Partial discharge classification using learning vector quantization network model

This paper represents a partial discharge (PD) classification technique by using learning vector quantization (LVQ) network model. LVQ model was not only implemented with the PD test data but also combined with the principal component analysis (PCA), so called, PCA-LVQ for a data reduction. In this research work, both LVQ model and PCA-LVQ model were investigated. PD phenomena, corona at high voltage side in air (CHV), corona at low voltage side in air (CLV), surface discharge (SF), and internal discharge (IN) were simulated in the shielding room. Electromagnetic wave due to PD phenomena was detected using a log-periodic antenna and recorded employing a spectrum analyzer. 80 experiments in total were performed for CHV, CLV, SF and IN. The original independent variables for each classification model, skewness and kurtosis of each period of the captured signals, were calculated. Then, 60% of the experimented data was used as a training data for the PD classification models. Another 40% experimented data was used to evaluate the performance of the designed PD classification models both LVQ and PCA-LVQ models. Besides, noise signals were generated with computer program for testing the efficiency of these models. 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 generated 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 LVQ model can predict PD patterns without noise signal with the accuracy 100% whereas the PCA-LVQ model provided more than 87.5% accuracy of PD classification. The prediction ability of LVQ PD classification models decreased sharply especially for CHV when this model was tested by the mixed noise-PD signals. Whereas the prediction accuracy of PCA-LVQ model was more robust to the noise signals than LVQ model.