Blind Sources Separation For Mixing Signals Excited By Partial Discharge In Gis

Partial discharge(PD) in Gas Insulated Switchgear(GIS) is one of the important factors for arising the fault. The complex structure of GIS and the randomness of partial discharge lead to the existence of multiple insulation defects in GIS. Extracting the single partial discharge information from the signals of multiple insulation defects partial discharge is significant for safe operation of GIS. And the separation of partial discharge mixing signals is the main puzzle.This paper designed a T-shape GIS model for multiple insulation defects partial discharge experiment based on the analysis the shortage of the existing GIS test model. And this GIS model can simulate L-shape and line-shape GIS. This paper also designed three kinds of defect models and built the UHF detection system of partial discharge. Borrowing the ideas of operation methods of"cocktail lounge", introduced blind sources separation(BSS) into the research of separation for the mixing signals of multiple insulation defects partial discharge. This paper chose the algorithm of second order non-stationary source(SONS) to fit the non-stationary UHF pulse signals of partial discharge. To restrain the interferes of white noise, pre whitened the mixing signals gained from the sensors. This procedure makes the algorithm of SONS more robust. Performance Index(PI) and Signal Interference Ratio(SIR) were used to value the results of separation.The suitable physical size and the installation site of the insulation defects were determined to obtain the partial discharge UHF signals, the amplitudes of witch are similar. To make the research easy, two styles of multiple insulation defects discharge mixing signals were created via the linear instant mixing matrix H by single partial discharge signals. The algorithm of SONS was used to separate the two kind of mixing signals. By adjusting the parameters of separation, the optimum domains for the parameters of separation were found. And these domains made a reference for the separation of random mixing signals.Monte Carlo method was made a use of statistic separation for mixing signals mixed by random linear instant mixing matrix H. The results show that the algorithm of SONS separate the mixing signals generated by the multiple insulation defects in GIS perfectly. Also the algorithm was tested by the real mixing signals. PI and SIR appear that the effectiveness of separation is acceptable. And the algorithm of SONS can be used to separate the mixing signals which the mix units were more than two. The results of this paper make a convenience for the identification of the multiple insulation defects.