DETECTION AND CLASSIFICATION OF

ELECTRICAL POWER SIGNAL DISTURBANCE

USING SOFT COMPUTATION TECHNIQUE

Dissertation submitted

in partial fulfillment of requirements

for MASTER OF TECHNOLOGY

in

POWER SYSTEM of

MAULANA ABUL KALAM AZAD UNIVERSITY of TECHNOLOGY

by

SUBHRODIPTO BASU CHOUDHURY

Roll No.-10913414019

Reg. No. - 141090410027

Under the guidance of

Prof. SANKHADIP SAHA

DEPARTMENT OF ELECTRICAL ENGINEERING

NETAJI SUBHASH ENGINEERING COLLEGE

TECHNO CITY, GARIA, PANCH POTA

KOLKATA-700152

May, 2016

Dedicated

to

almighty

Guide

Parents

grandparents

CERTIFICATE

This is to certify that this dissertation titled “DETECTION AND CLASSIFICATION OF ELECTRICAL POWER

SIGNAL DISTURBANCE USING SOFT OMPUTATION TECHNIQUE” submitted in partial fulfillment of

requirements for award of the degree Master of Technology (M. Tech) in POWER SYSTEM of Maulana Abul

Kalam Azad University of Technology is a faithful record of the original work carried out by,

SUBHRODIPTO BASU CHOUDHURY, Roll no 10913414019 Regd. No. 141090410027, Year-2014

under my guidance and supervision.

It is further certified that it contains no material, which to a substantial extent has been submitted for the award

of any degree in any institute or has been published in any form, except the assistances drawn from other

sources, for which due acknowledgement has been made.

___________

Date:……………… Guide’s Signature

Prof. Sankhadip Saha

Sd/__________________

Head of the Department

ELECTRICAL ENGINEERING

NETAJI SUBHASH ENGINEERING COLLEGE

TECHNO CITY, GARIA, KOLKATA – 700 152

DECLARATION

I hereby declare that this dissertation titled

DETECTION AND CLASSIFICATION OF ELECTRICAL POWER SIGNAL

DISTURBANCE USING SOFT COMPUTATION TECHNIQUE

is my own original work carried out as a post graduate student in NETAJI SUBHASH ENGINEERING

COLLEGE (109) except to the extent that assistances from other sources are

duly acknowledged.

All sources used for this dissertation have been fully and properly cited. It contains no material which to a

substantial extent has been submitted for the award of any degree in any institute or has been published in any

form, except where due acknowledgement is made.

Date:………………… ……………………………………….

Student’s signature

SUBHRODIPTO BASU CHOUDHURY

Roll No. – 10913414019

CERTIFICATE OF APPROVAL

We hereby approve this dissertation titled “DETECTION AND CLASSIFICATION OF

ELECTRICAL POWER SIGNAL DISTURBANCE USING SOFT COMPUTATION

TECHNIQUE” carried out by SUBHRODIPTO BASU CHOUDHURY, Roll no. 10913414019 Regd.

No.-141090410027 & Year- 2014 under the guidance of Prof. Sankhadip Saha of Netaji Subhash

Engineering College (109), in partial fulfillment of requirements for award of the degree

Master of Technology (M. Tech) in POWER SYSTEM of Maulana Abul Kalam Azad University

of Technology (M.A.K.A.U.T).

Date:……………………………

Examiners’ signatures:

1. ………………………………………….

2. ………………………………………….

3. ………………………………………….

CONTENTS

CHAPTER NAME Page No.

ACKNOWLEDGEMENT i

ABSTRACT ii

ORGANISATION OF THESIS iii

1. INTRODUCTION

1.1 Overview 1-2

1.2 Literature review 3-8

1.3 Aim of thesis 9

2. GENERAL DISTURBANCE SIGNAL IN ELECTRICAL POWER SYSTEM

2.1 SAG 12-15

2.2 SWELL 16-17

2.3 HARMONIC 18-19

2.4 FLICKER 20-21

2.5 NOTCH 22-23

2.6 OSCILLATORY TRANSIENT 24-25

2.7 SPIKE 26-27

2.8 INTERRUPTION 28-29

2.9 TEMPORARY OUTAGE 30-31

3. NECESSITY AND REQUIREMENT OF FAULT CLASSIFICATION 33

4. FEATURE EXTRACTION FROM DISTURBANCE SIGNAL 34-37

4.1 CORRELATION

4.1.1 AUTO-CORRELATION

4.1.2 CROSS-CORRELATION

5. FEATURE SELECTION AND CLASSIFICATION TECHNIQUE 38-43

5.1 ARTIFICIAL NEURAL NETWORK (ANN)

5.2 PROBABILISTIC NEURAL NETWORK (PNN)

5.3 RADIAL BASIS FUNCTION (RBF)

6. BLOCK DIAGRAM FOR CLASSIFICATION OF ELECTRICAL DISTURBANCE 44

7. RESULT 45

8. CONCLUSION 46

9. FUTURE ADVANCEMENT 47

10. REFERENCE 48-49

APPENDIX 1 50-51

APPENDIX 2 52

ACKNOWLEDGEMENT

I am grateful to the Almighty for providing me a caring and helpful guide Prof. Sankhadip Saha (Dept. of

ELECTRICAL ENGINEERING) in carrying out my project activities. He constantly extended his helping

hand during problems and took updated news regarding project. He also paved me a way to my future research

work on SOFT COMPUTATION as it is subject of my taste. He also introduced me to this vast field and

informed me on present research on SOFT COMPUTATION.

I equally pay my gratitude to Prof. Subrata Biswas (Incharge, MTECH, POWER SYSTEM) for

provision of constant updated information regarding POWER SYSTEM department. I am also grateful to Head

of Electrical Engineering department Prof. Tridibesh Nag and Dean Prof. Dr. S. K. Bhattacharya.

Finally, I am thankful to my parents and late grandparents for their well wish and constant motivation during

any tension for project progress.

Dated:………………………...

SUBHRODIPTO BASU CHOUDHURY

Roll No.: 10913414019

i

Abstract

Electrical power disturbances are undesirable interruption of electrical power signal. They are mainly caused by

switching of loads. According to IEEE there are several types of disturbance that occur in real time situation: Sag, Swell,

Interruption, Transient, Harmonic, Notch, Spike, Flicker, and Temporary Outage.

Nowadays, electrical consumers are very aware of paying for only the electrical energy they are consuming. Since, cost

of generation of electrical power is economical area of concern to electrical utility companies; hence wastage of

electrical energy is avoided. This can be achieved by identification n powers of electrical disturbances occurring in

electrical power signal and swiftly mitigate the disturbance.

The proposed method constantly extract the features from disturbance signal and detect the type of disturbance

occurring. These disturbances actually cumulatively impair the insulation and aggravate electrical energy wastage, which

is undesirable. The method uses cross-correlogram of power signal and electrical disturbance signal to extract features.

These features are distinct or different for different class or type of electrical disturbance. If any of the features does not

match with features of that class of electrical disturbance, the classifier machine cannot identify the electrical

disturbance. The cross-correlogram is utilized to extract certain distinct features.

These features are used to train classifier machine and it is tested using data of some other set of disturbance and

accuracy is computed.

Classifier is delicate and soft computation part evolved in the middle of 21st century. They are primarily used for

classification purposes. Several researchers put forward new algorithms to design new classification technique to

improve classification accuracy. Earlier only ARTIFICIAL NEURAL NETWORK was used for classification purpose. But,

present era witnesses several new classification techniques like kNN, SUPPORT VECTOR MACHINE, CORE VECTOR

MACHINE, ENSEMBLE METHODS, DEMPSTER-SHAFER METHOD OF CLASSIFICATION, BOOTSTUP METHODS, etc.

A new advanced technology can be used to classify the electrical disturbances. GAME THEORY technology can

also be used.

Keywords: disturbance, classification, computation, features

ii

ORGANISATION OF THESIS

Chapter 1 deals with overview and vitality for electrical signal disturbance classification. It also describes the earlier

classification method used by earlier eminent personalities with their classification accuracy. They are presented

chronologically year wise. It also describes about the method used here to classify the disturbance signal in brief.

Chapter 2 describes twenty different types of electrical faults occur in real life situation. It also guides about the reason

for occurance of specific faults. Their appearance in real life field is presented pictorially using modern simulation

software. They are presented using IEEE standard 1159-1995.

Chapter 3 guides the significance of disturbance classification as mitigation methods are distinct pertaining to type of

occurring electrical disturbance.

Chapter 4 elaborates signal processing technique. Here CROSS-CORRELATION technique is used. It enlightens about

correlation method and type of CORRELATION used. Mostly, as two signals are different i.e. one is normal electrical

power signal and other one is disturbed signal, therefore, CROSS-CORRELATION technique is used. Cross-correlogram is

also attached for SAG disturbance and normal power signal.

Chapter 5 describes the significance of feature selection technique. The presented method compares two classification

technique using ARTIFICIAL NEURAL NETWORK (ANN), PROBABILISTIC NEURAL NETWORK (PNN) and RADIAL BASIS

FUNCTION NEURAL NETWORK (RBFNN). Twelve features are extracted from CROSS-CORRELOGRAM of disturbed signal

with normal electrical power signal. These features are extracted using certain feature extraction code in simulation

program. A feature matrix is formed which is used to train classifier and validate it by using some other number of

samples.

