ABSTRACT

Transmission and distribution lines are vital links between generating units and consumers.
They are exposed to atmosphere, hence chances of occurrence of fault in transmission line is
very high, which has to be immediately taken care of in order to minimize damage caused by
it. This report focuses on detecting the faults on electric power transmission lines using
artificial neural networks. A feed forward neural network is employed, which is trained with
back propagation algorithm. 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. The developed neural network is capable of detecting single line to
ground and double line to ground for all the three phases. So, an important application of
artificial intelligence is the diagnosis of faults of mechanisms, systems, in general. Based on
the ability of this network to distinguish reliably between different types of faults that may
occur in converter, the feature can be suitably integrated with artificial neural network based
algorithms to improve the dynamic performance of AC-DC power system. Neural networks
are inherently not fault tolerant and hence they can be used to differentiate the faults that may
occur in converter. Also they are capable of handling huge amount of data fast and
efficiently. Therefore, taking into account all the advantageous features of neural network an
attempt has been made to detect the fault in HVDC converter using artificial neural network.

1
 Chapter1

INTRODUCTION

1.1 Prospective of Seminar:-

Confronting with an increasing demand of power, there is a need to explore the most efficient
and reliable bulk power transmission system. High voltage DC current (HVDC) transmission
is an economic way for long distance bulk power transmission and interconnection of
asynchronous system with different frequency. Remote generation and system
interconnections lead to a search for efficient power transmission at increasing power levels.
The problem of AC interconnections particularly in long distance transmission has led to the
development of DC transmission. HVDC allows power transmission between unsynchronized
AC distribution systems, and can increase system stability by preventing cascading failures
due to phase instability from propagating from one part of a wider power transmission grid to
another. HVDC transmission system has become a mature and well accepted technology. It is
recognize as an effective and efficient means of transmitting bulk power over long distance.
The main circuitry comprises of converter transformers, converter bridges, smoothing
reactors and filters. For safe operation of AC-DC systems requires the monitoring of
appropriate system signals and accurate and rapid classification of any perturbations so that
protective control decisions can be made. Power system reliability improves when HVDC
converter faults are detected and eliminated before they deteriorate to severe state.

The synchronized measurement technology emerged as a promising prospect in achieving

real time protection with global positioning system (GPS). The phasor measurements units
(PMU) are the most widely used synchronized measurement devices for power system
applications, whose measurements are synchronized with respect to a GPS clock and PMU
based fault locators are more accurate than the methods based on unsynchronized phasors.
There is a need for the measuring algorithms that have the ability to adapt dynamically to the
system operating conditions such as changes in the system configuration, source impedances
and fault resistances. Keeping this in view the intelligent techniques are under investigation
to increase reliability, speed and accuracy of existing system based on Artificial Neural
Network (ANN), Fuzzy Logic and Wavelet Methods.

Earlier systems use conventional method for the fault detection which results in the late
detection and inaccurate results. Conventional algorithms are based on deterministic

2
computations on a well-defined model for transmission line protection. Conventional distance
relays consider power swing as a fault and tripping because of such malfunctioning would
lead to serious consequences for power system stability. To improve the performance, Neural
Network architecture is used which results in the earlier fault detection. Artificial neural
networks are found to be suitable for above requirements. Neural networks are massively
parallel distributed processors that has natural propensity for storing experimental knowledge
and making it available for use. An important feature of fault diagnosis using neural networks
is that they can interpolate among the training to give an appropriate response for cases
described by neighboring or noisy input data.

Neural network have emerged as a powerful technique for pattern classification, function
approximation, optimization, prediction and automatic control. They exhibit mapping
capability also possess the capability to generalize. They can process the information in
parallel at high speed and in distributed manner. Also abstract the features from inputs
containing irrelevant data. As the neural networks can handle large amount of data fast and
efficiently they can be used for fault detection application as stated above.

1.2 Objectives of Seminar:-

 To empower the use of artificial intelligence technologies in power system.
 To evaluate the performance of HVDC transmission system using artificial neural
network.
 To compare various fault identification techniques in HVDC system.
 To implement the artificial neural network technology in HVDC transmission system
through MATLAB.

1.3 Organization of Seminar:-

Chapter1:- It gives a brief introduction of fault detection technologies used in HVDC

transmission system.

Chapter2:- It describes HVDC transmission system, different types of fault occurs in HVDC
system and their identification.

Chapter3:- This chapter defines the fault identification using artificial intelligence
techniques in HVDC transmission systems.

3
Chapter4:- In this chapter recent scenario of artificial neural network in electrical
engineering is described.

Chapter5:- Applications of artificial neural network in electrical engineering are included in

this chapter.

Chapter6:- In this chapter result analysis and future scope of fault detection using ANN are
shown.

4
 Chapter2

HVDC TRANSMISSION SYSTEM AND THEIR FAULT

DETECTION

2.1. HVDC Transmission System and Model:-

The most common reason for choosing HVDC over AC transmission is that it is more
economical than AC for transmitting large amount of power point to point over long
distances. A long distance, high power HVDC transmission scheme generally has lower
capital cost and lower losses than an AC transmission link. Even though HVDC conversion
equipment at the terminal station is costly, overall savings in capital cost may arise because
of significantly reduced transmission line costs over long distance routes. HVDC needs fewer
conductors than an AC line, as there is no need to support three phases. Also, thinner
conductors can be used since does not suffer from the skin effect. Depending on voltage
levels and construction details, HVDC transmission losses are quoted as about 3% per 1000
km, which is less than typical losses in an AC transmission system. Changes in load that
would cause portions of an AC network to become unsynchronized and separate would not
similarly affect a DC link, and power flow through DC link would tend to stabilize the AC
network. The magnitude and direction of power flow through DC link can be directly
commanded, and changed as needed to support the AC networks at either end of the DC link
this has caused many power system operators to contemplate wider use of HVDC technology
for its stability benefits alone. HVDC technology has become popular and widespread due to
the advantages it provides over AC transmission. Advantages of HVDC over AC
transmission include lower net cost for long distance, greater power per conductor, and
possible use of ground as a return path, hence each conductor can be operated as an
independent circuit. Besides that, it doesn’t require any charging current, no skin effect;
hence the line doesn’t require reactive compensation, elimination of system synchronization
and simpler transmission towers.

A HVDC transmission lines are mainly overhead transmission lines lay over long distances.
The existence of fault on the transmission line may lead to the tripping of the transmission
line. As the HVDC is a bulk power transmission system the short duration faults may lead to
the total block out of the region. Therefore, reliability of HVDC systems has always been of
primary concern in planning and operation of power systems. The total reliability of HVDC

5
system depends on the components in the line and at the station. The HVDC system model is
designed for 12-pulse 1000MW, with the help of MATLAB software. A 1000 MW (500 kV,
2kA) DC inter connection is used to transmit power from a 500 kV, 5000 MVA, 60 Hz
network to a 345 kV, 10,000 MVA, 50 Hz network. The rectifier and the inverter are 12-
pulse converters using two 6-pulse thyristor bridges connected in series. At both the ends, at
rectifier end as well as at inverter end AC filters are installed. Filters are installed to filter out
the harmonics also to provide reactive power to the converter. Rectifier and inverter are
connected via two 0.5H smoothing reactor. The identification of faults in HVDC converter
will be done using back propagation training algorithm. Here we develop an M-file program
in MATLAB software using back propagation algorithm to identify the faults that may occur
in converter such as valve short circuit and arc through.

