Separation and Classification of Concurrent Partial Discharge Signals Using Statistical-Based Feature Analysis

Abstract

In this paper, an algorithm for feature extraction and classification of high-pressure gas insulation system defects based on statistical analysis of time-domain parameters of partial discharge (PD) signals is presented. The algorithm focuses on the measurement and interpretation of PD signals in the time domain in completion of our previous paper on multiple-source phase resolved patterns. In this procedure, the PD measurements are conducted in different artificial defects that commonly happen in gas-insulated substations (GIS) such as corona, moving particles, floating electrodes, and metallic protrusions. To overcome the noise problem, wavelet transform (WT) technique is applied on the recorded signals. The PD pulse waveform parameters, namely rise time, fall time, slew rate, and pulse width are calculated, investigated and used as the discriminative features to represent each type of PD signals. The separation of PD sources is implemented based on Weibull distribution and K-means unsupervised clustering technique. The higher level statistical based features of each cluster are calculated, studied and used as the inputs of a kernel support vector machine (KSVM) classifier in order to classify multiple PD sources in a robust way. The results of this work demonstrate that the presented probabilistic diagnostic algorithm to extract features from time-domain PD pulse waveforms and their corresponding probabilistic distribution can be employed to cluster and classify PD signals based on their source of origin.