Discrimination of three or more partial discharge sources by multi-step clustering of cumulative energy features

Partial discharge (PD)-based diagnosis is extensively employed in condition assessment of electrical equipment. In the case of multiple PD sources, discrimination of mixed signals is significant for reliable PD interpretation. To improve the separation performance of three or more PD sources, a multi-step discrimination method is proposed. The cumulative energy functions are exploited to characterise wave shapes of PD signals, and width and sharpness features are extracted and classified to separate mixed patterns. With the objective of maximising a novel evaluation parameter of separation capability, the oblique line and length of structure element are optimised in the feature extraction stage. For the multi-step discrimination method, the feature extraction and clustering procedures are repeatedly applied to the whole dataset, sub-classes, sub-sub-classes etc., until no more clusters are generated. To evaluate the separation performance of the proposed algorithm, a mathematical model for PD pulse is proposed, which is the multiplication of Heidler enveloping function and an oscillating function whose frequency spectrum confirms Gaussian distribution. In the end, the multi-step discrimination method is tested with PD current pulses and ultra-high-frequency signals of three artificial defects in transformer and gas-insulated system, and the results prove the effectiveness of the proposed algorithm.