Fig2.1:- HVDC transmission model

Direct voltage and direct current that results after simulation of the HVDC system model are
shown in figure 2.2

Fig2.2:-Simulation model of HVDC transmission system

6
2.2 Converter Faults in HVDC System:-

HVDC system consists of rectifier at sending end and inverter at receiving end. Converters
are connected with a DC transmission line represented by a resistance and DC inductor on
either side of the line.

Fig2.3. HVDC system model

There are three basic types of faults that can occur at converter station:

1. Faults due to malfunctions of valves and controllers:-

(i) Arc backs (or back fire)

(ii) Arc trough (Fire through)

(iii) Misfire

(iv) Quenching or current extinction

2. Commutation Failures in inverters

3. Short circuits in a converter Station

The arc back is the failure of the valve to block in the reverse direction and result in the
temporary destruction of the rectifying property of the valve due to conduction the reverse
direction. This is a major fault in mercury arc valve and is of random nature. This is non self
clearing fault and result in severe stresses on transformer windings as the incidence of arc
backs is common. Fortunately, thyristor don’t suffer from arc back which has led to the
exclusion of mercury arc valves from modern converter stations.

2.2.1. Commutation Failure:-

This type of faults occurs in thyristor as they required a definite turn – off time so there is a
need to maintain a minimum value of extinction angle defined by

𝛾 = 180 − 𝛼 − 𝛽 (… 2.1)

7
Where the overlap angle (𝛽) is a function of the commutation voltage and the DC current.
The reduction in the voltage or increase in current or both can result in an increase in the
overlap angle which can result in 𝛾<𝛾𝑚𝑖𝑛. This give rise to commutation failure.

Fig2.4:- Thyristor based rectifier

In the fig.2.2 If the current in the incoming valve (say valve Q3) which starts conduct after
valve Q1 will diminish to zero and the outgoing valve (valve Q1) will be left carrying the full
sink current. As the firing of next valve in sequence is of valve Q4 this will result in a short
circuit of the bridges as both the valve Q1and Q4of the same arm will conduct. If the
conduction from valve Q2 to Q4 is successful, only valve Q1 and Q4 are left conducting and
this state continues until the valve Q6 is fired. The firing of valve Q6 is unsuccessful as the
valve Q5 is reverse biased at the time of firing. If the commutation from valve Q4 to Q6 is
successful, the conduction pattern returns to normal except the bridge voltages is negative at
the instant where Q4 ceases conduction. If the causes which led to commutation failure in
valve Q1 in the first instant have disappeared, the bridge operation returns to the normal state.
Thus, a single commutation failure is said to be self clearing. The wave form of the bridge
voltage and valve voltage are shown in fig.2.3.

Fig.2.5:- Bridge voltage waveform for a double commutation failure

The following are the effects of a single commutation failure:-

1. The bridge voltage remains zero for a period exceeding 1/3 of a cycle, during which the
DC current tends to increase.

8
2. There is no AC current for the period in which the two valves in an arm left conducting.

The recovery from a commutation failure depends on the following factors:-

1. The response of the gamma controller at the inverter

2. The current control in the link

3. The magnitude of the AC voltage

2.2.2. Arc Through:-

This is a fault likely to occur at the inverter station, where the valve voltages are positive
most of the time (when they are not conducting). A malfunction in the gate pulse generator or
the arrival of spurious pulse can fire a valve which is not supposed to conduct, but is forward
biased. For such fault, the firing delay angle of the faulted valve is reduced from its normal
value to a smaller value or zero. For example in a bridge, when valve 1 has successfully
commutated its current valve 3, then initial current across it is a negative (for the duration of
the extinction angle) and then become positive. If the valve 1 is fired at this time, the current
will transfer back to valve 1 from valve 3. The effect of an arc through are similar to that of
commutation failure – the voltage across the bridge falls as valve 4 is fired (with valve 1 is
conducting) and the AC current goes to zero when valve 2 current goes to zero. The firing of
valve 5 is unsuccessful and the bridge recovers to its normal operation after valve 6 is fired
and the subsequent firings are according to the normal sequence. Such fault also introduces a
significant increase in the harmonic contents of the turbine-generator shaft torsional torques.

2.2.3. Misfire:-

Misfire occurs when the required gate pulse is missing and the incoming valve is unable to
fire. The probability of the occurrence of misfire is very small in modern converter stations
because of duplicate converter controls, monitoring and protective firing of valves. While
misfire can occur in rectifier or inverter stations, the effects are more severe in the latter case.
This is due to the fact that in inverters, persistent misfire leads to the average bridge voltage
going to zero, while an AC voltage is injected in to the link. This result in large current and
voltage oscillations in the DC link as it presents a lightly damped oscillatory circuit viewed
from the converter. The DC current may even extinguish and result in large overvoltages
across the valves. The waveform of the DC voltage and current in the link of persistent
misfire in the inverter. Also such fault introduces significant distortion to the electromagnetic

9
torque.The effects of a single misfire are similar to those of commutation failure and arc
through. When valve 3 in a bridge misfires, the valves 1 and 2 are still conducting until valve
4 is fired. However, at the end of cycle the normal sequence of valve firing is restored. Thus a
single misfire is also self clearing.

It is clear from graph shown in fig 2.4 and 2.5 that output voltage of an inverter reduces
during misfire fault. A misfire is done on valve 1 for the time interval T = 0.02 to T =0.08 sec
to obtain the above shown result. After removal of misfire fault voltage become normal.

Fig2.6:- Normal output line to line voltage of an inverter

Fig2.7:- Line to line output voltage of an inverter during misfire fault

2.2.4. Current Extinction:-

The extinction of current can occur in a valve if the current through it falls below the holding
current. This can arise at low value of the bridge currents when any transient can lead to

10
current extinction. The current extinction can result in overvoltages across the valve due to
current chopping in an oscillatory circuit formed by the smoothing reactor and the DC line
capacitance.

The problem of current extinction is more severe in the case of short pulse firing method.
However in modern converter stations, the return pulses coming from thyristors levels to the
valve group control, indicate the build up of voltage across the thyristors and initiate fresh
firing pulses when the valve is supposed to be conducting. It may happen that a number of
firing pulses may be generated during a cycle when then current link is low.

2.2.5. Short Circuit in a Bridge:-

This fault also has very low probability as the valves are kept in a valve hall with air
conditioning. However, bushing flashover can lead to a short circuit across the bridge and
produce large peak currents in the valve that are conducting. The short circuit currents are
significant only in rectifier bridges. The worst case is when the short circuit occurs at the
instant of firing a valve at α = 0o. Assuming that there is no inductance in series with the
bridge, the peak short circuit current (ipeak) is given by

𝑝
𝑖𝑝𝑒𝑎𝑘 (1+sin 𝜋/𝑝) ( )−1 𝑖𝑑𝑜
3
=3∗ 𝜋 + 𝑝 ∗( ) …eq. (2.2)
𝐼𝑠 𝑝 sin (3) 𝐼𝑠
𝑝

Where

p = pulse number of the converter (6 or 12)

ido = the dc current at the instant of firing the valve

Is=√2 ELL / 2 Xc

For a six pulse converter, the peak current is

Ipeak= ½ [3Is+ido]

The bridge voltage and current waveform are shown, the effect of network impedance in
limiting the current is neglected. The maximum peak current in a valve results when it is
conducting in to a valve fault.

