Wavelet Based Detection and Classification Power Quality

Disturbance using SVM and PSO

1DR. P. KANIRAJAN

1Department of Electrical and Electronics Engineering,

NPR College of Engineering and Technology, Natham, Tamilnadu, INDIA.

Abstract: This paper introduces a novel approach to detect and classify power quality disturbance in

(PSO). The simulation results possess significant improvement over existing methods in signal

detection and classification with lesser number of features

Keywords: Support Vector Machine, Radial Basis function Neural Networks, Wavelet

Transformation, Power Quality and Particle Swarm Optimization

Received: October 27, 2022. Revised: May 9, 2023. Accepted: June 15, 2023. Published: July 18, 2023.

1. Introduction

flicker, spike and transients causing problems

such as malfunction, instability, short life

times, failure of electrical equipment and so

on. Switching off of large load and

energization of large capacitor may cause

voltage swell. Whereas the faults leading to

voltage sag or momentary interruption,

harmonic distortion and notching in the

voltage and current are caused because of the

usage of solid state switching device and

In power system, these disturbances need to be

identified in order to improve the power

quality. PQ events identification is difficult

because it involves wide range of disturbance

categories. Therefore, the decision boundaries

of disturbance features may overlap. For these

reasons, the need of power quality analysis has

axis, while the localizing scalable Gaussian

window dilates and translates. As a result, the

phase spectrum is absolute in the sense that it

always referred to the origin of the time axis,

the fixed reference point. ST is found to be

superior (Edward reid, 1996). However, the

computational time is very large compared to

Wavelet Transform (WT), which is

undesirable for on-line applications. WT based

approach, such as Wavelet MRA, has been

al.,2003) and hence, provides an excellent time

frequency resolution(Chia-Hung Lina and

chia-Hao Wang,2006).This allows WT for

analysis of signals with localized disturbances

components and also for classifying low and

high frequency power quality problems. Using

the properties of WT technique and the feature

of the decomposed waveforms along with

ANN algorithm, it is possible to extract

important information from a disturbance

deviation can be considered as a measure of

energy signal with zero mean ( Gaing, 2004).

The classification of seven types of

PQ disturbances with self-organizing learning

array system considering 11 features, besides

22 families of wavelet are tested to identify the

best one for a better classification .

Classification of eleven types of PQ events

using wavelets and Probabilistic Neural

Network (PNN) is discussed (Mishra et

overhead. Further, eleven types of PQ events

are also classified with the help of ST and

To summarize, the paper shows the power

quality problems classification using WT and

SVM-PSO. First the work handles with

wavelet transformation and feature extraction

from WT needed by the neural networks for

training and for effective classification for all

the 8 types. Next the paper describes the

structure and results and discussion about

detection and classification PQ events using

SVM and similarly for SVM-PSO. Finally, the

2. Wavelet Transforms

Wavelet transformation has the ability to

analyse different power quality disturbances in

both time and frequency domain. The wavelet

transform is useful in extracting features of

various power quality disturbances. Wavelet

analysis deals with expansion of function in

terms of a set of basis function. However,



 (5)

Table 1: Results of Selection of Wavelet Function

power quality events are statistically different

from those that are calculated during an event

free time period.

In this paper, features based on mean, standard

deviation, Norm Entropy and Skewness of

transformed signals are extracted and energy at

each decomposition level, which has the

ability to quantify the magnitude of variation

extracted features help to distinguish one

extract feature of these signals, the standard

solution by nonlinearly projecting the input

space in to high dimension space. The kernel

function K (xi, x) the input vector and support

xi drawn from the test data. In the SVM the

optimal decision function constructed to

predict and classify accurately the un seen data

in to two classes and minimizes the

classification error. This is achieved by SRM(

V.N.Vapnik, 1998).The over fitting is

mitigated because of good generalization

that maximizes the margin as well as

minimizes the number of misclassified

patterns.

The optimal weight vector W* is givens in

equation (7).

the solution of quadratic programming

problem.

The classification of a new data vector x can

be done with equation (8 )

y=sign(f(x)) (8)

Where, f(x) is the optimal decision boundary

deserved from the set of training samples

󰇛󰇜 󰇛   󰇜

combined wavelet transformation with SVM

produce better detection and classification rate,

since it was trained with vast data with wide

variations. Whereas 10 numbers of orientation

may not be adequate in real cases for detection

and classifications of PQ events.

The PSO algorithm is different than any other

technique, rather than training one network

PSO trains a network of networks. It initializes

all weights to random values and starts

training each other, on each pass, PSO

compare the networks fitness. Each network

contains position and velocity. The position is

related to weight and the velocity refers to

updating of neural networks weights. Getting

variables are mean, standard deviation, energy,

Norm Entropy and Skewness. To speed up the

training, the variables are normalized. The

function of PSO is to get the best set of

weight. 80% of the generated inputs were used

for training and remaining 20% were used for

testing. For SVM-PSO with different initial

weight, a population of networks was

constructed and sum of square error in each

iteration over the training data set were

calculated and compared to find the best

is the ratio of correctly classified events to that

the of total number of events. The overall

classification accuracy is 98.75 %.

Table 5: Comparison of Proposed SVM on Number of Features with Other Work

From the Table 4 and Table 5 it is inferred that

the proposed wavelet based feature selection

gives better classification rate with lesser

number of features when compared with other

However, the signals in the real system will

always have noise. In order to test the

robustness of SVM and SVM-PSO, the white

is added to normal signal to test the

performance of SVM-PSO under noisy

environment. The signal to noise ratio (SNR)

.