Partial discharge classification using probabilistic neural network model

Abstract

The aim of this paper is to propose the probabilistic neural network (PNN) model for classification partial discharge (PD) patterns, which comprised of corona discharge at high voltage side and at low voltage side in air, corona discharge at high voltage side and at low voltage side in mineral oil and surface discharge in mineral oil. Partial discharge signals were investigated by conventional method according to IEC60270. Independent parameters such as skewness, kurtosis, asymmetry, and cross correlation of the Φ-q-n PD patterns were analyzed. The PNN PD classification model was constructed. Moreover, the principal component analysis (PCA) was utilized to reduce the input dimension of the developed PD classification model. After that, 60% of the experimented data was used as a training data for the PD classification models. Another 40% experimented data was used for evaluation the performance of the designed PD classification models. Effects of spread parameters and input neuron numbers on the PD classification performance were examined. It was found that the first four score variable was appropriate to be used to construct the designed PNN model with the optimal spread value of 1.2. The proposed PD classification model can classify PD types with the accuracy of 100% of 40 tested data.