GIS Partial Discharge On-Line Monitoring System And Kernel Principle Component Analysis Pattern Recognition Research

The most common feature of Gas Insulated Switchgear (GIS) electircal fault is before the insulation was completely brokedown, a partial discharge (PD) occurres. Through the online monitor of GIS PD, we can find the interior defect of GIS in time and avoid serious accidents. So, it is significant to research on the GIS PD monitor and the way of its pattern recognition. It is significant not only for the safe and reliable running of GIS, for the diagnose of GIS insulation and the character of its defect style, and also for the overhaule and repair of GIS.This paper designed a T sharp GIS experiment model, and emulated the electromagnetic field inside and outside the model with FDTD mothod;we researched the effects of distance, angle,pulse width,pulse amplitude to partial discharge waveform and the distribution of GIS electromagnetic field inside and outside together with its influence to online monitor.Developed a power frequency trigger circuit and GIS PD UHF online monitor software,Constructed a GIS PD UHF online monitor system.This system accomplished the following functions: signal acquisition and processing,spectrum analyzing and complex wavelet analyzing, pattern recognition of defects type,discharge locating,monitoring data query and and primary evaluation of insulation condition.Developed a composition data management system base on file management system and database management system, accomplished the function of real-time and quick storage,query and management, solved the problem of large quantity data managing.This system gain satisfied effective in laboratory and on-site work.Base on bi-spectrum analysis,implemented locating algorithm of two PD signal using time differencing mothod and developed a PD locating system using LabVIEW.This system acquired two PD signal quickly and synchronizly,and then locate using time differencing mothod automaticly.This system gain satisfied effective in laboratory.Aquired four PD samples produced by GIS model defect, constructed GIS PD surface and gray image. The method called PD pattern recognition basing on the KPCA is proposed.KPCA is used to abstract PD gray image, While maintaining the original model for the structure of distribution at the same time from the non-linear features of the principal component, And then use minimum distance Classifier for classification. Laboratory studies showed that the recognition rate is about 80 percent, be a good effect.