Pattern Recognition of Partial Discharge in GIS Based on Improved BP Network

Summary

As the rapid development of electrical industry, higher standards have been set for the reliability of electrical equipment. Gas insulated substation (GIS) is widely used nowadays because of its small volume and high operational reliability, and online monitoring of partial discharge (PD) in GIS has become the focus of current research. Pattern recognition of the PD signals acquired is proved to be effective to grasp the different characteristics of typical defects in GIS. In this paper, three defect models are tested by acoustic method including the metallic protrusion, the floating potential and inner defect model, the PD data of which acquired after repeated experiments were analysed. Four spectrograms were calculated and displayed including the pulse count distribution Hn(f), the maximum pulse height distribution Hqmax(f), the mean pulse height distribution Hqn(f) and the discharge amplitude distribution H(q). The graphic differences of three defects are preliminarily summarized. 32 PD parameters were calculated as the possible input of improved back propagation (BP) neutral network, including discharge asymmetry, phase asymmetry, the cross-correlation factor, and the number of peaking, the phase median, skewness and kurtosis. Combined with three parameter-selecting schemes, which are respectively proposed based on the repeatability, the statistical features of the PD data and the distance criterion, the results of pattern recognition were compared. It is found that the scheme based on the class distance criterion is comprehensively optimal.

Additional informations

Authors

Temma, Morishima, Shimomura, MASUDA, MACHIDA, SHINOZAKI