Chapter 6 shows a pictorial representation of classification technique in block diagram. Each block represents a method

or activity that a classification technique passes by.

Chapter 7 describes about classification accuracy of different single and multiple electrical disturbance signals. It is

noted that classification accuracy degrades on addition of noise of higher magnitude.

Chapter 8 concludes and specifies the reason for degradation of disturbance classification accuracy on addition of noise

level (db). It also specifies the reason for different classification accuracy in case of different classification techniques.

Chapter9 gives idea about other signal processing techniques to extract features. They can be merged with features

extracted from cross-correlogram and use several other feature selection algorithms to classify them.

Chapter 10 lists about the research work paper or journal from where inspirations have been drawn to urge for the

presented real life electrical disturbance classification method.

iii

1. INTRODUCTION

1.1 Overview

Power equipments demand better power quality for smooth and reliable operation. Due to excessive use of non-linear

electronic loads such as adjustable speed drive, energy efficient lamp, PLC operated devices, etc. Equipments that are

used to generate renewable energy also indulge contamination to the electrical power signal. These electrical

disturbances cause unreliable operation of electrical equipment and require frequent maintenance of electrical

equipment which highlights economical background. Reliable operation of electrical devices is desirable and essential to

consumers as they pay for the usage of electrical energy.

It is essential to sense electrical disturbance (ED) (if any) present in the signal. Different methods are suggested like,

wavelet transform based in [1] and [7], S-transform analysis in [2], wavelet based neural network in [3], etc. in recent

years to detect and classify electrical disturbances in signal. Due to presence of noise in signal classification accuracy of

wavelet network degrades. Noise attachment in the electrical signal is an inherent property which arises due to

switching operation and discrepant connection between terminals of electrical equipment. Switching is employed to

convert high voltage alternating electrical quantities to high voltage direct electrical quantities in substations to reduce

transmission losses. During continuous conduction if switching operation is done, a discontinuation in flow of electrical

quantities is observed for infinitesimal small time interval. Hence, noise is observed as if during switching, the

conversion from alternating quantities to direct electrical quantities is not possible at the instant of switching operation.

But emphasis is not given on recognition of multiple electrical disturbances (MED) occurring simultaneously. Most of the

analysis in references uses high speed computer to handle the huge computational burden within small time. Moderate

trend is to build a detecting device using microcontroller as the features of the microcontroller can be easily changed by

modifying its firmware instead of changing its hardware.

The field of SOFT COMPUTATION utilizes several methodologies to detect certain set of patterns or events and hence

classify them. This field nowadays witnesses advanced research in the event of classification. Several renowned curious

persons indulge themselves in finding new advanced technology by gaining experiences from previous knowledge of

classification techniques.

This paper is an outcome of a comparative study between three classifier machines namely ARTIFICIAL NEURAL

NETWORK (ANN), PROBABILISTIC NEURAL NETWORK (PNN )and RADIAL BASIS FUNCTION NEURAL NETWORK (RBFNN)

classifier. The technique used in method has two parts. Primary part involves feature extraction from ELECTRICAL

DISTURBANCE SIGNAL using different signal processing techniques and secondary part employs classification of

disturbance signal based on extracted features.

1

Apart from other two classifier machine ARTIFICAL NEURAL NETWORK (ANN) is s better robust and gives better result.

Suitable features are extracted from cross-correlation sequence of two signals, viz. one is normal sinusoidal power

frequency electrical signal and the other is ELECTRICAL DISTURBANCE signal. Correlation gives the extension of similarity

between two signals.

This paper presents a comparative study between different electrical disturbance classification techniques. The

method successfully classifies disturbances like, sag, swell, spike, harmonic, notch, flicker and other single electrical

disturbance.

2

1.2 Literature review

A.M. Youssef et. Al. in [1] in 2004 utilises time warping classifier for detection of electrical distortion of power signal.

Vector quantization and fast match is used to accelerate classification process. Walsh transform and fast Fourier

transform are used to extract features. This method computes better result of classification than traditionally artificial

neural networks and fuzzy logic controllers. The method has lower sensitivity to noise levels.

P. Dash et. Al. in [2] in 2004 proposed a method of localization and classification of short duration disturbances in power

networks using phase-corrected wavelet transform known as S-transform and extended Kalman filter. It has time-

frequency resolution characteristics and provides detection, localization and visual patterns perceptible for automatic

recognition of automatic classification of electrical faults. The method is applied to practical disturbance and reveals

significant classification accuracy.

T. K. Abdel Galil et. Al. in [3] in 2004 presents a novel approach of classification of electrical disturbance signal based on

inductive learning using decision trees. Wavelet transform uses representative features that usually capture unique and

distinct features for each class of disturbance signal. Decision tree is formed using feature vectors from wavelet analysis

by inductive learning process. Decision tree has some predefined rules which are used to classify faults which are

interpreted form the extracted features. Classification accuracy is incomparable with artificial neural networks based

classification.

Haibo He et. Al. in [4] 2006 classifies the PQ disturbance based on WAVELET TRANSFORMATION and SELF-ORGANISING

LEARNING ARRAY (SOLAR) system. Features from electrical PQ disturbance signal are extracted utilizing the WAVELET

TRANSFORMATION method based on MULTIRESOLUTION ANALYSIS (MRA). Feature vector of PQ disturbance signal is

constructed from energy of the DETAIL and APPROXIMATION levels after multi-resolution analysis of sampled PQ

disturbance signal for future testing and training. Neuron parameters and connections of SOLAR are specified pertaining

to the minimum entropy adaptively set for each neuron. The accuracy for PQ disturbance classification is 94.3%.

Ömer Nezik Gerek et. Al. in [5] in 2006 presents higher order cumulants as feature parameter and quadratic classifiers

for classification purpose. A feature vector of six local maxima and minima of higher order cumulants starting from

second variance to fourth cumulant is constructed. Local vector magnitudes and simple threshold provides an

immediate detection criterion. Local maxima and minima around the vicinity of electrical disturbance of a particular type

are used for classification purpose. 2-D classifiers have classification success of 98% whereas 6-D classifiers have

classification success of 100%.

Inigo Monedero, Carlos Leon et.al.in [6] in 2006 is capable of classifying disturbances of electrical signal by usage of

ARTIFICIAL NEURAL NETWORK. Fourier and wavelet methods can be used to generate good signal features but they are

complex in comparison to feature extraction using ANN method. Fault signal is decomposed in APPROXIMATION

WAVELET COEFFICIENTS and other set of DETAIL WAVELET COEFFICEINTS. The obtained APPROXIMATION

COEFFICIENTS are further decomposed in several levels of successive APPROXIMATION and DETAIL WAVELET

COEFFICEINTS to increase the level of resolution. After feature extraction, classification process is executed using

method of pattern recognition by ANN. It is achieved by forming adequate number of training patterns to train ANNs

correctly so that they can classify future inputs (FEATURES FROM FUTURE OCCURING FAULT).

3

A problem with ANN applied to PQ is obtaining indispensible useful real training patterns directly from power grid due

to irregular apparition of these disturbances and problem in capturing them. The scheme satisfactorily classifies the

POWER QUALITY DISTURANCES in electrical signal with partial accuracy of 89%.

Ömer Nezih Gerek et. Al. in [7] in 2006 uses covariance method of classifying voltage disturbance. A feature vector is

constructed using selected features like, local wavelet transform at various decomposition levels, spectral harmonic

ratios local extrema of higher order statistical parameters. The proposed scheme distinguishes the loading conditions

within same class of electrical disturbance with accuracy of 70%.

Fengzhan Zhao et. Al. in [8] in 2007 detects disturbance in electrical power signals and classify them using the S-

transformation technique. Features like No. of main frequency, No. of peak in STANDARD DEVIATION (STD)curve, No.

of peaks in high frequency of (STD) curve, mean of POWER DISTURBANCE signal and a composite feature after S-

transformation of POWER DISTURBANCE signal. Composite features composed of 3 intermediate features like degree

of SAG, INTERRUPTIONS or SWELL in electrical signal. Classification is done based of some rule generated which is

required to classify the POWER SIGNAL DISTURBANCES. This scheme classifies 98.5% of SNR=30dBofPOWER SIGNAL

DISTURBANCE of electrical signal accurately.

S. Suja et. Al. in [9] in 2007 are capable of classifying POWER SIGNALDISTURBANCE based on features extracted using

DISCRETE WAVELET TRANSFORMATION (DWT). DWT technique is applied to extract time of disturbance occurance and

frequency features from POWER DISTURBANCE signal. These DWT coefficients are applied as inputs to neural networks

along with MRA technique and statistical manipulations to extract features from disturbance signal at different

resolution levels. The SQUARED APPROXIMATE COEFFICIENTS after DWT at primitive level are calculated. After that,

APPROXIMATE COEFFICIENTS and DETAIL COEFFICIENTS are further subjected to CUMULATIVE DISTRIBUTION

FUNCTION (CDF) and PROBABILITY DENSITY FUNCTION (PDF) at subsequent 2 levels for clear distinction between

different POWER SIGNAL DISTURBANCES. Their attempt has classification accuracy of 86% in pertaining to Power Signal

disturbance.

