Online recognition method of partial discharge pattern for transformer bushings based on small sample ultra-micro-CNN network

Abstract

Oil impregnated paper bushing is the key equipment connecting a transformer and a power grid. Insulation deterioration may cause partial discharge, which poses a great threat to the safe operation of power systems. In order to realize the online diagnosis for partial discharge and automatic identification of insulation defects of transformer bushings, an ultra-micro-convolutional neural network with only more than 3000 parameters is designed, which adaptively extracts partial discharge characteristics based on small samples, so as to judge the defect category and the reasons. The accuracy rate can reach 97.1%, the computational complexity is lower, the real-time performance is stronger, and it can be easily deployed on various embedded platforms.