Localization Method And Pattern Recognition Of Substation Partial Discharge Based On UHF Method

Partial discharge detection can detect the early insulation defects of power equipment in time,provide targeted suggestions for equipment maintenance,and avoid power system failures due to further deterioration of insulation performance.In recent years,the partial discharge detection method based on UHF is widely used in partial discharge detection and location,However,the electromagnetic environment of substation is complex,which greatly interferes with the detection accuracy and positioning accuracy of UHF partial discharge positioning system.In this thesis,the principle of PD location based on UHF is discussed firstly,and the UHF PD positioning system is also introduced.Because of the interference of the noise signal,the positioning system has fluctuation phenomenon to the positioning result of the same discharge power supply,and to solve this problem,this paper proposes a jitter cancellation algorithm based on K-Means.At the same time,the problem of multi-source separation is studied,and the method of cluster analysis directly to the system positioning result is proposed;With the increase of the PD source distance,the detection of UHF weak signal is more difficult,and the positioning error increases gradually,which is nonlinear with the increase of the PD source distance.To solve this problem,this paper proposes a partial discharge location error correction method based on BP neural network.On the basis of admitting the inherent error of the positioning system,the distribution of error is studied and reproduced by using the ability of BP neural network's highly fault tolerance and the infinite approximation of nonlinear function,and then the error compensation curve is constructed to correct the measured value of the positioning system.Experimental results show that the proposed method can significantly improve the accuracy of UHF PDpositioning system.At the same time,in order to make the positioning system can locate the source of PD,but also can judge the type of discharge PD source,this paper presents a partial discharge pattern recognition method based on random matrix theory.The characteristic roots distribution of high dimensional random matrix constructed by partial discharge signal time series and its covariance matrix spectral distribution rule are used as the characteristic of PD pattern recognition.Compared with traditional pulse waveform or pulse peak value as characteristic quantity recognition method,This method is less affected in waveform time series and has has good practicability in engineering for the large amount of data acquisition(Big data characteristics)and different analysis mechanism.