Pulses separation and recognition strategy for multiple partial discharge sources of oil-paper insulation based on time-frequency similarity

Multiple partial discharge (PD) sources are often generated within power transformers due to the complexity of liquid-solid insulation system and operation condition, which would give rise to crossover and overlap of the registered PRPD patterns and incorrect diagnosis. To solve the problem of multiple PD sources recognition, a new method based on time-frequency analysis (TFA) of PD pulses and affinity propagation clustering (APC) is proposed for pulses separation and recognition of multiple PD sources of oil-paper insulation in transformers. The multiple PD pulses are firstly separated by input the S transform (ST) based time-frequency similarity matrix into affinity propagation clustering (APC) algorithm. Then, a support vector machine with particle swarm optimization (PSO-SVM) classifier based on PRPD statistical features is employed to obtain the recognition results of PRPD sub-patterns relevant to each PD source, and thereby examine the separation effectiveness. The PD data of artificial defect models acquired in laboratory are adopted for algorithms testing. It is shown that ST combined with APC can effectively eliminate pulse-shaped noises (PSN) and separate pulses of multiple PD sources.