Research On Separation Method Of GIS Multi-source Mixed Partial Discharge Signals

Partial discharge(PD)will occur when there are several insulation defects in GIS,and the online monitoring and diagnosis for PD have been the major method of judging the insulated state of GIS.The Ultra-high Frequency method is the most direct and effective method for the detection and diagnosis of PD,and has made many achievements in faults location and pattern recognition.However,the existing research mostly assumes that there is only a single insulation defect in GIS.The complex and independent layout of gas chambers in GIS makes it possible for multiple independent insulation defects to occur simultaneously,then the mixed PD signal is generated which brings great challenge to pattern recognition.In this paper,based on the analysis of the statistical and mixed characteristics of mixed partial discharge signals,two improved blind source separation algorithms are proposed to separate two kinds of mixed partial discharge signals,respectively.The research results of pattern recognition of single PD signals is still available for separating signals.The main work and contribution are as follows:The research results of GIS insulation defects,PD detection technology and the separation algorithm of mixed PD signals are summarized.In view of the characteristics of UHF PD signal,the research route of this paper is determined by comparing the existing research: the blind source separation algorithm is employed to separate the mixed PD signals into independent and single PD signals,which solves the problem of difficult identification of mixed PD signals.Two types of mathematical models of GIS PD signals and their time-frequency domain statistical characteristics are analyzed.The mixed model and the corresponding separation theory of GIS mixed PD signal are given.A mixed partial discharge signal classifier based on BP neural network is designed to realize the recognition and classification of the two kinds of mixed partial discharge signals.Finally,the theory of blind source separation is introduced and the feasibility of solving the problem of mixed PD signal separation is discussed.For the mixed PD signals composed of no more than 3 typical GIS PD signals,An improved blind source separation algorithm based on kurtosis maximization is proposed for mixed GIS PD signals with no more than 3 source signals,and artificial bee colony algorithm is applied as the optimization function.Givens rotation algorithm is introduced to optimize the separation matrix,which avoids the defect that the traditional blind source separation algorithm based on maximum kurtosis can only extract one source signal at a time,realize the separation of all source signals at the same time and effectively reduce the calculation amount.Aiming at the uncertainty of the amplitude of separation signal in blind source separation,the amplitude correction link is designed.Simulation results show that the proposed algorithm has good separation performance and robustness for mixed partial discharge signals with time delay greater than 3 ns.For mixed PD signals with no more than 2 sources signals and composed of exponential PD signals,a blind source separation algorithm based on improved JADE algorithm is proposed.An artificial bee colony algorithm is introduced to improve the joint approximation diagonalization algorithm of the cumulative amount matrix.Finally,the effects of source signal type,time delay and noise on the performance of the proposed algorithm are analyzed by simulation.The simulation results show that the proposed algorithm has good separation effect and certain robustness for mixed partial discharge signals with time delay greater than 4.5 ns.