COV Based Classification of PQ Disturbances Using Wavelet and Optimized SVM Hyper-Parameters

Abstract

This paper presents study of a feature, Coefficient of Variation (COV) based on wavelet transform (WT) for classification of power quality (PQ) disturbances. A five level wavelet decomposition using Db4 wavelet is chosen for all feature extraction processes. Six types of power quality disturbances (PQDs) have been considered. The feasibility of COV has been evaluated along with four other features using linear and Gaussian kernel support vector machine (SVM). Classification accuracy has been improved by optimizing the hyper-parameters for a Gaussian kernel SVM through Bayesian optimization (BO). BO-SVM provides better accuracy than a linear kernel SVM. From the experimental analysis it is revealed that the proposed method can effectively detect and identify PQDs.