24-01-2010, 06:09 PM
[attachment=1448]
AUTOCONFIGURING ARTIFICIAL NEURAL NETWORK APPLIED TO FAULT DIAGNOSIS IN POWER SYSTEMS
INTRODUCTION:
The fault diagnosis of a power system provides an effective means to
get information about system restoration and maintenance of the power
system. Artificial intelligence has been successfully implemented on
fault diagnosis and system monitoring. Expert systems are used by
defining rules, for a fault diagnosis. In the present work particularly
a new method of AI namely Artificial Neural Network is used as
diagnosing to power system faults.
A study has been made by taking a sample models of power systems. The
all possible faults of the system were diagnosed and predicted with the
help of Auto-Configuring Artificial Neural Network namely Radial
Basis Function Network and the comprehensive study reveals that the
proposed method is more efficient, faster and reliable than any other
method used for fault diagnosis of power systems.
DEFINITION:
Artificial intelligence (AI) is simply the way of making the computer
think intelligently. It there by provides a simple, structured approach
to designing complex decision-making programs. While designing an AI
system, the goal of the system must be kept in mind. There exists a
more sophisticated system; which guides the selection of a proper
response to a specific situation. This process is known as Pruning,
as its name suggests eliminates path way of thoughts that are not
relevant to the immediate objective of reaching a goal.
AI has made a significant impact on power system research. Power system
engineers have applied successfully AI methods to power system research
problems like energy control, alarm processing, fault diagnosis, system
restoration, voltage/var control, etc. for the last couple of years a
new AI method namely Artificial Neural Network (ANN) has been used
extensively in power system research. In comparison to the AI method,
which tries to mimic mental process that takes place in human
reasoning, ANN on the other hand tries to stimulate the neural activity
that takes place in the human brain. ANN has been successfully applied
to economic load dispatch, shot term load forecasting, security
analysis, alarm processing, capacitor installation and EMTP problems.
An attempt has been made here to solve the fault diagnosis problem in
power systems using ANN.
The principal functions of these diagnosis systems are:
1) Detection of fault occurrence
2) Identification of faulted sections
3) Classification of faults into types:
HIFs (high impedance faults) or LIFs(low impedance faults)
This has been achieved through a cascade, multilayered ANN structure.
Using these FDS accurately identifies HIFs, which are relatively
difficult to identify in the other methods.
FAULT ANALYSIS AND PROTECTIVE SYSTEM
A fault in electrical equipment is defined as a defect in its
electrical circuit due to which the current is diverted from the
intended path. Breaking of conductors or failure generally causes
fault. The other causes of fault include mechanical failure, accidents,
excessive internal and external stress the faults can be minimized by
inputting the system, design, quality of equipment and maintenance
Voltage and current unbalanced, Over voltage, Under frequency, Reversal
of power, Power swings, Instability. However the faults can be
eliminated completely.
For the purpose of analysis the faults can be classified as
1) Single line to ground fault
2) Line to line fault
3) Double line ground fault
4) Simultaneous fault
5) Three phase fault
6) Open circuit fault etc
Some of the abnormal conditions are not serious enough to call for
tripping of the circuit breaker. In such cases the protection relaying
is arranged for giving an alarm where as in other cases it is harmful
in such cases the fault should be disconnected immediately without any
delay. This function is performed by protective relaying and switch
gear.
FAULT CALCULATION:
The knowledge of the fault current is necessary for selecting the
circuit breakers of adequate rating, designing the sub “station
equipment, determining the relay setting, etc. The fault calculation
provides the information about the fault currents and the voltages at
various points of the power system under different fault conditions.
