Partial discharge source classification by support vector machine

Abstract

The interpretation and recognition of partial discharges due to different sources were studied by Support Vector Machine, which is a statistical learning technique. The classification based on the Support Vector Machine technique, requires a preprocessing of the input data for different partial discharge sources, obtained by phase resolved partial discharge analysis. The phase resolved partial discharge data was divided into smaller phase windows and the average magnitude, maximum magnitude and the number count in each phase window is determined. The high dimensionality of the feature data set is reduced by adopting Principle Component Analysis method. This reduced feature data set is given as the input vector to the SVM and the partial discharge sources are classified by adopting, Radial Basis Function kernel.