Mamun Bin Ibne Reaz et. Al in [10] in 2007 shows a different type univariate randomly optimized neural network in

conjunction of discrete wavelet transform and fuzzy logic to achieve better electrical disturbance classification accuracy.

The system is composed of VHSIC Hardware Description Language (VHDL) a hardware description language, succeeded

by extensive testing and simulation to justify functionality of system that allows hardware implementation.

4

The method reaches 98.19% classification accuracy for application of system on software-generated disturbance signal.

Ameen M. Gargoomet.al.in [11] in 2008 automatically classifies electrical disturbances using multi-resolution S-

transform and Parseval’s theorem. S-transform is utilized to produce instantaneous frequency vectors. Energy of these

vectors are used to classify disturbance signal using Parseval’s theorem. The significance of method is magnitude,

duration and frequency content can be effectively identified to characterize the electrical disturbance. The method was

tested using two typical case studies and produced favourable and desired classification accuracy of disturbance.

Suryannarayana Chandaka et. Al. in [12] in 2009 introduces a pattern recognition technique called CROSS-CORRELATION

aided SVM based classifier. The scheme is utilized for binary classification of EEG signals. The features like PEAK VALUE,

INSTANT OF PEAK OCCURANCE, CENTROID, EQUIVALENT WIDTH and MEAN SQUARE ABSCISSA are extracted from

CROSS-CORRELATED sequence between two different EEG signals. The classification of EEG signals is done using

SUPPORT VECTOR MACHINE (SVM) for solving a binary classification problem in a supervised manner and learning

problem is formulated as a quadratic optimization problem where the error surface is free of any local minimum and has

global optimum. SVM creates OPTIMAL SEPERATING HYPERPLANES (OSH) between different distinct CLASSES on basis

of STRUCTURAL RISK MINIMIZATION (SRM) principle. The method achieved ECG signal classification accuracy of

95.96%.

Chiung-Chou Liao et. Al. in [13] in 2009 introduces classification of noise influenced disturbance with integrated feature

extraction and neuro-fuzzy network. Wavelet transformation coefficients ahs numerous features needed for transient

signal identification of electrical signal variation. This scheme reduces features of noise, noise-suppression and energy

spectrum in wavelet transform coefficients in different scales are computed using Parseval’s theorem. Neuro-fuzzy

network is applied to frame fuzzy-rule and classify disturbance signal.

P. Rajamani et. Al. in [14] in 2011 shows the capability of identifying fault characteristics of dynamic insulation failure

during impulse test. Features viz. FAULT TYPE, CONDITION OF FAULT and LOCATION OF OCCURANCE OF FAILURE

ALONG LENGTH OF WINDING OF THE TRANSFORMER are extracted from cross-correlated sequence of fault signal and

electrical signal of fundamental frequency (50Hz). A stochastic gradient based MULTI-DIMENSIONAL WAVELET

NETWORK (MDWN) is developed and employed for identification of fault characteristics of dynamic insulation failure.

Features like MAXIMUM VALUE OF CROSS-CORRELATED SEQUENCE, EQUIVALENT WIDTH, CENTROID, ABSOLUTE

CENTROID, ROOT MEAN SQUARE ABSCISSA, KURTOSIS OF CORRELATED SEQUENCE, SKEWNESS OF CORRELATED

SEQUENCE and SKEWNESS OF FAULT SIGNAL are extracted from cross-correlated sequence of electrical power signal of

fundamental frequency of 50Hz and fault signal of transformer under insulation test and used to train the MDWN to

generate the targets. This MDWN is used in future to test the classification of fault characteristics. The approach has

classification accuracy of 90.70%.

5

Now the classification accuracy attained variability on use of different mother wavelet to train MDWN.

Chun-Yao Lee et. Al. [15]in 2011 uses PROBABILISTIC NEURAL NETWORK BASED FEATURE SELECTION(PFS) method to

classify PQ fault in electrical signal. The scheme combines global optimization algorithm with an ADAPTIVE

PROBABILISTIC NEURAL NETWORK (APNN) for removing redundant and irrelevant features from PQ disturbance signal.

Total 62 features were extracted from 11 different types of PQ disturbance signals from ST and TT transformations. A

PNN (PROBABILISTIC NEURAL NETWORK) based feature extraction was implemented by combination of FULLY

INFORMED PARTICLE SWARM (FIPS) with an APNN. FIPS was used for optimization of smoothing parameters of APNN

by gradual removal of irrelevant features from PQ disturbance signal. This method classifies the PQ disturbance signal

with accuracy of 96.3%, 90.8% and 96.1% of SNR of ∞Db, 20Db and 30Db.

Milan Biswal et. Al. in [16] in 2013 proposed a fast variant of S-transform algorithm for extraction of relevant features

used to distinguish different electrical signal deviations by a fuzzy decision tree based classifier. It has higher recognition

success rate for identification of simultaneous disturbances in electrical signal. Fast dynamic S-transform is used to for

appropriate time-frequency localization, decision tree algorithm finds its use in selection of optimal features and fuzzy

decision rules complements overlapping patterns. It has an accuracy rate of 98.6% for classifying electrical distortion

with 40Db noise level.

Sovan Dalai et.al.on [17] in 2013 successfully selects the optimal features from cross-correlation sequence between PQ

disturbance signal and power signal of fundamental frequency(50Hz). Several signals with single fault and combination

between different fault signals are used to generate other different categories of fault signals using simulation

software(MATLAB). Their work emphasize on successful extraction of 12 features namely MAXIMUM VALUE OF

CORRELATION SEQUENCE, INDEX VALUE OF CORRELATION SEQUENCE, EQUIVALENT WIDTH, CENTROID, ABSOLUTE

CENTROID, ROOT MEAN SQUARE WIDTH, MEAN VALUE OF CORREATION, STANDARD DEVIATION, SKEWNESS OF

CORRELATION SEQUENCE, KURTOSIS OF CORRELATION and PQ DISTURBANCE signal. ROUGH SET THEORY is used as it

successfully selects optimal features from a situation where knowledge of features is less and also redundant features

exist. CONDITION ATTRIBUTES is tabulated containing the features from CORRELATION SEQUENCES of PQ disturbance

signal and power signal of fundamental frequency. DECISION ATTRIBUTE is created based on ROUGH SET THEORY

concept. Their proposed approach has the classification accuracy of 97.10%.

Shufan He et. Al. in [18] in 2013 uses hybrid method based on S-transform and dynamics. Hybrid method primarily

dynamics to identify the signal components in frequency spectrum of Fourier transform and uses inverse Fourier

transform to some signal components. Features are extracted from Fourier transform, S-transform and dynamics and

decision tree is used to classify the faults. Reduction of Heisenberg’s uncertainty is achieved by windowing signal

components by different Gaussian windows, results in better adaption and flexibility. DSP-FPGA hardware platform is

used to test run time and justification of the proposed method.

6

Swati Banerjee et. Al. in [19]2014 successfully classifies ECG signal based on features extraction from CROSS-WAVELET

transformation. CROSS-WAVELET transformation is used for analysis of different stationary ECG signals. Single lead data

from ECG signal is first deionised and beat segmentation is done to normalize the cardiac beats in time domain.

CONTINUOUS WAVELET TRANSFORMATION of NORMAL TEMPLATE SIGNAL and SIGNAL TO BE ANALYSED is carried

out for CROSS-WAVELET TRANSFORMATION of two continuous wavelet transformed of signals. CROSS-WAVELET

TRANSFORMAITON is succeeded by CROSS-WAVELET SPECTRUM and WAVELET COHERENCE for classification of

different ECG signals. The datasets that is used to classify the ECG signals composed of three sets of parameters namely

(pa1,pp1), …..,(pa3,pp3). The ECG signals were classified by THRESHOLD-BASED CLASSIFICATION technique. The work

incurred an ECG signal classification accuracy of 97.6%.

Martin-Valtierra Rodriguez et. Al. in [20] in 2014 proposed a new method of classifying single and multiple electrical

disturbances using dual neural network. It uses adaptive linear network for estimation of harmonic and inter-harmonic

for computation of root mean square voltage and total harmonic indices. These indices are used to classify sag, swell,

outages and harmonic and inter-harmonic disturbances whereas feed-forward neural network is used to recognize

pattern of identification of spike, flicker, notch and oscillatory transients from vertical and horizontal histograms of

specific voltage waveform. It reaches classification of accuracy of 98% for noiseless signal and 90% for contamination of

20Db SNR noise level.