11
 Fig2.8:-Line to line voltage of inverter during fault

2.3. Transmission Line and Asymmetric Faults in HVDC System:

Transmission lines are used for long distance power supply. Transmission lines, when
interconnected with each other, become transmission networks. High-voltage overhead
conductors are not covered by insulation. The conductor material is nearly always an
aluminum alloy, made into several strands and possibly reinforced with steel strands. Copper
was sometimes used for overhead transmission but aluminum is lighter, yields only
marginally reduced performance, and costs much less. Overhead conductors are a commodity
supplied by several companies worldwide. Improved conductor material and shapes are
regularly used to allow increased capacity and modernize transmission circuits. High-voltage
direct current (HVDC) is used to transmit. In an electric power system, a fault is any
abnormal electric current. For example, a short circuit is a fault in which current bypasses the
normal load. An open-circuit fault occurs if a circuit is interrupted by some failure. In three-
phase systems, a fault may involve one or more phases and ground, or may occur only
between large amounts of power over long distances or for interconnections between
asynchronous grids. The transmission line model, three phase source and three phase load is
simulated using MATLAB/Simulink powergui. Sampling is done to reduce complexity and
increase simulation speed. phases. In a "ground fault" or "earth fault", charge flows into the
earth. The prospective short circuit current of a fault can be calculated for power systems. In
power systems, protective devices detect fault conditions and operate circuit breakers and
other devices to limit the loss of service due to a failure.

12
An asymmetric or unbalanced fault does not affect each of the three phases equally. There are
three common types of asymmetric faults, namely line to line, line to ground and double line
to ground. Line-to-line fault is a short circuit between lines, caused by ionization of air, or
when lines come into physical contact, for example due to a broken insulator. Line-to-ground
- is a short circuit between one line and ground, very often caused by physical contact, for
example due to lightning or other storm damage. Double line-to-ground fault is where two
lines come into contact with the ground (and each other), also commonly due to storm
damage. Single line to ground and double line to ground fault is simulated using three phase
fault block. It is capable of introducing short circuit fault at the different interval of time. It
can be visualized using the waveform of the input and output current and voltages through
scope.

2.4. HVDC Transmission Side Fault Detection:-

We can use different techniques for fault identification like wavelet transform, artificial
neural network, fuzzy logic, genetic algorithm, expert system etc. First of all we take wavelet
transform method for detection of fault and artificial neural network for classification of
faults.

The neural network is provided with six inputs during the fault detection process. The inputs
are three voltages of respective three phases and three currents of the respective three phases.
The value of input voltages and input currents are normalized with respect to the pre-fault
values of the voltages and currents respectively. The output of the neural network is in simple
yes or no form, i.e. 1 or 0, which indicates whether the fault has been occurred or not.

The 12 pulse HVDC system model is studied for various faults like (a) DC line short circuit,
(b) short circuit on the AC side of inverter station and (c) normal operation as reference case.
The short circuit faults on the AC side of the inverter are Single Line to Ground fault (LG
fault), Line to Line fault (LL fault), Triple line (symmetrical) fault on the AC side of the
inverter (LLL fault). It is not easy to identify the faults and to make correct protection
decision within 3-5ms by using traditional methods, as the fault generated travelling waves,
will be similar in nature. It is even more difficult if there is noise. If the converter fault is
considered then Fault detection of HVDC converter is based on the fact that every operation
of the converter valve is associated with a set of conduction pattern of the valves. Fault
detection is basically treated as a problem of pattern recognition and neural networks are
extensively used where pattern recognition is needed. Therefore, by integrating the neural

13
network algorithm with HVDC system model the faults in converter can be detected. The
faults such as valve short circuit, misfire and arc through faults are discussed in this paper.
With the help of ANN and wavelet transformation, the fault is easily classified and detected
respectively. Wavelet analysis expands functions not in terms of trigonometric polynomials
but in terms of wavelets, which are generated in the form of translations and dilations of a
fixed function called the mother wavelet. Compared with Fourier transform, wavelet can
obtain both time and frequency information of signal, while only frequency information can
be obtained by Fourier transform. Each wavelet is created by scaling and translation
operations in a special function called mother wavelet. A mother wavelet is a function that
oscillates, has finite energy and zero mean value. When fault occurs on a DC line, the
travelling waves propagate along the line and are reflected at discontinuous points of surge
impedance. This leads to an abrupt change in the voltage and current and hence in the reverse
voltage travelling wave. The sudden changes are the edges in signal processing. The wavelet
theory is used to detect the sudden changes, and the fault location can be identified along
with the recognition of the fault by obtaining the time delay Δt in seconds between the two
absolute maximum values of the wavelet coefficients. By denoting

…eq.(2.3)

Where,

V= is the velocity of travelling wave in km/s and

L =is the fault distance in km from the measuring point.

The time delay between the first two maximum wavelet coefficients of the reverse voltage
travelling wave is Δt = 0.00102 seconds, thus the fault location from the measuring terminal
can be calculated and is equal to 151 km. One of the fault commutation failures can be
identified clearly from the HVDC line fault with the polarity change of wavelet coefficient.

2.5. Converter Side Fault Detection Schemes:-

Fault detection of HVDC converter is based on the fact that every operation of the converter
valve is associated with a set of conduction pattern of the valves. Fault detection is basically
treated as a problem of pattern recognition and neural networks are extensively used where
pattern recognition is needed. Therefore, by integrating the neural network algorithm with

14
HVDC system model the faults in converter can be detected. The faults such as valve short
circuit, misfire and arc through faults are occur in converter side of HVDC transmission.

2.5.1. Valve Short Circuit Detection:-

When valve short circuit occurs high amplitude current developed on secondary of converter
transformer which is conducted by a faulty valve and thereby a healthy valve which is
attempted to commutate. This is avoided by fast detecting the fault and order blocking
without bypassing pair before the next valve is fired. The valve short circuit is detected when
the dc current goes to zero and ac current tends to increase. Here the valve short circuit fault
is created in a six pulse converter bridge using breaker across valve 1 shown in figure2.9

Fig2.9:- Valve short circuit fault

For detecting all these three faults an M-file program is developed using back propagation
training algorithm. After execution of this program we get the results for valve short circuit
fault which are shown below:

Fig2.10:- Valve current during short circuit

2.5.2. Misfire Fault Detection:-

A misfire is when a valve fails to take over the conduction at its programmed instant although
its voltage has the correct polarity. For a single valve misfire fault, the normal valve
conduction sequence of the rectifier will be disturbed once in a cycle and reappear in every
cycle if the fault is sustained. As a result, a harmonic current in rectifier side of the system
frequency will be induced on dc line. Misfire fault created in a six pulse converter bridge is
shown in figure2.11

15
 Fig2.11:- Misfire fault

Fig2.12:-Valve current during misfire fault

16
 Normal conduction sequence

Faulty condu

ction sequence

Fig2.13:- Normal and faulty conduction sequence

2.5.3 Arc through Fault Detection:-

Arc through fault occur when the valve fails to block the conduction at its programmed
instant i.e. the valve is continuously is in ON state. Due to this reason the conduction
sequence will be disturbed. Arc through fault in six pulse bridge is shown in figure2.14.
Single arc through is self clearing if the causes led to it are removed. In case of persistent arc
through fault differential protection scheme is used.

