Quantification of feature importance in automatic classification of power quality distortions

Automatic classification of power quality distortions has been studied extensively. Many studies adopted the Stockwell Transform as an appropriate signal processing technique. In this paper, features extracted from the Stockwell Transform are used in two classification techniques. Some of these features have not been seen before in any study on power quality classification. The contribution of this paper is the analysis of these features to determine their importance in classification results. This analysis is not common in power quality studies. As a result, a feature based on computing the contour of the third harmonic was found to be the most discriminant feature. For the study, datasets at different noise levels were generated using a public model. They were uploaded to a public repository to be reused by any interested researcher.