Multiple PD sources separation of oil-paper insulation using S transform based similarity

A new algorithm for multiple partial discharge sources separation is proposed in this paper. ST is firstly employed to obtain the time-frequency representation of original PD pulses. Then, PD pulses similarity matrix is computed based on ST amplitude matrix. By using the pulses similarity matrix, affinity propagation clustering is applied to perform the automatic separation of PD pulses. The proposed algorithm is tested by multiple PD sources generated by manually combining pulses from different single sources. It is shown that the successful rates of pulses separation 99.63% and 98.35% are obtained in two application cases. Results demonstrate that it is feasible to apply ST and APC to realize multiple PD source separation.