The basic protection against all the converter faults is provided by valve group differential
protection, which compares the rectified current on the valve side of converter transformer to
the dc side current measured on the line side of the smoothing reactor. The differential
protection is employed because of its selectivity and fast detection. The overcurrent circuit is
used as a backup. The level of overcurrent required to trip must be set higher than that of the
valve group differential protection to avoid tripping with faults outside the station.

17
 Fig2.14:- Arc through fault

Fig2.15:- Arc through rectifier

18
2.6. Fault Detection Techniques in HVDC System:-

There are different techniques which are used in fault detection in HVDC system:-

2.6.1. Travelling Wave Phenomenon:-

Travelling wave fault location methods are usually more suitable for application to long lines.
Travelling wave methods for transmission lines fault location have been reported since a long
time. It is well known that when a fault occurs in overhead transmission lines systems, the
abrupt changes in voltage and current takes place, at the point of the fault. When a fault
occurs on the HVDC line, according to travelling wave theory, voltage and current travelling
waves transmits on the line. During the occurrence of a fault, travelling waves carry the
information about the fault which can be used for fault detection and line protection. If the
times of arrival of the travelling waves in the two ends of the transmission line can be
measured precisely, the fault location then can be determined by comparing the difference
between these two arrival times of the first consecutive peaks of the travelling wave signal.
Travelling-wave-based line fault location principle has been successfully applied to
transmission line fault location in the conventional HVDC systems with two terminals.

2.6.2. Wavelet Transform:-

A wavelet is a waveform of effectively limited duration that has an average value of zero.
Wavelet transform has received greater attention in fault analysis due to its ability in
analyzing the travelling waves than conventional methods of analysis. The method using
wavelet transforms for detecting a HVDC Transmission line faults is proposed after
simulating the HVDC system for various faults. The simulation results show that the
application of wavelet technique leads to a more reliable solution for recognition of faults and
provides a good basis for the new protection scheme for the HVDC lines. Wavelet analysis is
a relatively new signal processing tool and is applied recently by many researchers in power
system due to its strong capability of time and frequency domain analysis. The definition of
continuous wavelet transform (CWT) for a given signal x(t)with respect to a mother wavelet
(t) is,

…eq(2.4)

19
Where a, is the scale factor and b is the translation factor. For CWT, t, a, b are all continuous.
Unlike Fourier transform, the wavelet transform requires selection of a mother wavelet for
different applications. One of the most popular mother wavelets for power system transient
analysis found in the literature is Daubechies‟s wavelet family. In the new scheme, the db5
wavelet is selected as the mother wavelet for detecting the short duration, fast decaying fault
generated transient signals.

The application of wavelet transform in engineering areas usually requires discrete wavelet
transform (DWT), which implies the discrete form of t, a, b eqn. The representation of DWT
can be written as:

…eq.(2.5)

Where original a and b parameters in equation are changed to be 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 Fig. below. 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 2to obtain the detail coefficients and the
approximation coefficients at level 1 (D1 andA1). The approximation coefficients are then
sent to the second stage to repeat the procedure. Finally, the signal is decomposed at the
expected level.

2.6.3. Distance Protection Method:-

The proposed method is based on the frequency dependent nature of the line parameters.
Hence, by considering frequency dependent parameter line model transformation matrix is
separated into two parts (1) Constant parameter model known as distributed parameter model
(2) Compensation matrix consisting of frequency dependent parameters. Determination of
fault location is necessary to distinguish whether the fault is internal or external to protection
zone, especially for the end zone faults. Therefore, in distance protection some degree of
measurement error is acceptable as relay will operate only when the fault lies in the
protection zone and if the measured fault distance is not more than the setting distance. The
condition for protection method to operate correctly is that the measurement error in the
location of fault should be less than the difference between the fault distance and setting
distance that is given as,

20
 Em < |lf – lset | …eq(2.6)

where, Em is the measurement error, If and Iset is the fault distance and the setting distance
respectively. This method is based on the accuracy of calculation of setting point voltage and
current using frequency dependent parameter model. It reduces the error in the measurement
of distance to a greater extent for far end faults.

2.6.4. Discrete Fourier Transform Method:-

DFT Algorithm The most commonly used method of calculating phasors from sampled data
is that of Discrete Fourier Transform (DFT). It is performed by the implementation of
following equation,

...eq(2.7)

In this equation, ‘X’ is the phasor of a signal, ‘k’ is the order of harmonic, ‘x’ is the
instantaneous value of sinusoidal signal such as voltage and current, ‘n’ is the nth sample of
data window and ‘N’ is the number of samples in data window. Recursive implementation of
DFT is a more computationally efficient method which computes the estimated phasor
recursively by adding the contribution made by the new sample, and subtracting the
contribution made by the oldest sample. Phasor Measurement Units, which are being
deployed on power systems as a measurement tool in many countries, use recursive DFT
algorithm.

2.6.5. Independent Component Method:-

Independent component analysis is a kind of high efficient blind source separation method
that is developed gradually in recent years. The independent component analysis has been
widely used in the feature extraction and the speech recognition aspects due to the little
requirement of the environment and the target. we can use Fast Independent Component
analysis to blind separate the DC voltage and DC current signal that are measured by multi-
channels so that the system fault source signal is restored, and the key feature of this method
can effectively extract the fault. Due to Independent component analysis algorithm, we can
process the current signal of DC transmission lines after fault, then decompose the current
characteristic signal, and finally detect the time of the initial Bob and the second arrive Bob
measurement point and polarity relationship, so we can realize fault location. The constraints

21
of an independent component method mainly require that source signals are independent each
other, and there is a Gaussian signal at most.

2.6.6. Natural Frequency Method:-

Since DC transmission line is only a route, and thus we can extract a natural frequency
method to realize fault locations. It proposes that there is a mathematical relationship between
the fault travelling wave spectrum of transmission line and the fault distance, and a fault
travelling wave spectrum can be used to realize a fault location. Based on travelling wave
natural frequency range of the principle component, this method isn’t restricted by the Bob
recognition; using transient voltage information after the line occurred in fault, through the
spectrum analysis, we can get the natural frequency of the travelling wave principle
component, and we can realize a fault location. Using the data on both ends of the line, the
Pronny algorithm is utilized to extract the wave natural frequency; we can’t use line
parameters to calculate the wave velocity, but we can realize a precise fault location.
Integration testing model decomposition is used to extract the high frequency component of
the wave, identify the Bob and extract the precise time parameter, and finally solve the
problem of wave velocity selection.

2.6.7. Artificial Neural Network:-

The reach accuracy of an electromechanical, static or a microprocessor based distance relay is

affected by different fault conditions and network configuration changes. So ANN techniques
are under investigation over the past 15-20 years, which can adapt dynamically to the system
operating conditions at a high speed. The ability of ANN to learn by training any complex
input/output mapping and recognize the noisy patterns gives them the powerful property of
pattern recognition and classification. ANNs can solve the overreach and the under reach
problems which are very common in the conventional distance relay design. ANN utilizes
samples of currents and voltages directly as inputs without computation of phasors and
related symmetrical components. Various kinds of neural network such as multi-layer
perceptron (MLP), recurrent, radial basis function (RBF), probabilistic neural network etc.
are being applied for fault classification and fault location. These are designed by different
training algorithms like back propagation, orthogonal least square, extended kalman filter etc.
The use of ANNs can extend the first zone of distance relays and enhance system security.
For selecting the appropriate network configurations, the performance criteria are fault
tolerance, minimal response time and generalization capabilities. ANN approach has been

22
used to improve some of the standard functions used in protection of transmission lines. They
have been related to fault direction discrimination, fault detection and classification, distance
protection, improvements in fault distance computation, protection of series compensated
lines, adaptive distance protection and adaptive reclosing.