Sovan Dalai et. Al. in [21]in 2015 enables to classify MULTIPLE POWER QUALITY DISTURBANCES in electrical signal. Their

work comprises of two steps of feature extraction from MULTIPLE POWER QUALITY signal and classification of the

signal. FLDA (Fischer Linear Discriminant Analysis) method is used for feature extraction on its suitability of non-linear &

non-stationary MULTIPLE POWER QUALITY signal and its capability of fetching smaller set of data from higher order

statistical data sets. Features are extracted from CROSS WAVELET SPECTRUM of MULTIPLE POWER QUALITY SIGNAL

and POWER SIGNAL of standard fundamental frequency (50Hz). Classification of MULTIPLEPOWER QUALITY signal is

done using SUPPRT VECTOR MACHINE (SVM) as it ensures reduction of structural and experimental risk in search of

superior generalization when the large dimension set of training set is provided but the dimension of test set is limited.

Their scheme successfully classifies MULTIPLE POWER QUALITY DISTURBANCE of an electrical signal with 99.09%

accuracy.

Zhigang Liu et. Al. in [22]2015 develops a combination method for recognition of complex electrical disturbances based

on ensemble empirical mode decomposition and multi-label learning. EEMD is used to extract features suitable for non-

stationary disturbance in signals. Rank wavelet support vector machine is finds its application in classification of

electrical distortion. Features are extracted from complex disturbance using EEMD by defining standard energy

difference of each intrinsic mode function. Optimization of rank based SVM based of wavelet kernel function ranking

function and multilabel functions are prepared.

7

Finally, rank-WSVM finds its classification task of electrical disturbance.

C. –H. Lin et. Al.in [23] in 2005 proposes harmonic detection and voltages using wavelet-based network. It is a two-layer

architecture containing wavelet layer and probabilistic network. Wavelet transformation carries out feature extraction

from various disturbances and probabilistic network to analyse translation patterns form time-domain distorted wave

and perform classification task. The test result concludes that simplified network architecture enhances classification

performance and curtails processing time for disturbance event detection. It achieves an average classification accuracy

of 98.7%.

8

1.3 Aim of thesis The thesis provides the importance of fault classification technique. The noticeable fact about the

classification method is the type of significant features extracted from cross-correlogram.

It also uses modern simulation software to generate nine different types of electrical

disturbances that generally occur in power system. The mathematical equations of the fault are in

accordance with IEEE 1159-1995 standards. The features extracted from faults should be distinct otherwise,

it classifier will be unable to detect the type of electrical disturbances.

Other features can also be extracted. But the computational time for feature extraction would

become more and also memory requirement would be large. Thus, here only twelve features are extracted.

Power system network is frequently subjected to transients which cause generation of electrical

disturbance. These are cumulative in nature and relay should be fast acting to send tripping alarm signal to

circuit breaker. Hence, relay has to transmit which type of disturbance a power system network is subjected

as mitigation methods are different for different type of disturbances.

Nowadays power system uses smart relays which can transmit signal to telecommunication

device specifying the type of disturbance.

9

2. GENERAL DISTURBANCE SIGNAL IN ELECTRICAL POWER SYSTEM

Disturbance in power signal occurs due irregular deviation in argument (magnitude, frequency, etc.) of power signal.

This disturbance is manifested by electrical energy wastage and imparts damage to user’s electrical equipments.

Disturbance occurs due to continuous switching on/off of user’s electrical equipments. As power system is subjected to

continuous load demand change hence, disturbance in power signal transmission is inevitable situation.

Based on observation and study the list entails the categories of disturbance in power signal.

SAG

SWELL

OSCILLATORY TRANSIENT

HARMONIC

FLICKER

NOTCH

SPIKE

MOMENTARY INTERRUPTION

TEMPORARY OUTAGE

Pertaining to the knowledge on occurance of multiple electrical disturbances several other types of faults are

generated in combination of aforementioned faults. Electrical disturbance signals are generated according to

the mathematical formulation of each signal in accordance with IEEE 1159-1995. The mathematical formula for

single electrical disturbance is given in the table next page.

10

Table I

MATHEMATICAL FORMULATION OF DIFFERENT ELECTRICAL DISTURBANCE SIGNAL

Sl. No. TYPE of ELECTRICAL

DISTURBANCE (Event)

CAUSE MODEL EQUATION PARAMETER CONSTRAINT

1 Sag

Line Fault VSlf=[1-m(u1-u2)]sin(2πft+θ) 0.1≤m≤0.9,u1

, u2-unit step function

f-fundamental frequency

Transformer Energizing VSt=[1-m(u1e-a(t-t1))] sin(2πft+θ)

Induction Motor Starting VSm=[1-m(Fs-Fr)] sin(2πft+θ) Fs= u1(1-e-at1)+ u1(mrsin(2πfrt1

+θ))e-ϒt1

Fr= u2(1-e-bt1) 0.1≤m≤0.9 a-sag decay

rate b-sag recovery

rate

2 Swell Unbalance fault VSw=[1+ m(u1-u2)]sin(2πftt+θ) 1.1≤m≤ 1.8 u1, u2-unit

step function f-fundamental

frequency

3 Transient Capacitor switching VTc= sin(2πft+θ)+[u1mtsin(2πftt1)e-ϒt1] ft≤5kHz mt≤4

Line Energizing VTl=u1* sin(2πft+θ)+[u1mtsin(2πfrt1)e-ϒt1] ft≤5kHz mt≤4

Lightning Strike VTl= sin(2πft+θ)+[u1mte-ϒt1] 1.2/50µs

4 Notch Electronic switching VNT= sin(2πft+θ)-sign(sin(2πft+θ)) 𝛴09 𝜅[(u1(t-(t1+0.02n)-u2(t-

(t2+0.02n)]

5 Flicker Electric arc furnace and wind farm

VF=[1+mFsin(2πfFt+θF)] sin(2πft+θ) mF≤0.03 fF≤10Hz

6 Harmonic Non-linear electronic loads

VHm=sin(2πft+θ)+[ΣmNsin(2πfNt+θN)],n>1 mN=magnitude of n order

harmonic fN- n order harmonic frequency

7 Noise Electromagnetic interference

VN= sin(2πft+θ)+[nNLW(t)] nNL- noise level(%)

W(t)-white Gaussian

noise

11

DISCUSSION ON ELECTRICAL SIGNAL DISTURBANCE

2.1 SAG Sag in signal occurs when the amplitude of transmitting signal lessen its value than a normal value of electrical

power signal. According to IEEE 1159-1995 a fault is said to be SAG if its peak magnitude varies from 0.1-0.9 pu

(per unit) of system voltage and persists for 0.5-30 cycles. This situation occurs due to the following causes:

Single line fault carrying high capacity of electrical power

Starting of induction motor

Energizing of unloaded transformer

SINGLE LINE FAULT: Power system is stable by generation side and load side balance of active power and reactive power.

On occurance of single-ground fault, the generation and load side balance of both active and reactive power are not in

parity. Due to this disparity, the generation side suffers from oscillation of frequency of generator though it is mitigated

by AUTOMATIC LOAD FREQUENCY CONTROL (ALFC). But the disparity of reactive power results in dipping of voltage

signal through the transmission line which causes mal-operation of end user’s electrical equipments. A sample sag signal

for single line fault in per unit system is given in next page:

Fig.1. Voltage signal with Sag

The effects of faults are given below:

o Steady and transient instability

o Aging effects of electrical equipments

12

STARTING OF INDUCTION MOTOR: Induction motor is composed of ferromagnetic rotor and stator. It develops torque

on basis of DOUBLE REVOLVING FIELD THEORY as two revolving magnetic fields are sweeping across each other stator

field is revolving at synchronous speed. Hence, the induction motor requires large reactive power from power supply as

the both stator and rotor circuits are both mostly inductive. Thus during starting of induction motor the voltage

magnitude suddenly dips on consumption of reactive power .A sample sag signal for starting of induction motor in per

unit system is given below:

Fig.2. Voltage with Sag due to starting of induction motor

The fault creates transient state problem in the power system operation.

ENERGISING OF UNLOADED TRANSFORMER: Transformer is a static electric device that transforms electrical power from

one electrical circuit to another electrical circuit at constant electrical frequency. It actually consists of an

electromagnetic core wounded by two or more electrical conductors to energize the core and sets up electromagnetic

flux. Due to mutual flux linkage between different coils wound on the core, the coils gets electromagnetically energized

and electromagnetic force (emf) induced in the coil. This situation occurs when all coils are connected to some active

(closed) electrical equipment. But when one or more coils are not connected to active electrical equipments the mutual

flux linkage does not takes place between coils. The coil not connected to active electrical equipments gets highly

energized and consumes huge amount of reactive power from the electrical power supply causing voltage to dip in

magnitude.

The situation is manifested by effect listed under:

Transient instability

13

Fig.3. Voltage with Sag due to energizing of unloaded transformer

14

Fig.4 Samples of voltage signal with sag

15

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 1/2 cycle sag

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 2 cycle sag

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 1 cycle sag

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 3 cycle sag

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 1 cycle sag

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 1/2 cycle sag

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 3/2 cycle sag

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 3 cycle sag

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with 7/2 cycle sag

Time(second)

Am

plitu

de(p

u)

2.2 SWELL The swell in signal occurs when the magnitude of amplitude increases with respect to the normal value

of electrical power signal A swell generally has peak magnitude of 1.1-1.8 pu(per unit) of system voltage

and remains for 0.5-30cycles.The situation occurs due to following causes:

Switching of large capacitor banks

Switching off large loads

SWITCHING OF LARGE CAPACITOR: Charged capacitor plays a vital role to supply electrical reactive

power in its shortage. They also serve the role of power factor improvement of the whole electrical

power system. So when capacitor bank is switched on there is excess of electrical reactive power in the

power system. Thus the voltage suddenly increases in magnitude than its normal value.