The per. Unit (p.u) system normally used for fault calculations
The symmetrical faults such as three phase faults are analyzed on per
phase basis the unsymmetrical fault is calculated by the method of
symmetrical components
Network analyzer and digital computers used for fault calculation for
large systems
ARTIFICIAL INTELLIGENCE APPLIED TO FAULT DIAGONISIS AND POWER SYSTEM
RESTORATIONS:
AI is simply a way of making a computer think intelligently this
accomplished by studying how people think when they are trying to make
decisions and solve problems, breaking these thought processes down
into basic steps and designing a computer program that solves problems
using those some steps .AI thereby provides a simple , structured
approach to design complex decision making programs, human intelligence
is of complex function that scientists have only began to understand,
but enough is known for us to make certain assumptions about how we
think and apply these assumptions in designing AI problems.
SUPERFAST AUTOCONFIGURING ARTIFICIAL NEURAL NETWORK:
The reasons for adapting ANNs are as follows:
¢ Massive parallelism
¢ Distributive representation and computation
¢ Learning ability
¢ Adaptivity
¢ Inherent contextual information processing
¢ Fault tolerance
¢ Low energy consumption
BIOLOGICAL NEURON:
The concept of neuron in ANN structure is divided from biological
neurons. A neuron is special biological structure that process
information. The output area of the neuron is called axon through which
an impulse triggered by the cell can be sent. The input area of the
nerve cell is a branching fiber is called dendrites. When a series of
impulses is received at the dendrites area of the neuron the result is
usually an increase probability that the target nearer will fire an
impulse down its action.
ANN ARCHITECTURE:
ANNs can be categorized into two groups:
¢ Feed forward networks
¢ Recurrent networks
Feed forward networks are static; they produce one set of output values
rather a sequence of values from a given input. These networks are
memory less in the sense their response to an input is independent of
the previous network states. On the other hand recurrent network
systems are dynamic systems when a new input pattern is presented the
neuron outputs are computed, because of the feedback paths. The inputs
to each neuron are then modified, which leads the network to enter a
new state.
In a most common family of feed forward networks is called multilayer
perception, neurons are organized into layers that have unidirectional
connections between them. The bottom layer of units is the input layer,
the only units in a network that receives external inputs. The layer
above is the hidden layer in which the PUs is interconnected to layers
above and below. The top layer is the output layer .the layers are
fully interconnected to each PU is connected to every unit in the layer
above and below it; units are not connected to other units in the same
layer.
A THREE LAYER NEURAL NETWORK
LEARNING:
The ability to learn is a fundamental trait of intelligence. A
learning process in the ANN context can be viewed as the problem of
updating network architecture and connection weights, so that a network
can efficiently perform a specific task. There are three main learning
paradigms:
¢ Supervised
¢ Unsupervised
¢ Hybrid
In supervised learning the network is provided without a correct answer
for every input pattern weights are determined to allow the network to
produce answers as close as possible to the known correct answers.
In unsupervised learning doesnâ„¢t require correct answers associated
with each input pattern in the training dataset. It explores the
underlined structure in a data, or corrections between patterns in the
data and organizes patterns into categories from these correlations.
Hybrid learning combines both the supervised and unsupervised
learningâ„¢s.
TRAINING OF ANN:
There are several training methods used for training of ANN:
¢ Back propagation network(BPN)
¢ Radial basis function network(RBF)
¢ Levenberg-Marquardt network(LMN)
¢ Hopfied network
SYSTEM UNDER STUDY:
Here a sample power system is selected to test the neural network
model.
POWER SYSTEM MODEL-I:
The below power system-I consists of bus bars, transformers,
transmission lines, CBs and protective relays with their back-ups. The
input pattern consists of status (on or off) of the protective relays
and the circuits breakers of the power system. The output pattern for
the training cases consists of the corresponding faults of the system
This power has 10 circuit breakers (CBs), 5 transmission lines (Ls), 2d
buses (Bs), 2 transformers (Ts) and 9 protective relays (Rs). It is
assumed that each protective relay for main and back-up protection and
each line has two protective relays.
POWER SYSTEM-I FOR FAULT DIAGNOSIS.
