Optimizing Power Quality Disturbance Classification With Higher Order Statistics: Social And Economic Impacts

Abstract

In this research, four signal decomposition techniques (Empirical mode decomposition (EMD), Multivariate singular spectral analysis (MSSA), Wavelet Packet Decomposition (WPD), and Discrete wavelet transform (DWT) are studied for the optimal selection of signal processing technique to classify the Power Quality disturbances (PQD). Twelve types of single, multiple and synthetic PQD dataset is simulated from MATLAB R2020b, and real data is acquired from IEEE power quality guideline 1159.3-2019. Statistical parametric analysis for feature decomposition and selection are also explained as well. These statistical parameters are then subdivided into three groups to examine the contribution of each parameter to the selected feature extraction methods. However, MSSA and WPD have the highest accuracy of 99% and 99.9%, with the inclusion of higher-order statistical (HOS) features. Finally, features from each group were fed into a convolutional neural network (CNN) based classifier to classify the power quality disturbances. This study compares the selected techniques for optimal features with and without HOS and highlights the fundamental properties of each method. The proposed method was found to have reliable high classification accuracy under noisy and noiseless conditions.