Power quality event identification using wavelet packet transform: A comprehensive investigation

Abstract

The present study focuses on the selection of appropriate wavelet basis which would result in better classification of Power Quality (PQ) events. The accuracy of the classification is often critically dependent on the nature of wavelet basis and the induction algorithm. This study, therefore, comprehensively investigates the performance of several wavelet families including Daubechies, Coiflets, Symlets, Fejer-Korovkin, Bi-orthogonal and Reverse Bi-orthogonal from the prospective of PQ event identification. The performance of these wavelets was evaluated through fourteen distinct single and simultaneous PQ events which were generated following IEEE Std. 1159. Further, to investigate the interaction between induction algorithm and wavelet basis, two induction algorithms were included: k-Nearest Neighbor (k-NN) and Naive Bayes (NB). The results of the investigation convincingly demonstrate that 'db18' from the Daubechies family provides the best overall performance among 110 wavelet bases included in this study.