Study Of Denoising And Pattern Recognition In Online PD Detection Of10kV Cables

With the wide application of cables in city power grid, the insulating status of cables hasbeen given more and more attention.The PD online detection of cables is meaningful toensure the reliable operation of power systems. This paper carries out the related technologyresearch based on current researches of PD online detection at home and abroad.At the present stage, emphasis and technical problems in online detection are mainlyconcentrated in two aspects, which are denosing and pattern recognition. In this paper, adetecting equipment which is called Longshot is bought and used to detected hundreds ofcable lines in Dongguan City. Based on experiments the characters of typical PD and noiseare summed up. In view of the two common noise (the white noise and narrow-bandinterference), a method based on wavelet packet is proposed to restrain the interferes of whitenoise. Another method of suppressing narrow-band interference using partial energy ratiopretreatment is also proposed in this paper. In this method the different characteristics of fastFFT power spectrum between simulation PD signals and PD signals mixed with thenarrow-band interference were analysed. The notion of partial peaks is brought in and partialenergy ratio is defined. The denoising simulation shows that the proposed method could picknarrow-band interference signal from real PD signal. This modified method is used in waveletpacket power spectrum to denoise narrow-band interference, comparison with that of waveletpacket denoising shows that the FFT is superior in suppressing narrow-band interference.The detected online PD may come not only from the cable and its joints, termination, butalso from the switchgear connected with it. Different PD source is with different consequenceand has its own criteria. Therefore it is necessary to recognize different PDs. Based onwavelet packet, a method, combined energy spectrum with1.5-order norm entropy, areproposed here, and the recognition rate reaches97%. Besides, in view of the difficulty offeature selection, a new pattern recognition method based on adaptive wavelet neural networkwas proposed and an adaptive wavelet neural network with four-layer was given to recognizePD source in this paper. Particle swarm optimization algorithm was used to optimize thenetwork first, and then BP algorithm was used to make a second optimization, whoseperformance is remarkably better than that only using BP algorithm. Influences of different wavelets and different structures to the performance of the adaptive wavelet neural networkwere discussed. The results show that the adaptive wavelet neural network which is optimizedby both particle swarm optimization and BP algorithms is able to recognize PD sourceaccurately and reliably.Da Hongshan10kV cable line is taken as an example of online PD detection.Thedetecting results are analysed and proposals are offered.