This cause has subsequent effects listed under:

Insulation breakdown of electrical equipments:

Fig.5. Voltage signal with Swell due to switching of large charged capacitor

16

Fig. 6 Samples of voltage signal with swell

17

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-2

-1.5

-1

-0.5

0

0.5

1

1.5

2Voltage signal with 2 cycle swell

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-2

-1.5

-1

-0.5

0

0.5

1Voltage signal with 1/2 cycle swell

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with 1 cycle swell

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with 7/2 cycle swell

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-2

-1.5

-1

-0.5

0

0.5

1

1.5

2Voltage signal with 3 cycle swell

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-2

-1.5

-1

-0.5

0

0.5

1

1.5

2Voltage signal with 3/2 cycle swell

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.5

0

0.5

1

1.5

2Voltage signal with 1/2 cycle swell

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with 3 cycle swell

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-2

-1.5

-1

-0.5

0

0.5

1

1.5

2Voltage signal with 2 cycle swell

Time(second)

Am

plitu

de(pu)

2.3 HARMONIC

It is caused due to introduction to higher order harmonics of fundamental component of the electrical power signal.

Frequent switching of electrical equipments introduces harmonics in power system. Also mal-operation of electrical

equipments like motors causes harmonics to merge with the system electrical power signal. According to IEEE 1159-

1995, harmonic has peak magnitude of maximum 0.2 pu (per unit) system voltage and has integral multiple of system

frequency (in India it is 50 Hz).

The causes for the harmonic introduction into the power signal are listed below:

excessive usage of non-linear electrical motor loads and speed drives

large UPS (UNINTERRUPTIBLE POWER SUPPLY) system

The effects of harmonic introduction are given below:

overheating of electrical equipment and conductor

produces noise at excessive level of harmonic introduction

Fig.7. Voltage signal with Harmonic

18

Fig. 8 Samples of voltage signal with harmonic

19

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal 1st,2nd & 3rd harmonic

Time(second)

Am

plitude(pu)

2.4 FLICKER

Flicker in electrical power signal is situation occurs due to irregular dip and hike in magnitude of amplitude

in electrical power signal. The causes for flicker in electrical power signal are listed below:

arcing in electrical conductor

usage of large electrical motor loads with variable loads

Flicker produce consequence listed below:

Disturbance in monitoring equipment

Fig. 9: Voltage signal with flicker

20

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plit

ude(p

u)

Fig. 10 Samples of voltage signal with flicker

21

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal flicker

Time(second)

Am

plitude(pu)

2.5 NOTCH

Fig. 11 Voltage signal with notch

Notch signal is caused due to sudden switching operation. It can be viewed as sudden voltage sag for infinitesimal time.

This type of operation is employed when heavy electrical loading is the consequence of 1 single-ground fault. During this

type of fault, the terminal voltage of alternator should be lowered for reduction of electrical loading. Depth of notch

depends on switching instant or ignition angle of convertor devices in substations.

Notch can be avoided in following ways:

Decreasing switching angle

Reduction of switching operation

22

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

Notch

Time(seconds)

Ampl

itude

(pu)

Fig. 12 Samples of voltage signal with notch

23

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)A

mplitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with notch

Time(second)

Am

plitude(pu)

2.6 OSCILLATORY TRANSIENT

Fig. 13.Voltage signal with oscillatory transient

Transient disturbance occurs due to irregular switching of equipments. This type of disturbance mainly occurs in

energizing characteristics of transformer. This phenomenon is familiar with electrical equipments producing

harmonics. These disturbances results in unusual energy wastage and insulation breakdown. They may also switch

off a load in case of excessive transiency of the electrical signal.

It is defined as a sudden; non-power frequency change in the steady state condition of voltage, current or both

that have both positive and negative values. It exhibits a damped oscillation, with a voltage or current waveform

continuing to oscillate for half a cycle to three cycles. Oscillatory transients event occur due to switching of circuits,

specially the energizing and de-energizing of capacitors and inductors.

The disturbance can cause temporary or permanent damage to electrical equipment. A transient overvoltage can

damage components, especially semiconductors, which are not designed to withstand over voltages. They can cause

data errors in data processing and storage equipment because transients in electrical power circuits can couple to

communications and control circuits or cause adjustable drives to trip accidentally.

24

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5

Oscillatory transient

Time(second)

Am

plitu

de(p

u)

Fig.14 Samples of voltage signal with oscillatory transient

25

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5

2Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with transient

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.5

0

0.5

1

1.5Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5

2Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1

1.5Voltage signal with transient

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1.5

-1

-0.5

0

0.5

1Voltage signal with transient

Time(second)

Am

plitude(pu)

2.7 SPIKE

Fig. 15 Voltage signal with spike

Voltage spikes are sudden and momentary increase in voltage. It may be caused by lightning strikes power

outages, tripping of circuit breakers, short circuits, power transitions in other large equipment on the same

power line etc. Due to voltage surge electronic devices may operate at decreased efficiencies.

Spikes permanently damage the equipments.

26

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

Spike

Time(second)

Am

plitu

de(p

u)

Fig. 16 Samples of voltage signal with spike

27

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.5

0

0.5

1

1.5Voltage signal with spike

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with spike

Time(second)

Am

plitude(pu)

2.8 INTERRUPTION

Fig.17 Voltage signal with interruption

It may be seen as a loss of voltage in power system. This disturbance describes a drop of 0.9-1 pu of rated

voltage for duration of 0.5 cycles to 1 minute as per IEEE 1995. It occurs from accidents like faults and

component malfunctions. Short voltage interruptions are typically the result of a malfunction of a switching

device or a deliberate or inadvertent operation of a fuse, circuit breaker in response to faults and disturbance.

Thus a voltage interruption can cause a permanent shutdown of equipment and also can lead to damage. A

voltage interruption over a large geographical area lasting for long time is usually undesirable.

28

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Interruption

Time(second)

Am

plitu

de(p

u)

Fig. 18 Samples of voltage signal with interruption

29

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)A

mplitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with interruption

Time(second)

Am

plitu

de(p

u)

2.9 TEMPORARY OUTAGE

Fig.19 Voltage signal with outage

Outage is defined as a temporary disconnection of load from supply. This event stops electrical equipments

that are running. Sudden outage may damage mechanical parts of electrical motors. This occurs due to

undesirable governor action of alternator. If the governor does not open properly this temporary

disconnection is observed.

Revival from outage takes a long time. During this situation all load are disconnected from transformer

and after repair of governor control system the substations are informed via telemetry system to connect

load. A outage for long time from the supply is usually termed as BLACKOUT, which in undesirable.

30

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Outage

Time(second)

Am

plit

ude(p

u)

Fig. 20 Samples of voltage signal with temporary outage

31

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)A

mplitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitude(pu)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitu

de(p

u)

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Voltage signal with temporary outage

Time(second)

Am

plitude(pu)

Fig. 21 Flowchart for generation of electrical disturbance signal and formation of feature matrix

32

3. NECESSITY AND REQUIREMENTS FOR DISTURBANCE CLASSIFICATION

Disturbances are some undesirable events occurring in electrical signal which enhance unusual wastage of energy. Fault

can occur at any time with disruptive operation of instruments. Electrical energy finds no substitute of its existence as it

can be easily generated and transferred to distant places.

Several disturbances have their own mitigation methods. It is necessary to identify type of electrical disturbance

so as to ensure the type of mitigation method required.

The process or methods employed here for classification of faults are given below:

Several types (9) of disturbances are generated using modern simulation software (MATLAB) as occurs in real life

situation.

Signal processing techniques are used to correlate fault signal and normal electrical power signal. Here, CROSS-

CORRELATION technique is used.

Twelve features are extracted from cross-correlogram. These features are distinct with respect to type of

disturbance.

Three modern classification techniques like ARTIFICIAL NEURAL NETWORK (ANN), PROBABILISTIC NEURAL

NETWORK (PNN) and RADIAL BASIS NEURAL NETWORK (RBFNN) are used to classify disturbance.

The fault classification is necessary for the exact mitigation methods that is to be applied for type of

disturbance occurring. Nowadays PLC based mitigation methods are widely used where the location and type of

faults are sent to control room of generating or substation by telemetry system. The fault types are computed in

microprocessor based PLC based on logic provided for classification.

It is responsibility of power utility company to assure reliability of power to their consumers. Hence,

disturbance free power signal is expected from these companies. Since, disturbance is inheritable occurance,

thus detection and classification of fault is indispensible for provision of power reliability to consumers.