To make the ANN responsive to time varying voltage and current waveforms different types
of recurrent networks were considered that allow the hidden units of the network to see their
own previous output, so that the subsequent behavior can be shaped by previous response.
Such an Elman recurrent network designed to act as the fault direction detection module of a
transmission line. Further development is the concept of supervised clustering to reduce the
number of iterations in the learning process of multi layer feed forward networks. It is
observed that the ANN based distance relays need much larger training sets and hence the
training of these networks is time consuming and further research work shall produce a
hardware realization with proper modification in the learning methodology and preprocessing
of input data that would improve the learning rate performance, efficiency and the reliability
many folds. Presently research efforts are in the direction of evolutionary computational
techniques such as genetic algorithms (GA) for determining the neural network weights and
thereby avoid training of ANN.

2.6.8. Fuzzy Logic:-

Zadeh introduced the concept of fuzzy set theory in 1965 for dealing with uncertain and
ambiguous properties of events (Zadeh, 1965). It was introduced in power system networks
to solve uncertainty problems that arise due to the continuously varying power system
parameters. The key benefit of fuzzy logic is that its knowledge representation is explicit,
using simple “IFTHEN” relations. The fuzzy set theory is used for fault type identification on
a transmission line. These algorithms are fairly accurate only under certain assumptions of
fault distance, prefault power flow, fault resistance and line length [6]. Design of a fuzzy
logic controller does not need an accurate mathematical model of the system under
consideration. A qualitative knowledge about the system behavior is adequate to design a
fuzzy logic controller to achieve a desired control objective. In addition, it is easy to add
expert / heuristic knowledge about the system behavior in the controller structure. Moreover,
the performance of a fuzzy logic controller is not significantly affected due to changes in
system operating conditions and parameters. The output of Fuzzy Logic Controller can be
utilized to modulate the power order of the DC control, which in turn modulates the DC

23
power. The stabilizing control is implemented through large signal modulation of power in
response to a control signal derived from the AC system variables. The effectiveness of the
control can be enhanced by increased overload rating of the converters which permit short-
term overloads.

In fuzzy logic based protection system, accuracy cannot be guaranteed for wide variations in
system conditions. So consequently a more dependable and secure relaying algorithm during
real time implementation is needed for classifying the faults under a variety of time-varying
network configurations. The fuzzy-neuro approaches aresensitive to system frequency
changes and require large training sets and a large number of neurons affecting their accuracy
and speed in protecting large power networks.

24
 Chapter3

FAULT IDENTIFICATION OF HVDC CONVERTER USING

ARTIFICIAL NEURAL NETWORK

3.1. Introduction of Artificial Neural Network:-

An Artificial Neural Network (ANN) is an information processing paradigm that is inspired

by the way biological nervous system process information. It is composed of large number of
highly interconnected processing elements called neurons working in unison to solve specific
problems. This highly connected, parallel interface forms the hardware of the ANN. One of
the major advantages associated with the use of neural network is the parallel processing
capability. The high degree of parallel connectivity of an artificial neural network also brings
about desirable properties such as noise reduction and fault tolerance [8]. The processing
units or neurons of ANN consist of three main components; synaptic weights connecting the
nodes, the summation function within the node and the transfer function. Synaptic weights
characterize themselves with their strength, which corresponds to the importance of the
information coming from each neuron. In other words, the information is encoded in these
strength-weights. The summation function is used to calculate a total input signal by
multiplying their synaptic weights and summing up all the products. The artificial neuron
model is shown in fig.3.1

Fig3.1:-Neuron model

ANN based methods do not require a knowledge base for the detection of faults unlike the
other artificial intelligence based methods. The prime motive behind this work is that a very
accurate fault detector could make if employed in a power transmission and distribution

25
system, in terms of the amount of money and time that can be saved. The main goal of Fault
Detection is to detect a fault in the power system with the highest practically achievable
accuracy. When the physical dimensions and the size of the transmission lines are considered,
the accuracy with which the designed fault detector detects faults in the power system
becomes very important.

As power system grow both in size and complexity, it becomes necessary to identify different
faults faster and more accurately using more powerful algorithms. Back propagation
algorithm of neural network is used for the fault diagnosis system. Different fault like single
line to ground fault and double line to ground faults are detected using ANN. Fault detector
module of a transmission line protective scheme can be used to start other relaying modules.

ANNs are non linear information (signal) processing devices, which are built from
interconnected elementary processing devices are called neurons. The arrangement of
neurons into layers & the pattern of connection within & in-between layer are generally
called as the architecture of the net. ANNs have been used for the protection of power
transmission lines. The excellent pattern recognition and classification abilities of neural
networks have been cleverly utilized to address the issue of transmission line fault location.
Artificial neural network (ANN) can be applied to fault detection and classification
effectively because it is a programming technique, capable to solve the non linear problems
easily. They are widely accepted and used in the problem of fault detection and fault
classification because of the following features:

ANN is made of massive interconnection of elementary processing units, information

processing can be carried out in a parallel distributed manner. This makes real time
processing of large volumes of data more readily realizable.

ANN can model any degree of nonlinearity and thus are useful in solving these problems
which are inherently nonlinear.

ANN approach is non-algorithmic and requires no prior knowledge functions relating the
problem variables. Also being non algorithmic, they do not make any approximations unlike
as the case with most of the mathematical models.

ANN is capable of handling situations of incomplete information, corrupt data and thus is
highly fault tolerant.

26
Each neuron could receive several inputs from neighboring neurons through interconnections.
The net input of the neuron is the weighted sum of inputs into the neuron and then neuron
uses net input, together with information on its current activation state to determine its new
state of activation. The activation function is used to calculate the output response of a
neuron. Neural network computing characteristics are distinguished from conventional
pattern recognition by their ability to map complex and highly non linear input-output
patterns. A neural network can be trained to perform a pattern association or classification. It
can respond to input data and provide output that can provide quantitative information about
the system.

3.2. Architecture of Artificial Neural Network:-

The fault location tool in the present work is radial basis neural network. We have references
for modeling our system. Fault diagnosis of HVDC with pre-classifier is discussed in [25].
We closely followed the use of RBNN from [25-27]. Comparison of BPNN and RBNN is
discussed in [27] which motivated us for using RBNN in the present work. The RBNN has
less execution time, fastest creation of network object etc. We followed the creation of
network with training parameters, in [28]. The concept of zoning in the present work ,is
similar to the approach in [28]. Other works referred have different application background
[31] but the way the neural network configured is same [28-31]. RBNN is based on
supervised learning. The network consists of input layer, hidden layer and output layer. The
input layer feeds the input data, to the hidden layer. In the present work, the input data is the
processed wavelet sample from HVDC system which will be discussed in next section. The
hidden layer works by neurons loaded with radial basis activation function. The output layer
is powered with neurons of linear activation function. The architecture of RBNN shown in
Fig3.2

Fig3.2:- Structure of radial basis neural network

27
The network inputs K input nodes and m hidden neurons. For a K dimension W’ wavelet data
the network computes the scalar value Z.

Where W0 is the weight bias, Wi is the weight parameter Di is the Radial basis function. The
radial basis function for input is .The plot of the radial basis function is Gaussian
double sided bell curve. In the present work the network iteratively creates radial basis
network one neuron at a time. Neurons are added to the network until the sum-squared error
falls beneath an error goal or a maximum number of neurons have been reached. The above
network is created by using neural network tool box in MATLAB.

