Generalized regression networks for partial discharge classification

Abstract

This document represents a partial discharge (PD) classification by using Generalized Regression Networks (GRNN) model. Two PD classification models, GRNN1 with 11 input variables and GRNN2 with 3 selected derived statistic parameters, were investigated for classification of PD signals into 5 patterns, corona at high voltage side in air, corona at low voltage side in air, corona at high voltage side in mineral oil, corona at low voltage side in mineral oil and surface discharge in mineral oil. The conventional PD measurement was performed for measuring PD signals of the artificial PD models. The statistical parameters of the PD signals such as skewness, kurtosis, asymmetry, cross correlation and so on were calculated from the developed computer program. Then, 60% of the experimented data was used as a training data for the developed PD classification models. Another 40% experimented data was used to evaluate the performance of the designed PD classification models. It was found that the GRNN1 model can classify PD patterns better than GRNN2 model. The accuracy for PD classification of GRNN1 model was 100% while the accuracy of GRNN2 model was 97.5% of 40 testing data.