APPLICATION OF RADIAL BASIS FUNCTION NETWORK TO THE PROBLEM:
As mentioned earlier the radial basis function network model is adopted
for solving the system under study problems. The network can be
represented by a number of inputs, hidden layer and outputs are
calculated and subsequently, radial basis algorithm is applied to
determine the weight element changes. The more efficient batching
operation is applying Q input vector simultaneously and get the network
response to each of them. The inputs and outputs can be represented by
matrices called P and T, which can be written in the following form:
The network also produces the output in matrix form.
P= T=
PERCEPTRON:
The perceptron is the simplest form of the neural network used for
classification. It consists of single layer with adjustable synaptic
weights and a threshold. A single layer perceptron is limited to
performing pattern classification with only two separate classes.
¢ The model of each neuron in the network includes a non linear
element at the output end.
¢ The network contains one or more layers of hidden neurons that
are not of a part of the input or output of the network. The hidden
neurons enable the network to learn complex tasks by extracting
progressively more meaningful features from the input patterns..
The simulation of perceptron consists of two phases
¢ Initialization
¢ Training
INITIALIZATION: The MATLAB function for the initialization is rad. This
function is used to initialize the weights and bias elements to small
positive and negative values.
TRAINING:
The major steps in the training phase can be summarized as follows:
i. The presentation phase: presented the inputs and calculate the
network outputs.
ii. Checking phase: check to see if each output vector is equal to
the target vector associated with the given input.
iii. Training algorithm: training is done by orthogonal least square
algorithm for radial basis function network.
iv. Learning phase: adjust weight and bias accordingly using
perceptron learning rule.
FOR POWER SYSTEM 1:
The input layer of the neural network contains informations about the
above mentioned 10 circuit breakers and 9 protective relays.
The input layers are (from the left):
CB1, CB2, CB3, CB4, CB5, CB6, CB7, CB8, CB9,
CB10,LIM,LIB,L2M,L2B,L3M,L3B,L4M,L4B,L5M,L5B,T1M,T2M,X1B, X2B
Where:
CB*=circuit breaker
L*M=main relay associated with line
L*B=back up relay associated with line
T*M=main relay associated with transformer
X*=main relay associated with bus
The possible faults associated with the given power system are
transmission line faults, transformer faults and bus bar faults
Therefore the variables of the output layer of the neural network1(from
the left)
B1, B2, L1, L2, L3, L4, L5, T1, T2
Where:
B*=fault of bus bar*
L*=fault of line
T*=fault of transformer
The on/off status of the circuit breakers and the relays are
represented by 1s and 2s as defined in the below table.
DEFINITION OF THE STATUS OF THE NEURON
NEURON STATUS
1 2
Relay Not operated Operated
Circuit breaker Not tripped Tripped
Fault components
No fault
fault
The typical input patterns and the corresponding output pattern that