3.3. Fault Identification using Artificial Neural Network:-

Artificial intelligence, cognitive modeling, and neural networks are information processing
paradigms inspired by the way biological neural systems process data. Artificial intelligence
and cognitive modeling try to simulate some properties of biological neural networks.
Artificial neural networks have been applied successfully to speech recognition, image
analysis and adaptive control, in order to construct software agents (in computer and video
games) or autonomous robots and specially in fault detection system. Neural network theory
has served both to better identify how the neurons in the brain function and to provide the
basis for efforts to create artificial intelligence.

HVDC converter faults are mainly depends upon the operation of the converter valve and
conduction pattern of the valve. Fault identification is based on valve parameter. Fault
identification method is treated as a problem of input data pattern recognition and pattern
recognition is well handled by artificial neural networks.

Neuron consists of three basic components, namely weights, thresholds/biases and a single
activation function. Values w1, w2……wn are weights to determine the strength of input
vector X = [x1, x2,….., xn]T. Each input is multiplied with its associated weight of the
neuron XT.W.

…eq. (3.1)

28
 Fig3.3:-Block diagram for fault identification in transmission line using ANN

Threshold, φ is the neuron‟s internal offset. The neuron fire or produces positive output only
if the total input I is above the threshold value. It affects the activation of the node output y as

…eq.(3.2)

To generate the final output Y, the sum is passed on to a non- linear filter f called activation
function or transfer function, which releases the output Y. There are various activation
functions which are chosen depending on the type of problem to be solved by the network.
The most common activation function includes linear function, tangent hyperbolic function,
threshold function and sigmoidal function. Most popular sigmoidal function follows the
transition equation shown below.

…eq.(3.3)

Where 𝛼 is the slope of sigmoid function.

The capability of the neural network increases as number of neurons increases. This
capability multiplies as number of layers in neural network structure increases. Figure 3
shows multilayer neural network with one hidden layer. The weights connecting neurons are
varied continuously while training the neural network. In NN applications, the challenge is to

29
find the right values for the weights and the threshold. Various algorithms are developed in
neural network field depending on different problems and applications where it has been
used. Back Propagation, Radial Basis Functions, Multi-Layer Perceptron algorithm, Adaptive
Resonance Theory (ART), Self Organizing Maps (SOM) and Counter Propagation Networks
(CPN) are few algorithms of neural networks.

Fig3.4:- A single neuron and sigmoidal activation function

Various neural network architecture, training algorithms and transfer functions were studied
to decide upon the final neural network model for fault detection system. Back propagation
algorithm, a supervised learning is used as the network will be trained using the data created
from the simulation model of transmission. Number of hidden layers and number of neurons
in each layer can be varied depending on the complexity of the problem to be solved. The
training data was created using Simulink model of 300 km long transmission line for various
fault cases. This training data consists of various vectors of the output and input current and
voltage. Using these values and the created target vector, neural network is trained to detect
the fault at the different instants of time. Training is done using back propagation training
algorithm using “traingd” function of neural network toolbox of MATLAB. The
trained/updated neural network model for fault detection is converted to Simulink block using

30
neural network toolbox. This is done to make it usable with Simulink model of transmission
line.

3.4. Fault Detection in HVDC System using ANN in MATLAB:-

In MATLAB there is a fixed procedure to make simulation model of artificial neural

network:-

3.4.1. Training of Neural Network:-

The neural network architecture with three hidden layers was fixed for the simulation
purpose. Hidden layers consist of 7-10-3 neurons. Input layer consists of three neurons which
takes the current of three phases of the transmission line. Output layer consists of one neuron
which indicates if fault is there. A threshold of 0.9 is fixed based on simulation in order to
avoid false alarms. As shown in figure, the training algorithm used is gradient descent variant
of back propagation algorithm. Figure shows the number of iteration, time taken to train the
Neural Network. It also shows the different performance plots of the Neural Network after
training of the network. The different training parameter encountered during the training
process is gradient, mean square error and validation check. For better training purpose, mean
square error value should decrease when the training process continues; validation check
gives us maximum number of fails in the Neural Network training process. These training
parameter plots are shown in Fig.3.5

Fig.3.6:- Training of neural network

31
3.4.2. Performance Plot of Trained Neural Network:-

The training data is created using simulation model of transmission lines. Single line to
ground and double line to ground fault was introduced in different phases at different time
instants. This training data, when fed for training the neural network, is divided into three
parts. The data samples are divided into train, test and validation sets randomly by the
program, in the ratio 0.7:0.15:0.15. The figure 3.8 shows the performance of Neural Network
when Neural Network is trained with particular input vectors and target vectors. The plot
shows that the variation of different parameters throughout the training process. After
training is completed satisfactorily, the neural network is tested with the available sample
data to evaluate the performance of the updated trained neural network. If the performance is
not up to the expectations, some variations may me experimented with the neural network
structure by varying number of hidden layers of number of neurons in each layer. Different
training algorithm may also be experimented with various activation functions.

Fig.3.7:- Training parameters plot

Fig.3.8:- Performance plot

Once the neural network with satisfactory performance is developed, its Simulink model is
developed using „gensim‟ function so that it can be used with other block in the simulation

32
model. Neural network simulation model is shown in Fig.3.9. The complete simulation set up
is shown in Fig.3.10.

Fig.3.9:- Neural network simulation model

3.4.3. Simulation Set-up for Trained Neural Network:-

As discussed in previous section, transmission line with three phase source and load, three
phase fault and neural network based fault detection simulation models are developed to test
the capability of the proposed method. Transmission line model is generated with discrete
state and pi section line. Using this model samples are reduced from 1000 samples to 50
samples, which is feasible for real time application and easy for implementation. Sampling
time of this model is Ts= 0.02ms and simulation is done for 1 sec. With these specifications,
the current and voltage vectors consist of 51 samples. This simulation model of transmission
line with various fault cases is used to create training data which was used to train neural
network. A feed forward neural network with 3 hidden layers is used for fault detection
purpose. The number of neurons in all hidden layer can be varied to get the optimum
performance. First hidden layer is fixed with 7 neurons, second with 10 neurons and last
hidden layer with 3 neurons as shown in Fig. 3. Output layer has one neuron, output indicates
whether fault has occurred or not at different instants of time. This neural network structure is
trained using back propagation training algorithm for the training data of transmission line
model. This trained/updated neural network is capable of detecting faults. Detection of fault
is successfully done for single line to ground and double line to line ground for all the three
phases. The fault detection system successfully detects single line and double line to ground
fault in transmission line.

33
3.4.4. Hardware Implementation:-

Fault detection system using neural network is implemented on DSP processor. Neural
network model parameters are studied and programmed using „C‟ language and the same is
implemented in TMS320C6713 using code composer studio. DSP processor is used to verify
the result for timing constraints. Implementation of Neural Network on DSP Processor
TMS320C6713 is difficult, because of the size of the generated Neural Network. To
overcome the difficulties faced while doing implementation, Neural Network was modified
as mentioned below:

 Changing the input from floating point to the binary form. For eg. Input in floating point
=0.04 is changed to the input in binary form= 0 and Input in floating point = 0.92 is
changed to the input in binary form=1.As detection of line to ground fault is short circuit
fault, input can be transformed to binary values.
 The size of the neural network is reduced, by changing the number of hidden layer to 1
and number of neurons in the hidden layer to 4. This neural network could give
satisfactory results for the transformed binary values.
 Implementation of this neural network with one hidden layer of 4 neurons is preferred for
the hardware, which contains 16 weight values and 5 biases. Totally 21 values are
required, which is easier for hardware implementation and able to detect faults
successfully.

