Separation Technique of Multiple Source PD Signals by Employing Feature Optimization Extraction of Cumulative Energy Function

Multiple partial discharge (PD) resulted from different defects may occur simultaneously in power equipments, and cause the interpretation problems such as pattern recognition and risk assessment of mixed PD sources. In this paper, a mixed partial discharge (PD) signals separation algorithm based on feature optimization extraction of cumulative energy (CE) function was proposed in order to diagnose the insulation condition of equipments. The CE functions in time domain (TCE) and frequency domain (FCE) were calculated to characterize the pulse current and ultra-high frequency (UHF) waveforms and their FFT spectrums, respectively. The mathematical morphology gradient (MMG) operation was applied to the TCE and FCE, and the energy rising steepness features were extracted. The standard deviation of extracted features was adopted to evaluate their separation performance. The length of structure element in MMG was optimized with the goal of maximum separation performance. The energy rising steepness features with optimized S_EL was adopted in the separation method. The experiments with three kinds of typical multi-defect models were performed on a 252kV GIS in laboratory, and the separation performance of the proposed algorithm was verified with the acquired mixed UHF signals. Moreover, the algorithm was successfully applied to the separation of multiple PD UHF signals detected from an on-site 1100kV GIS. The separation results indicate that the proposed method is effective for both internal and external sensors, which shows good performance of multiple PD UHF signals separation.