33

4. FEATURES EXTRACTION FROM ELECTRICAL DISTURBANCE SIGNAL

Several methods were proposed by earlier researchers for features extraction from the electrical disturbance signal. The

approach by Haibo He et. Al. in [4] utilizes the CROSS-WAVELET TRANSFORMATION TECHIQUE based on MULTI-

RESOLUTION ANALYSIS (MRA) for features extraction form the electrical disturbance signal achieving an accuracy of

94.3% in 2006. Noble attempt by Inigo Monedero, Carlos Leon et.al.in [6] classifies the electrical fault signal with an

accuracy of 89% using WAVELET DECOMPOSITION for feature extraction and ARTIFICIAL NEURAL NETWORK (ANN) for

classification of the fault signal. The scheme of Fengzhan Zhao et. Al. in [8] detects disturbance in electrical power signals

and classify them using the S-TRANSFORMATIONTECHNIQUE with an accuracy of 89.5%. S. Suja et. Al. in [9] are capable

of classifying power signal disturbance based on features extracted using DISCRETE WAVELET TRANSFORMATION

(DWT) with accuracy of 86%.

But the recent year’s attempts by Suryannarayana Chandaka et.al.in [12], P. Rajamani et.al. in [14] and Sovan Dalai et.

Al. in [16] used CROSS-CORRELATION TECHNIQUE for feature extraction form the disturbance signal which increases the

accuracy percentage to 98% approximately.

The present attempt of ELECTRICAL DISTURBANCE CLASSIFICATION also utilizes CROSS-CORRELATION TECHNIQUE for

features extraction from the disturbance signal.

4.1 CORRELATION

Correlation interprets statistical relation between different broad classes of dependent objects. They indicate predictive

relationship that can be dispensed in practice. DEPENDENCE refers to any situation in which random variables do not

satisfy a mathematical condition of probabilistic independence. There are several methods for measuring the degree of

correlation. Among them PEARSON CORRELATION TECHNIQUE is used broadly for measuring degree correlation

between two different situations or sets of data.

The population correlation coefficient 𝝆(𝒙, 𝒚) between two random sets of variable 𝑥and 𝑦 with expected values µ𝑥 and µ𝑦 and standard deviations 𝜎𝑥 and 𝜎𝑦 is defined as:

𝝆(𝒙, 𝒚) = 𝒄𝒐𝒓𝒓(𝒙, 𝒚) =𝒄𝒐𝒗(𝒙,𝒚)

𝝈𝒙𝝈𝒚=

𝑬[(𝑿−µ𝒙)(𝒀−µ𝒚)]

𝝈𝒙𝝈𝒚

Now there are two types of correlation technique proposed by PEARSON. They are auto-correlation and cross-correlation.

34

4.1.1 AUTO-CORRELATION

When the two same signals are correlated with provision of arbitrary lag to the sample of disturbance signal, the

resulting sequence is called AUTO-CORRELATION SEQUENCE.

N−m−1

Cxx(m)= Σ Xn+m X n, m ≥ 0 n=0 Cxx(-m)=0, m <0. Where m indicates the shift parameter of correlation sequence and Cxx(m)indicates the auto-correlation sequence of the signal. If the signal has N no. of samples, then the correlation correlation sequence would have (2N-1) no. of samples. 4.1.2 CROSS-CORRELATION

When the two different signals are correlated with provision of arbitrary lag to the sample of disturbance signal, the resulting sequence is called CROSS-CORRELATION SEQUENCE.

N−m−1 Cxy(m)= Σ Xn+m X n, m ≥ 0 n=0 Cxy(-m)=0, m <0 Where m indicated the shift parameter of correlation sequence and Cxy(m) indicates the cross-correlation sequence of the signal. If the signal has N no. of samples, then the correlation correlation sequence would have (2N-1) no. of samples.

Fig. 22. Plot of a typical cross-correlogram

The present problem utilizes 10 cycles of power frequency electrical signal during multiple electrical disturbance signal data acquisition and analysis. Thus 20 half cycles of power frequency exist in the recorded signals. These cycles are utilized for cross-correlation between electrical power signal and electrical disturbance signal as shown in Fig. 14 below.

35

Fig. 23.Cross-correlogram of voltage sag signal

Several features were extracted from cross-correlation sequence between electrical disturbance signal and electrical

power signal. They are designated in Table II as F1-F12 listed below:

Table II FEATURE DESIGNATION LIST FOR DISTURBANCE SIGNAL

Sl. No. Feature Index Feature Designator

1 F1 Maximum value of correlation sequence

2 F2 Index of the maximum value of correlation sequence

3 F3 Equivalent width of correlation sequence

4 F4 Mean value of correlation sequence

5 F5 Standard deviation of correlation sequence

6 F6 Skewness of correlation sequence

7 F7 Kurtosis of correlation sequence

8 F8 Variance of disturbance signal

9 F9 Kurtosis of disturbance signal

10 F10 Kurtosis of correlation coefficient

11 F11 Variance of disturbance signal

12 F12 Kurtosis of disturbance signal

The mathematical interpretation pertaining to features F3-F10 and F12 are formulated below:

1. F3=[∑ 𝒏∗𝑪𝒏𝑵

𝒏=−𝑵 ]

𝑹𝒎𝒂𝒙

2. F4=[∑ 𝒏∗𝑪𝒏𝑵

𝒏=−𝑵 ]

[∑ 𝑪𝒏𝑵𝒏=−𝑵 ]

3. F5=[∑ |𝒏|∗𝑪𝒏𝑵

𝒏=−𝑵 ]

∑ 𝑹𝒏𝑵𝒏=−𝑵

4. F6=√[∑ 𝒏∗𝒏∗𝑪𝒏𝑵

𝒏=−𝑵 ]

∑ 𝑪𝒏𝑵𝒏=−𝑵

5. F7=Σ𝑪𝒏

(𝟐𝑵+𝟏)

6. F8=𝜮(𝑪𝒏−𝑭𝟕)∗(𝑪𝒏−𝑭𝟕)

(𝟐𝑵+𝟏)

7. F9=𝜮(𝑪𝒏−𝑭𝟕)∗(𝑪𝒏−𝑭𝟕)∗(𝑪𝒏−𝑭𝟕)

𝟐𝑵∗𝑭𝟖∗𝑭𝟖∗𝑭𝟖

8. F10=𝜮(𝑪𝒏−𝑭𝟕)∗(𝑪𝒏−𝑭𝟕)∗(𝑪𝒏−𝑭𝟕)∗(𝑪𝒏−𝑭𝟕)

𝟐𝑵∗𝑭𝟖∗𝑭𝟖∗𝑭𝟖∗𝑭𝟖

9. F12=𝜮(𝒀−𝒚)∗(𝒀−𝒚)∗(𝒀−𝒚)∗(𝒀−𝒚)

𝟐𝑵∗(𝒀𝒔∗𝒀𝒔∗𝒀𝒔∗𝒀𝒔)

where 𝒏 represents index of correlation coefficients,𝑵 represent no. of samples in electrical power signal,𝑪𝒏 represents correlation coefficient and 𝒀𝒔 is standard deviation of disturbance signal. The above 12 distinct features were extracted from disturbance signal after cross-correlation operation is employed.

Features to be chosen are calculated by examination process and utilizing previous experience.

36

However, user may apply different other features depending on the hurdle passes by the user. These features are extracted for all the electrical disturbances that are used to train and they are stored in a matrix as input of ANN based feature selection method. Large amount of data is required for training and testing. Electrical disturbance data obtained either from condition monitoring of different electrical equipment for long period of time or it is obtained directly from monitoring of electrical power signal of the any large industrial plant.

The present method is an initiative to create a prototype for monitoring electrical disturbance signal. Several fault

signals are generated by simulation software using suitable IEEE standard 1159-1995and mathematical model with

some constraints. Table I gives the mathematical model of electrical fault signals.

Table III

DESIGNATION OF ELECTRICAL DISTURBANCE

Sl. No. ELECTRICAL DISTURBANCE CLASS

COMBINATION OF DISTURBANCE

1 C1 Sag+Noise

2 C2 Swell+Noise

3 C3 Transient+Noise

4 C4 Flicker+Noise

5 C5 Harmonic+Noise

6 C6 Notch+Noise

7 C7 Spike+Noise

8 C9 Interruption+Noise

9 C9 Temporary outage +Noise

The classifier machine employs here to classify the above mentioned 9 types of electrical disturbances occurring in electrical signal.

37

5. FEATURE SELECTION AND DISTURBANCE CLASSIFICATION TECHNIQUE

Experts used several feature selection and classification techniques in earlier years with acceptable accuracy. Haibo He

et. Al. [4] in 2006 uses SELF-ORGANISING LEARNING ARRAY (SOLAR) system yielding classification accuracy of 94.3%.

Electrical disturbances are classified with accuracy of 89% Inigo Monedero, Carlos Leon et.al in [6] in the same year.

Scheme formulated by Fengzhan Zhao et. Al. in [8] classifies the disturbance signal with accuracy of 98.5% of SNR=30Db

utilizing S-TRANSFORMATION technique in 2007. The succeeding years witness the electrical disturbance classification

accuracy of 86%, 90.70%, 96.3%, 97.10%, and 99.09% by S. Suja et. Al., P. Rajamani et. Al., Chun-Yao Lee et. Al., Sovan

Dalai et. Al., Sovan Dalai et. Al. in [7], [14], [15], [17] and [21] respectively.