can be used to train the neural network are given below:
TRAINING PATTERNS:
PATTERN-1:
INPUT PATTERN:
112211111111111111111121
OUTPUT PATTERN:
211111111
PATTERN-2:
Failure of line L1, due to over current
Relay operated: L1M
Circuit breaker operated: CB1
INPUT PATTERN:
211111111121111111111111
OUTPUT PATTERN:
112111111
TEST PATTERN:
Failure of main line L1 relay
Relay operated : L1B
Circuit breaker operated: CB1
In this way the patters were computed assuming that only one single
fault occurs at any time. The total number of pattern chosen for the
training sets were equal to 9,in addition to this 5 patters were
selected for the testing of neural network. This test pattern consists
of one piece of equipment malfunction for the single fault(failure of
main relay). Refer the below given tables:
PATTERN GENERATION FOR POWER SYSTEM-1:
TRAINING PATTERN FOR SIMULATION
OUTPUT/
PATTERN 1 2 3 4 5 6 7 8
9
1 B1 2 1 1 1 1 1
1 1 1
2 B2 1 2 1 1 1 1
1 1 1
3 L1 1 1 2 1 1 1
1 1 1
4 L2 1 1 1 2 1 1
1 1 1
5 L3 1 1 1 1 2 1
1 1 1
6 L4 1 1 1 1 1 2
1 1 1
7 L5 1 1 1 1 1 1
2 1 1
8 T1 1 1 1 1 1 1
1 2 1
9 T2 1 1 1 1 1 1
1 1 2
INPUT (9 patterns,24 inputs)
INPUT/
PATTERN 1 2 3 4 5 6 7 8
9
1 CB1 1 1 2 1 1 1 1
1 2
2 B2 1 1 1 2 1 1 1
1 2
3 CB3 2 1 1 1 1 1 1
2 1
4 CB4 2 1 1 1 1 1 1
1 2
5 CB5 1 2 1 1 1 1 1
2 1
6 CB6 1 1 1 1 1 1 1
1 2
7 CB7 1 2 1 1 2 1 1
1 1
8 CB8 1 2 1 1 1 1 1
1 1
9 CB9 1 1 1 1 1 2 1
1 1
10 CB10 1 1 1 1 1 1 2
1 1
11 L1M 1 1 2 1 1 1 1
1 1
12 L1B 1 1 1 1 1 1 1
1 1
13 L2M 1 1 1 2 1 1 1
1 1
14 L2B 1 1 1 1 1 1 1
1 1
15 L3M 1 1 1 1 2 1 1
1 1
16 L3B 1 1 1 1 1 1 1
1 1
17 L4M 1 1 1 1 1 2 1
1 1
18 L4B 1 1 1 1 1 1 1
1 1
19 L5M 1 1 1 1 1 1 2
1 1
20 L5B 1 1 1 1 1 1 1
1 1
21 T1M 1 1 1 1 1 1 1
2 1
22 T2M 1 1 1 1 1 1 1
1 2
23 X1M 2 1 1 1 1 1 1
1 1
24 X2B 1 2 1 1 1 1 1
1 1
OUTPUT (9 pattern,9 output)
INPUT/PATTERN 1 2 3 4 5
1 CB1 2 1 1 1 1
2 CB2 1 2 1 1 1
3 CB3 1 1 1 1 1
4 CB4 1 1 1 1 1
5 CB5 1 1 1 1 1
6 CB6 1 1 1 1 1
7 CB7 1 1 1 1 1
8 CB8 1 1 1 1 1
9 CB9 1 1 1 2 1
10 CB10 1 1 1 1 2
11 L1M 1 1 1 1 1
12 L1B 2 1 1 1 1
13 L2M 1 1 1 1 1
14 L2B 1 2 1 1 1
15 L3M 1 1 1 1 1
16 L3B 1 1 2 1 1
17 L4M 1 1 1 1 1
18 L4B 1 1 1 2 1
19 L5M 1 1 1 1 1
20 L5B 1 1 1 1 2
21 T1M 1 1 1 1 1
22 T2M 1 1 1 1 123
X1M 1 1 1 1 1
24 X2B 1 1 1 1 1
OUTPUT/
PATTERN 1 2 3 4 5
1 B1 1 1 1 1 1
2 B2 1 1 1 1 1
3 L1 2 1 1 1 1
4 L2 1 2 1 1 1
5 L3 1 1 2 1 1
6 L4 1 1 1 2 1
7 L5 1 1 1 1 2
8 T1 1 1 1 1 1
9 T2 1 1 1 1 1
OUTPUT(TESTING PATTERN)
(5 patterns 9 outputs)
INPUT( TESTING PATTERN)5 pattern, 24 inputs)
CONCLUSION:
The salient features of the RBF networks are:
a) They are extremely fast due to the hybrid two stage training
scheme employed.
b) They have only a single hidden layer with growing number of
neurons during learning to achieve an optimal configuration.
c) Only a single network parameter called spread factor (SF) is
varied.
LIMITATIONS:
¢ Requires more training data for more accurate results.
¢ Unworthiness to detect multiple faults due to more piece of
equipment malfunctioning.
REFRENCES:
1) Sharestani.S.A, Silartis.J.Y.P Application of pattern recognition
to identification of power faults, electric power system research.
2) Fausett.L Fundamentals of neural networks, PHI, 1994 and several
other technical periodicals.