Fig.3.10:- Simulink model of intelligent fault identification system in transmission line model

34
3.4.5. Result Analysis of Intelligent Fault Identification System with ANN:-

In this work, an attempt was made to develop an intelligent fault identification system for
transmission lines using artificial neural network. Three phase current and voltages are in
analog form and the waveform in its analog form is shown in Fig.3.11. It shows the
continuous signals for all the three phases of the voltage before sampling. However, to design
a fast and accurate fault detection system, it is important to sample it. The sampled waveform
at 0.2 s is shown in Fig. 3.12 for 1 s. This figure shows the discrete signals for all the three
phases of the voltages after sampling.

Fig.3.11:- Waveform before sampling Fig.3.12:- Waveform after sampling

Various cases of single line to ground and double line to ground fault is tested for the
developed fault detection system. Each figure has four sub plots. First three sub plots show
current levels at phase A, B and C. Fourth plot show neural network output, which indicates
the occurrence of fault by switching to a high level. Fault in one phase disturbs other phase
current as shown in all the result plots.

Fig 3.13:- Fault detection for single phase to ground fault

35
 Fig3.14:- Fault detection for double line to ground fault

Fig.3.15. Neural network output from DSP processor

Neural network is generalized and thus capable of finding any type of fault in the
transmission line. The trained Neural Network can detect Single line to ground and Double
line to ground fault. Effects of the different system parameters and conditions are studied.

36
 Chapter4

IMPORTANCE OF NEURAL NETWORK IN ELECTRICAL

ENGINEERING

4.1. Load Forecasting using Neural Networks:-

Several key features that distinguish NNs from other AI techniques are: learning by example
in real time, distributed memory and associated amount fault tolerance and graceful
degradation, real time pattern recognition, intelligent association. Despite its numerous
advantages, distributed memory causes a major flaw in NNs, because knowledge in a NN is
structured as a pattern of weights and connections. Other stumbling blocks and how training
work is going on to tackle them are discussed in the section of stumbling block. This section
aids with typical ANN application areas in power systems. The various areas are: - Security
assessment, transient stability, fault Detection and diagnosis, control analysis, protection and
design. Most popular problems are: (i) load Forecasting (ii) security assessment and (iii) Fault
detection & diagnosis.

Load forecasting is a suitable problem for ANN application due to the availability of
historical load data on the utility databases. ANN schemes using perceptron network and
SOFM have been successful in short-term as well as long-term load forecasting with
impressive accuracy. A combined use of Unsupervised and supervised learning was done for
short-term load forecasting. Filter based an" algorithm for faster convergence and improved
prediction accuracy. The RBF (radial basis function) network was found superior to MLP or
BP model in terms of training time and its accuracy. ANN does not need additional memory
for storing the history of load patterns. 1% improvement in accuracy of STLF can save up to
Rs. 700 million for a typical power utility.

4.2. Security Assessment:-

Static and dynamic security assessment often requires on-line computation. In order to
evaluate solution efficiently, the nonlinear mapping of MLP is utilized to reduce
computational burden and &deal with the Characteristics of power systems. This allows us to
carry out on-line monitoring/assessment in transient, small signal stability, and voltage
instability. Though Contingency ranking and sensitivity factor methods have reduced the

37
Number of critical contingencies to be computed, ANNs has played a challenging role in
security area.

The security assessment is based on:-

* Knowledge of system dynamics

* Measurement of system variables

* Models for measurement uncertainty and system disturbances

* Well-defined criterion for security.

Maximal admissible currents of the transmission give the boundaries of the secure Domain of
the state space (operating point are defined by a vector whose components are Active and
reactive power).

4.3. Fault Diagnosis:-

Fault detection/diagnosis is one of challenging problems in power systems. MLP identifies

the type and location of faults with a given set of power system conditions, measurements,
alarms, etc. KN (Kohonen net) is applied to handle the classification of fault patterns. The
diagnosis of the power apparatus is done to judge what kinds of faults the apparatus suffered
from and then accordingly it is cleared. KN is inferior to MLP in terms of the solution
accuracy due to unsupervised learning. RBF and BP models were developed for fault
diagnosis problem. The BP network had given superior performance while training of RBF
network was much faster as compared to BP network.

4.4. Economic Load Forecasting:-

The economic load dispatch (ELD) problem is one of the important optimization problems in
a power system we presented a method to solve ELD problem with piecewise quadratic cost
function using Hopfield NN. The ANN based approach turned out to be much simpler and
accurate. It deals with combined ED and emission dispatch using improved BPNN. Adaptive
Hopfield NN is recently used for ELD. Traditionally, in the ELD problem; the cost function
for each generator has been approximately represented by a single quadratic function. It is
more realistic, however, to represent the generation cost function for fossil fired plants
(thermal, nuclear etc) as a segmented piecewise quadratic function. Some generation units,
especially those units which are supplied with multiple fuel sources (gas and oil), are faced

38
with the problem of determining which the most economical fuel to burn. As fossil fuel costs
increase, it becomes even more important to have a good model for the production cost of
each generator. Therefore a more accurate formulation is obtained for the ELD problem by
expressing the generation cost function as a piecewise quadratic function. This approach can
be applied to generators supplied with various fuels as well as valve point loading problems.

4.5. World Scenario of Neural Network:-

Many organisations are carrying out commendable work in the area. The Neural Network
Group at the Austrian Research Institute for Artificial Intelligence, for example, has always
aimed at turning their research results into practical applications, be they industrial, medical
or otherwise. A similar approach needs to be adopted in India also. A list of major real-world
applications for which neural network solutions have been developed by the Institute are:-

 Prediction of rainfall based on radar images.

 Prediction of traffic flow on highways.
 Contributions to the control of rotary blood pumps.
 Identification of coronary artery disease from stress ECG.
 Recognition of hand-written characters.
 Alarm detection in cardiotocograms (CTG) for foetal monitoring.

International Business Machines (IBM) has integrated neural network and A1 technology
into software applications, such as data mining packages. Such techniques enable business
users to sift through vast quantities of raw data to spot hidden trends or anomalies which
might otherwise be missed. IBM's business intelligence and data mining software can mimic
human thought when processing information called the verification model-or go beyond
human logic to discover correlations and similarities between seemingly disparate data-called
the discovery model. Before neural networks and artificial intelligence were used in business
intelligence applications, companies lacked the ability to leverage information they were
collecting in data warehouses. They were unable to connect consumer buying patterns,
customize promotions to target a range of customers with different needs, or identify
potentially fraudulent insurance claims based on past behavior.

Neural computing and Al techniques also lie behind many of the sophisticated search engines
and intelligent agents found on the Internet and corporate intranets. Who can deny the fact
that in business today, the ability to access the right information at the right time is absolutely

39
critical! They are used in case-based reasoning tools such as those developed by Inference
(California, USA) for intelligent problem solving in the help desk, customer support and
other computer telephony integration packages. Other Al and neural intelligence systems are
used for credit scoring and to help pinpoint and prevent fraud in areas as diverse as equity
trading and cellular telephony networks.