The present scheme is a comparative study of classification of electrical disturbances accuracy between

ARTIFICIAL NEURAL NETWORK (ANN)

PROBABILISTIC NEURAL NETWORK (PNN)

RADIAL BASIS NEURAL NETWORK (RBN)

Fig. 24. SVM divides space in 12D pertaining to each class of disturbance

38

5.1 ARTIFICAL NEURAL NETWORK

In machine learning and cognitive science, artificial neural networks (ANNs) are a family of models inspired by biological

neural networks (the central nervous systems of animals, in particular the brain) which are used to estimate

or approximate functions that can depend on a large number of inputs and are generally unknown. Artificial neural

networks are generally presented as systems of interconnected “neurons” which exchange messages between each

other. The connections have numeric weights that can be tuned based on experience, making neural nets adaptive to

inputs and capable of learning.

For example, a neural network for handwriting recognition is defined by a set of input neurons which may be activated

by the pixels of an input image. After being weighted and transformed by a function (determined by the network’s

designer), the activations of these neurons are then passed on to other neurons. This process is repeated until finally,

the output neuron that determines which character was read is activated.

Like other machine learning methods – systems that learn from data – neural networks have been used to solve a wide

variety of tasks, like computer-version and speech-recognition that are hard to solve using ordinary rule-based

programming.

Examinations of humans’ central nervous systems inspired the concept of artificial neural networks. In an artificial neural

network, simple artificial nodes, known as “neurons”, “neurodes”, “processing elements” or “units”, are connected

together to form a network which mimics a biological neural network.

There is no single formal definition of what an artificial neural network is. However, a class of statistical models may

commonly be called “neural” if it possesses the following characteristics:

1. contains sets of adaptive weights, i.e. numerical parameters that are tuned by a learning algorithm, and

2. is capable of approximating non-linear functions of their inputs.

The adaptive weights can be thought of as connection strengths between neurons, which are activated during training

and prediction.

Artificial neural networks are similar to biological neural networks in the performing by its units of functions collectively

and in parallel, rather than by a clear delineation of subtasks to which individual units are assigned. The term “neural

network” usually refers to models employed in statistics, cognitive psychology and artificial intelligence. Neural network

models which command the central nervous system and the rest of the brain are part of theoretical

neuroscience and computational neuroscience.

39

In modern software implementations of artificial neural networks, the approach inspired by biology has been largely

abandoned for a more practical approach based on statistics and signal processing. In some of these systems, neural

networks or parts of neural networks (like artificial neurons) form components in larger systems that combine both

adaptive and non-adaptive elements. While the more general approach of such systems is more suitable for real-world

problem solving, it has little to do with the traditional, artificial intelligence connectionist models.

What they do have in common, however, is the principle of non-linear, distributed, parallel and local processing and

adaptation.

Historically, the use of neural network models marked a directional shift in the late eighties from high-level (symbolic)

artificial intelligence, characterized by expert systems with knowledge embodied in if-then rules, to low-level (sub-

symbolic) machine-learning, characterized by knowledge embodied in the parameters of a dynamical representations.

40

5.2 PROBABILISTIC NEURAL NETWORK (PNN)

It is acronym of PROBABILISTIC NEURAL NETWORK. With context to real time electrical disturbance classification PNN

has proven to be more effective than traditional back propagation neural networks. In order to a feature pattern vector f

€ Fm, to assign the definite among predefined classes the conditional density P (f¦ Ck) of each class Ck is estimated since

it represents the uncertainty associated to class attribution. They are combined according to Bayes theorem that makes

optimal decision. Conditional density estimation is achieved by implementation of Parzen window. It can be viewed as a

sphere of influence p(s, x) around each training (known) sample s and to add them up for each of the k classes

P(x|C ) = p(s , x), p (s, x) = exp (-||x-s||2/2s 2) where the only free parameter is the width σ of the Gaussians.

ARCHITECTURE OF PNN

It consists of a node for each layer one for each training samples. Here 900 training samples are used for each class of

electrical disturbance. The weights leading from the input to a layer one node are the coordinates of the corresponding

sample. The node computes the distance d(s, x) from the test vector x to the training sample s and outputs the value of

the Gaussian.

The experimental results have been carried and it verify the ability of modified PNN in achieving good classification rate

in compared to traditional PNN or back propagation Neural Networks.

This paper presents a PNN based classification technique in order to compare with other classification process. The

pictorial representation of a basic PNN is given below.

Fig.25Basic architecture of PNN

41

5.3 RADIAL BASIS FUNCTION NEURAL NETWORK (RBFNN)

A radial basis function network is an artificial neural network that works on radial basis functions as activation functions. The

output from the network is a linear combination of radial basis functions and its neuron parameters. Radial basis function

networks finds its applications in function approximation, time series prediction, classification, and system control.

NETWORK ARCHITECTURE

Fig.26Architecture of RADIAL BASIS NEURAL NETWORK (RBNN)

Radial basis function (RBF) networks typically have three layers: an input layer, a hidden layer with a non-linear RBF activation

function and a linear output layer. The input can be modeled as a vector of real numbers .The output of the network is

then a scalar function of the input vector, , and is given by

where is the number of neurons in the hidden layer, is the center vector for neuron , and is the weight of

neuron in the linear output neuron. Functions that depend only on the distance from a center vector are radially symmetric

about that vector, hence the name radial basis function. In the basic form all inputs are connected to each hidden neuron.

The norm is typically taken to be the Eucledian distance (although the Mahalanobish distance appears to perform better in

general) and the radial basis function is commonly taken to be Gaussian

.

42

The Gaussian basis functions are local to the center vector in the sense that

i.e. changing parameters of one neuron has only a small effect for input values that are far away from the center of

that neuron.

43

6. BLOCK DIAGRAM FOR ELECTRICAL DISTURBANCE CLASSIFICATION

The above figure represents block diagram for the whole process of electrical faults classification. The process

commences with parameter initialization of electrical disturbance and proceeded by generation of samples of faults

giving range of parameter throughout the samples generated. The next task is feature extraction using signal processing

technique. Here cross-correlation technique is used. The feature matrix is formed for training neural network. The same

process is applied to create feature matrix for testing the classification accuracy. The training matrix is used to train the

neural network specifying definite class corresponding to feature matrix.

44

7. RESULT ANALYSIS Table IV

AVERAGE CLASSIFICATION ACCURACY FOR ANN WITH 0db, 30db AND 50db NOISE

45

CLASS SAG

(%)

SWELL

(%)

HARMONIC

(%)

FLICKER

(%)

NOTCH

(%)

SPIKE

(%)

TRANSIENT

(%)

INTERRUPTION

(%)

OUTAGE

(%)

SAG 95 0 0 0 0 0 0 0 0

SWELL 0 80 0 0 0 0 0 0 0

HARMONIC 0 0 99 0 0 0 0 0 0

FLICKER 0 0 0 91 0 0 0 0 0

NOTCH 0 0 0 0 100 0 0 0 0

SPIKE 0 0 0 0 0 87 0 0 0

TRANSIENT 0 0 0 0 0 0 100 0 0

INTERRUPTION 0 0 0 0 0 0 0 80 0

OUTAGE 0 0 0 0 0 0 0 0 81

8. CONCLUSION The presented method compares four types of classification techniques. Soft computation finds its

essential and efficient use in electrical fault classification. Among

the above methods mentioned, artificial neural network (ANN) classification is widely applied for its

robustness and low computational techniques.

Classification accuracy also draws attention on comparing noiseless fault signal and single electrical

disturbance signal. It reduces on introduction of noise. At higher noise level (db) the classification

accuracy reduces. This is due to the fact that on introduction of noise to fault signal the feature

matrix’s value changes drastically which becomes almost illegible the type of electrical faults occurring

by the neural network.

Proposed method classified nine types of single electrical disturbances. Accuracy increases on

iterative training of neural network with feature matrix of training sample of disturbance signal.

The most efficient soft classification method is as its classification accuracy is. The judgement on

best classification technique is based on average classification accuracy.

The above classification technique has been arranged in order of their decreasing average classification

accuracy:

Table V

ARRANGEMENT OF AVERAGE CLASSIFICATION ACCURACY

Sl. No. CLASSIFICATION TECHNIQUE AVERAGE ACCURACY

(%)

1 ARTIFICAL NEURAL NETWORK (ANN) 90.3

2 RADIAL BASIS FUNCTION (RBFNN) 94.3

3 PROBABILISTIC NEURAL NETWORK (PNN) 91

From the table, it can be concluded that RBFNN has more classification accuracy than other classification techniques

used. It has an average classification accuracy of 94.3%.

46

9. FUTURE DEVELOPMENT

Disturbance in electrical signal occur in association with several other types of disturbances. Those types of

signal are called multiple disturbance signals. Now the above classification techniques can be applied to

multiple disturbance signal and search for the most efficient method for fault classification.

