Partial discharge classification using principal component transformation

Abstract

Texture analysis algorithms have been applied to generate different features for identifying partial discharge (PD) sources. The algorithms utilised are the spatial gray-level dependence method, gray-level difference histogram method, gray-level run-length method and the power spectrum method. To reduce the identification time, it is important to minimise the number of features used to describe the PD sources in the feature space. Principal component transformation has been applied as a feature reduction technique on the features obtained from the texture analysis algorithms. The classification accuracy of the principal components generated has been established using a minimum-distance classifier for six different types of PD sources created in the laboratory as well as for a practical case simulating two types of PD on an actual cable length.