Study On Pattern Recognition Of Partial Discharge Using Combined Complex Wavelet Coefficient As Feature

Insulation breakdown is primary reason for GIS failures. Partial discharge(PD) phenomena is most common characteristic before insulation breakdown in GIS. By PD on-line monitoring systems, GIS internal defects can be discovered in time to avoid the occurrence of fatal failure. So the research of PD detection and its pattern recognition for GIS is very important to guarantee its safe operation, find out its insulation condition and defects type in time and guide its maintenance.A new method suitable for UHF PD pattern recognition is proposed based on analyzing researches about PD pattern recognition home and abroad. Complex wavelet is used for decomposing disposed UHF PD signal; and then combined complex wavelet coefficient is constructed by combining the real part and imaginary part of complex wavelet coefficients, finally, features can be extracted from combined complex wavelet coefficients of each scale with ameliorated FCM method. Optimal combined complex wavelet coefficients and complex wavelet are selected by comparing J criterion; Physical model of four defects in simulation of internal defects are developed, according to PD signal feature in GIS, and the characteristic of UHF PD signal is summarized; RBF network is used for recognizing UHF PD. The main achievements are as follows:Differences of UHF PD signal appear among different type of defect, and the signal's shape of the same type of defect don't change under different experiment condition. The subset R|I|n which is more abundant than real wavelet transform is constructed by researching complex wavelet coefficients, and features of UHF PD signal are extracted from combined complex wavelet coefficient with FCM method; It proves that combined complex wavelet coefficient containing more time-domain and frequency-domain of PD signal.It proves that feature extracting from combined complex wavelet coefficient is more excellent than from real part of imaginary part of complex wavelet coefficient; feature extracting from optimal combined complex wavelet coefficient is more excellent than from other combined complex wavelet coefficient; feature extracting from optimal combined complex wavelet coefficient of optimal complex wavelet transform have the best clustering ability. At the same time, It proves that using J criterion to select optimal combined complex wavelet coefficients and complex wavelet is feasible.