Research on UHF PD detection method based on improved DBN

Abstract

With the development of power grid technology and the widespread application of gas insulator switchgear (GIS) equipment, the power supply reliability of the power system has been greatly improved, but the problem of partial discharge (PD) faults in GIS has always been prominent, seriously affecting the safe and stable operation of the power grid. How to quickly determine the type and cause of GIS discharge is the key to online PD detection. In this paper, in order to deal with the very complicated data processing of ultra-high frequency (UHF) PD, the time-consuming and low efficiency of manually judging the type of PD, a classification model of UHF PD system based on deep confidence network (DBN) is established and an automatic classification method for UHF PD based on improved DBN is proposed; the activation function Sigmoid is improved to effectively prevent the occurrence of the gradient disappearance problem; the optimized DBN parameters are used to train and classify data of different PD types. The classification accuracy rate of the test results reached 96.7%, realizing the rapid classification evaluation of UHF PD types.