Research On Signal Detection And Pattern Recognition Of Partial Discharge In XLPE Cable

With the wide application of XLPE cable in power system, more and more attentions have been paid to its insulating state. Partial discharge detection can effectively achieve the status maintenance of XLPE cable, so it's the direction of development of equipment maintenance. This paper focuses on the partial discharge signal of XLPE cable, and researches on on-line partial discharge detection, anti-interference, pattern recognition and other aspects.In this paper, the design of wideband current sensor is completed by experiments, and signal conditioning circuit is used to achieve the amplification and filtering of sensor output signal. And then, use acquisition card and IPC to achieve the acquisition, display, analysis and preservation of partial discharge signal. In addition, this paper analysis the type of partial discharge of XLPE cable, and produced four simulation models to partial discharge test.Because partial discharge signal is weak and there is a lot of interference on-site, it is critical to extract partial discharge signals from the on-site collected data by the use of anti-interference technology. White noise and periodic interference are two main types of interferences in on-line partial discharge detection. CL multiwavelet denoising method is used to denoise. Multi- wavelet vector threshold method and the adjacent coefficient method are introduced to process the simulated PD signal superimposed random white noise. It verifies the effectiveness of multiwavelet denoising method and the superiority of the adjacent coefficient method. Then, the multiwavelet adjacent coefficient method is applied to study denoising of periodic interference. It is also be used to denoise the tested partial discharge signal.Besides, achieve the spectrum display and feature extraction of partial discharge test data after denoising.For the four defect models, this paper chooses the appropriate classification algorithm and uses support vector machines as pattern recognition classifier in order to recognize and judge them. By the analysis of the pattern recognition results, determine the optimal sequence of classification algorithm node selection in the system, and achieve better recognition result.