As these classification has become more common, new and advanced classification algorithm needs

to be evolved which has higher classification accuracy and takes low computational time.

A new technique needs to be evolved which works in online classification of fault. GSM based

technique can be used to obtain the type of fault that occurred in the power system. But nowadays PLC based

telemetry system is used to for signal communication purpose. GSM technique can be used for signal

communication system. GSM is faster in communication process. Wi-Fi technique can also be utilized for signal

communication purpose.

A recent theory by John Nash for which he received Nobel Prize in year 1991 can be also used for

classification. The theory is called GAME THEORY. It is feature selection algorithm where significant features

are selected to get more classification accuracy. This theory reduces computational time required to classify

the disturbance signal as the feature matrix gets reduced and also memory requirement squeezed and

compact. This is completely a new theory to this area of power system.

47

10. REFERENCES

[1]A. M. Youssef, T. K. Abdel-Galil, E. F. El-Saadany and M. M. A. Salama, “Disturbance Classification Utilizing Time

Warping Classifier,” IEEE Transactions on Power Delivery, vol. 19, no. 1, pp. 272-278, Jan. 2004

[2]P. K. Dash and M. V. Chilukuri, “Hybrid S-Transform and Kalman Filtering for Detection and Measurement of Short

Duration Disturbances in Power Networks,” IEEE Transactions on Instrumentation and Measurement, vol. 53, no. 2, pp.

588-596, April 2004

[3]T. K. Abdel-Galil, M. Kamel, A. M. Youssef, E. F. El-Saadany, M. M. A. Salama, “Power Quality Classification Using

Inductive Inference Approach,” IEEE Transactions on Power Delivery, vol. 19. No. 4, pp. 1812-1818, Oct. 2004

[4]Haibo He and Janusz A. Starzyk , “A Self Organizing Learning Array System For Power Quality Classification Based on

Wavelet Transform,” IEEE Journals on Power Delivery, vol.21, no.1, pp. 286-295, Jan. 2006

[5]ÖmerNezihGerek and DoganGökhanEce, “Power-Quality Event Analysis Using Higher Order Cumulants and Quadratic

Classifiers,” IEEE Transactions on Power Delivery, vol. 21, no. 2, pp. 883-889, April, 2006

[6]InigoMonedero, Carlos Leon, Jorge Ropero, Antonio Garcia, Jose Manuel Elena, Juan C. Montano, “Classification of

Electrical Disturbances in Real Time Using Neural Networks,” IEEE Transaction on Power Delivery, pp. 1-9, September

2006

[7]ÖmerNezihGerek ,DoganGökhanEce and AtalayBarkana, “Covariance Analysis of Voltage Waveform Signature for

Power-Quality Event Classification,” IEEE Transactions on Power Delivery, vol. 21, no. 4, pp. 2022-2031, October 2006

[8]Fengzhan Zhao and Rengang Yang, “Power Quality Disturbance Recognition Using S-Transform,” IEEE Transactions on

Power Delivery, vol.22, no.2, pp. 944-950, April 2007

[9]S.Suja and Jovitha Jerome, “Power Signal Disturbance Classification Using Wavelet Based Neural Network,” Serbian

Journal Of Electrical Engineering, vol.4, no. 1, pp. 71-83, June 2007

[10]Mamun Bin IbneReaz, Florence Choong, MohdSahimanSulaiman, Faisal Mohd-Yasin and Masaru Kamada, “Expert

System for Power Quality Disturbance Classifier,” IEEE Transactions on Power Delivery, vol. 22, no. 3, pp. 1979-1988, July

2007

[11]Ameen A. Gargoom, NesimiErtugrul, Wen. L. Soong, “Automatic Classification and Characterization of Power Quality

Events,” IEEE Transactions on Power Delivery, vol. 23, no. 4, pp. 2417-2425, Oct. 2008

48

[12]SuryannarayanaChandaka, AmitavaChatterjee, SugataMunshi, “Support Vector Machine Employing Cross-

Correlation for Emotional Speech Recognition,” Elsevier System with Application 36 (2009) pp. 611-618

[13]Chiung-Chou Liao and Hong-Tzer Yang, “Recognizing Noise-Influenced Power Quality Events with Integrated Feature

Extraction and Neuro-Fuzzy Neworks,” IEEE Transactions on Power Delivery, vol. 24, no. 4, pp. 2132-2141, Oct. 2009

[14]P. Rajamani, DebangshuDey and SivajiChakravorti, “Cross-Correlation Aided Wavelet Network for Classification of

Dynamic Insulation Failures in Transformer Winding during Impulse Test,” IEEE Transactions on Dielectrics and Electrical

Insulation, vol. 18, no. 2, pp. 521-532, April 2011

[15]Chun-Yao Lee and Yi-Xing Shen, “Optimal Feature Selection for Power Quality Disturbances Classification,” IEEE

Transactions on Power Delivery, vol. 26, no. 4, pp. 2342-2351, October 2011

[16]S. Dalai, D. Dey, B. Chatterjee, S. Chakravorti and K. Bhattacharya, “Rough-Set Based Feature Selection and

Classification for Power Quality Sensing Device Employing Correlation Technique,” IEEE Sensors Journal, vol. 13, no. 2,

pp. 563-573, February 2013

[17]Milan Biswal and P. K. Dash, “Measurement and Classification of Simultaneous Power Signal Patterns With an S-

transform Variant and Fuzzy Decision Tree,” IEEE Transactions on Industrial Informatics, vol. 9, no. 4, pp. 1819-1827,

Nov. 2013

[18]Shufan He, Kaicheng Li and Ming Zhang, “A Real-Time Power Quality Disturbances Classification Using Hybrid

Method Based on S-transform and Dynamics,” IEEE Transactions on Instrumentation and Measurement, vol. 62, no. 9,

pp. 2465-2475, Sept. 2013

[19]Swati Banerjee and MadhuchhandaMitra, “Application of Cross-Wavelet Transform for ECG Pattern Analysis and

Classification,” IEEE Transactions on Instrumentation and Measurement, vol. 63, no. 2, pp. 326-333, February 2014

[20]Martin Valtierra-Rodriguez, Rene de Jesus Romero Troncoso, Roque Alfredo Osornio-Rios and Arturo Garcia Perez,

“Detection and Classification of Single and Multiple Power Quality Disturbances Using Neural Network,”IEEE

Transactions on Industrial Electronics, vol. 61, no. 5, pp. 2473-2482, May 2014

[21]S. Dalai, D. Dey, B. Chatterjee, S. Chakravorti and K. Bhattacharya, “Cross-Spectrum Analysis-Based Scheme for

Multiple Power Quality Disturbance Sensing Device,” IEEE Sensors Journal, vol. 15, no. 7, July 2015

[22]Zhigang Liu, Yan Cui and Wenhui Li, “A Classification Method for Complex Power Quality Disturbances Using EEMD

and Rank Based Wavelet,” IEEE Transactions on Smart Grid, vol. 6, no. 4, pp. 1678-1685, July 2015

49

APPENDIX 1

LIST OF FIGURES

Sl.

No.

Figure

Number

Figure Name Page No.

1 1 Voltage signal with sag 12

2 2 Voltage with Sag due to starting of induction motor 13

3 3 Voltage with Sag due to energizing of unloaded

transformer

14

4 4 Samples of voltage signal with sag 15

5 5 Voltage signal with Swell due to switching of large

charged capacitor

16

6 6 Samples of voltage signal with swell 17

7 7 Voltage signal with Harmonic 18

8 8 Samples of voltage signal with harmonic 19

9 9 Voltage signal with flicker 20

10 10 Samples of voltage signal with flicker 21

11 11 Voltage signal with notch 22

12

12 Samples voltage signal with notch 23

13

13 Voltage signal with oscillatory transient 24

14 14 Samples of voltage signal with oscillatory transient 25

15 15 Voltage signal with spike 26

16

16

Samples of voltage signal with spike 27

17 17 Voltage with interruption 28

50

18 18 Sample of voltage signal with interruption 29

19 19 Voltage signal with outage 30

20 20 Sample of voltage signals with temporary outage 31

21 21 Flowchart for generation of electrical disturbance signal and

formation of feature matrix

32

22 22 Plot of a typical cross-correlogram 35

23 23 Cross-correlogram of voltage sag signal 36

24 24 SVM divides space in 12D pertaining to each class of

disturbance

38

25 25 Basic architecture of PNN 41

26 26 Architecture of RADIAL BASIS NEURAL NETWORK (RBNN)

42

51

APPENDIX 2

LIST OF TABLES

Sl. No. Table No. Table Name Page No.

1 I MATHEMATICAL FORMULATION OF

DIFFERENT ELECTRICAL DISTURBANCE

SIGNAL

11

2 II FEATURE DESIGNATION OF DIRRENT

DISTURBANCE SIGNAL

36

3 III DESIGNATION OF ELECTRICAL DISTURBANCE 37

4 IV AVERAGE CLASSIFICATION

ACCURACY FOR ANN WITH 0db,

30db AND 50db NOISE

45

5

V

ARRANGEMENT OF AVERAGE

CLASSIFICATION ACCURACY

46

52