Using K-Means Clustering and Parameter Weighting for Partial-Discharge Noise Suppression

The analysis of partial discharge (PD) contributes significantly to the evaluation of the status of insulation in terms of power equipment. In recent years, wavelet transform and equivalent bandwidth and time have been the most common various developing methods associated with noise suppression. This study introduces the use of membership weight function and k-means clustering to classify measured data, in order to assist engineers in making accurate statistical judgments. PD signals can be detected through the analysis of shape features (pulse equivalent bandwidth, rise time, pulse full width at half maximum, discharge magnitude, and pulse polarity). We applied this method to field and laboratory experiments in order to investigate the effectiveness of pulse identification and classification. We found that the proposed method could properly identify noise and a discharge pulse.