4.5. Indian Scenario of Neural Network:-

Lot of opportunities exist in the country for Al technologies, especially neural computing
applications. Though most of the work is being done around robotics and expert systems,
there are also people and organizations capable of developing neural system products. The
potential sectors of application range from manufacturing, banking and finance, defense,
telecommunications, pharmaceuticals to holiday industry. Substantial amount of work is
being done at the Centre for Artificial Intelligence and Robotics (CAIR, Bangalore) and the
Institute for Robotics and Intelligent Systems (IRIS, Bangalore). They have developed a
neural network for optical character recognition. The project is complete and awaits
commercialization. IRIS is working on functional electrical simulation using neural networks
to simulate the muscles of a handicapped person and allow him to walk. Scientists at the
Indian Statistical lnstitute (Machine Intelligence Unit), Calcutta, have figured out computer
simulated models, more advanced than human brain, for creating artificial entities more
intelligent than present day systems in performing cognitive tasks. This project will have far
reaching implications on medical research and robotics.

40
 Chapter5

APPLICATIONS OF NEURAL NETWORK

5.1. Applications in Electrical Engineering:-

Artificial Neural Network (ANN) is currently a 'hot' research area in electrical engineering.
The model used to simulate artificial neural networks is based on the biological nerve cell or
neuron shown in Figure 5.1. Electrical signals arising from impulses from our receptor organs
(e.g. eyes, ears) are carried into neurons on dendrites.

5.1.1. Signal Classification with Perceptron:-

A problem of particular interest to electrical engineers is that of signal detection, particularly

in a noisy environment. Methods such as filtering and signal averaging have been used
successfully.

Fig5.1:-A biological neuron

5.2. Agricultural Control System Engineering:-

Control and management of agricultural machinery offers many opportunities for application
of general purpose empirical models. The nature of agricultural machines creates the need for
modeling systems that are robust, noise tolerant, adaptable for multiple uses, and are
extensible. Artificial Neural Networks (ANNs) have these characteristics and are attractive
for use in control and modeling in agricultural machinery.

41
 Fig.5.2:-Sprayer sensor and nozzle element

5.3. Neural Networks in Control Engineering:-

The ever-increasing technological demands of our modern society require innovative

approaches to highly demanding control problems. Artificial neural networks with their
massive parallelism and learning capabilities offer the promise of better solutions, at least to
some problems. By now, the control community has heard of neural networks and wonders if
these networks can be used to provide better control solutions to old problems or perhaps
solutions to control problems that have withstood our best efforts.

Neural networks have been applied very successfully in the identification and control of
dynamic systems. The universal approximation capabilities of the multilayer perceptron have
made it a popular choice for modeling nonlinear systems and for implementing general-
purpose nonlinear controllers.

For the purposes of this work we will look at neural networks as function approximators. we
have some unknown function that we wish to approximate. We want to adjust the parameters
of the network so that it will produce the same response as the unknown function, if the same
input is applied to both systems.

For our applications, the unknown function may correspond to a system we are trying to
control, in which case the neural network will be the identified plant model. The unknown
function could also represent the inverse of a system we are trying to control, in which case
the neural network can be used to implement the controller.

42
 Fig.5.3:-Neural network as function approximator

5.4. Fixed Stabilizing Controllers:-

This scheme has been applied to the control of robot arm trajectory, where a proportional
controller with gain was used as the stabilizing feedback controller. We can see that the total
input that enters the plant is the sum of the feedback control signal and the feedforward
control signal, which is calculated from the inverse dynamics model (neural network). That
model uses the desired trajectory as the input and the feedback control as an error signal. As
the NN training advances, that input will converge to zero. The neural network controller will
learn to take over from the feedback controller. The advantage of this architecture is that we
can start with a stable system, even though the neural network has not been adequately
trained.

Fig.5.4:-Fixed stabilizing controller

43
5.5. General Applications of Neural Network:-

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:-

 Sales forecasting
 ndustrial process control
 Customer research
 Data validation
 Risk management.

44
 Chapter6

CONCLUSION & FUTURE SCOPE

6.1 Conclusion:-

 The reliability of HVDC system has always been of primary concern in planning and
operation of power system.
 The total reliability of HVDC system depends upon the components in the line and at the
stations.
 The aim of back propagation network is to train the network to achieve a balance between
the ability to respond correctly to the input patterns that are used for training and ability to
provide good responses to the input.
 Therefore, by developing a computer program using back propagation algorithm and
integrate it with HVDC system model we can identify the fault and improve the reliability
of the system. Safe operation of HVDC system depends on fast detection of faults.
 As neural network has capability to map complex and highly nonlinear input-output
behavior, the approach is widely used to recognize patterns in electrical circuits, fault
identification. Fault identification is basically treated as a problem of pattern recognition.

6.2 Future Scope:-

 In the HVDC system, long-distance transmission and complex environment are the main
factors which are responsible for fault occurrence.
 Hence the Transmission Line Protection is must be done to stop the faults occurs in the
HVDC system. A novel fault location strategy for HVDC transmission line is presented.
 The fault signal is preprocessed with wavelets and fed to the neural network bank tuned
for each zones.
 The outputs shows that proposed fault location technique result in accurate decision
making for fault localizing.
 The use of Radial basis network saves time as well as memory, and fast creation of
network objects.

45
 REFERENCES

[1] N. Bawane, A. Kothari, and D. Kothari, “ANFIS Based HVDC Control and Fault
Identification of HVDC Converter”, HAIT Journal of science and engineeringB, volume2,
2005.

[2] N.Kandil, V.K.Sood, K.Khorasani, R.V.Patil, “Fault Identification in an AC-DC

Transmission System Using Neural Network”, transaction on power system, vol.7, 2 May
1992.

[3] K.S.Swarup,H.S.Chandrasekharaiah ,”Fault Detection and Diagnosis of Power Converter

Using Artificial Neural Network”, Indian institute of science, Bangalore.

[4] TaharBouthiba, “Artificial Neural Network-based Fault Location in EHV Transmission

Lines”, Electric Power System Research.

[5] Narendra Bawane, A.G.Kothari,”Artificial Neural Network Based Fault Identification Of

HVDC Converter” synopsiumon diagnostics for electric machine ,power electronic and
drives,Atlanta,GA,USA,24-26 August 2003.

[6] Zheng Xiao-Dong, Tai Neng-Ling, James S. Thorp, and Yang Guang-Liang, “A Transient
Harmonic Current Protection Scheme for HVDC Transmission Line”, IEEE
TRANSACTIONS ON POWER DELIVERY,VOL.27,NO.4,OCTOBER2012.

[7] Ruchita Nale & Suresh Babu, “Distance Protection of HVDC Transmission Line with
Novel Fault location technique”, IJRET,2014.

[8] V.S.Kale , S.R.Bhide & P.P.Bedekar, “Comparison of Wavelet Transform and Fourier
Transform based methods of Phasor Estimation for Numerical Relaying”, International
Journal of Advances in Engineering Sciences Vol.1, Issue 1,2011.

[9] LIANG Wen-li and SHANG Li-qun,” The Review of High Voltage DC Transmission
Lines Fault Location”, International Journal of Computer, Consumer and Control (IJ3C), Vol.
3, No.4, 2014.

[10] Jing Liu, Jiandong Duan, Hailong Lu and Yuanbing Sun, “Fault Location Method Based
on EEMD and Traveling-Wave Speed Characteristics for HVDC Transmission Lines”,
Journal of Power and Energy Engineering, 2015.

46