The Classification of Power Quality Disturbances using Statistical S-Transform and Probabilistic Neural Network

Abstract

This article compares the ability of the Probabilistic Neural Network in the classification of several Power Quality Disturbances (PQD) using statistical parameters. The objective is to investigate the effectiveness of the classifier in modeling the low-dimensional feature vectors describing several PQD disturbances. In the process, several statistical parameters such as the mean, RMS value, standard deviation, skewness, Kurtosis, form factor, Crest factor, Energy, normalized entropy, log entropy, and Shannon entropy have been extracted using the Feature vectors of the well-known Stockwell Transform (ST). The statistical coefficients corresponding to ten-PQDs have been fetched and fed to the chosen PNN for efficient modeling. A comparison of the recognition accuracy of the PQDs has been made to that of the conventional statistical parameters extracted directly from the synthetic raw signals. The ST statistical parameters have shown to outperform with an average recognition accuracy of 92.6%. On the contrary, the conventional statistical parameters have provided a lower accuracy of 79.5%. In the case of PNN, the number of hidden layer neurons is made equal to the number of training data. A suitable selection of the spread factor leads to better recognition accuracy as revealed from our results.