Partial discharge classification using principal component analysis combined with self-organizing map

Abstract

This document proposes a statistical classification model using principal component analysis (PCA) for a data reduction approach combined with self-organizing map (SOM) for a classification purpose, so called, PCA-SOM model compared with SOM model to classify partial discharge pattern (PD) into four categories listed as corona at high voltage side, corona at low voltage side, surface discharge, and internal discharge. PD signals were investigated by using ultra high frequency (UHF) measurement technique. 12 independent parameters, skewness and kurtosis of each period of the measured electromagnetic signal, were calculated. 80 experiments in total were performed. PCA-SOM PD classification model was constructed. Then, 60% of the experimented data was used as a training data for the PD classification model. Another 40% experimented data was utilized to evaluate the performance of the designed PD classification model. Besides, noise signals were generated with a computer program and trained into the PD classification model as well. The peak of noise signal was set up at 10%, 20% and 30% of the peak value of the PD signal. These noise signals were added with the PD signals to generate a mixed noise - PD signal. Then, the mixed noise - PD signals were used to evaluate the performance of the PD classification models. It was found that the designed SOM model and PCA-SOM model can predict PD patterns without noise signal with the accuracy 100% of classification. The prediction ability of SOM for PD classification models decreased sharply when this model was tested by the mixed-PD signals with the noise level of 30% of the peak value of the PD signal. Whereas the PCA-SOM provided some degree accuracy reducing for PD pattern classification when it was verified with such data.