Control System Engineeri

28
 Chapter -8

FUTURE SCOPE AND CONCLUSION

A new ANN-based approach to real-time fault classification in power transmission

systems which can be used in digital power system protection has been proposed in this paper. It
uses calculated spectral energy as inputs. Simulation models of the transmission line system are
constructed and the generated information is then channeled using MATLAB software and
accompanying Power System Block Set. Various fault types (AG, BG, CG, ABG, BCG, ACG,
AB, BC, AC, and ABC) are detected and classified. The neural network performance of the
proposed scheme is evaluated using various fault types. It was shown that the network was able
to perform fast and correctly for different combinations of fault conditions, e.g. fault resistance,
fault type and fault location. The fault is identified just in a few milliseconds which prove that
the network is able to detect and classify the fault quite fast. All the test results show that the
proposed fault detector and classifier can be used for supporting a new generation of very high
speed protective relaying systems.

Some important conclusions that can be drawn from this thesis are:

a. Neural Networks are indeed a reliable and attractive scheme for an ideal
transmission line fault scheme specially in view of the increasing complexity of
the modern power transmission systems.
b. 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.
c. Back Propagation neural networks are very efficient when a sufficiently large
training data set is available.

29
 REFERENCE

[1] S. Saha, M. Aldeen, C.P.Tan, “Tan Fault detection in transmission networks of power
systems,” Scince Direct Electrical Power and Energy Systems 33, pp 887–900, 2011.

[2] Cichoki A, Unbehauen R, “Neural networks for optimization and signal processing”, John
Wiley & Sons, Inc., 1993, New York.

[3] Haykin S, “Neural Networks. A comprehensive foundation”, Macmillan Collage Publishing

Company, Inc., 1994, New York.

[4] Kezunovic M, “A survey of neural net applications to protective relaying and fault analysis.”
International Journal of Engineering Intelligent Systems for Electronics, Engineering and
Communications 5(4), pp. 185-192, 1997.

[5] Saha MM, Izykowski J, Rosolowski E, Fault Location on Power Networks, Springer
publications, 2010.

[6] El-Sharkawi M, Niebur D, “A tutorial course on artificial neural networks with applications
to Power systems”, IEEE Publ. No. 96TP 112-120, 1996.

[7] Djuric MB, Radojevic ZM, Terzija VV, “Distance Protection and fault location utilizing only
phase current phasors”, IEEE Transactions of Power Delivery 13(4), pp. 1020-1026, 1998

[8] Bhalja B.R, Maheshwari R.P., “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, pp. 801-812, 2007.

[9] Venkatesan R, Balamurugan B, “A real-time hardware fault detector using an artificial neural
network for distance protection”, IEEE Trans. on Power Delivery, vol. 16, no. 1, pp. 75 – 82,
2007.

[10] Howard Demuth, Mark Beale, Martin Hagan, The MathWorks user‟s guide for MATLAB
and Simulink, Neural Networks Toolbox 6.

30
31