Partial Discharge Pattern Recognition in GIS Based on CNN-SVM

Abstract

The phase resolved partial discharge (PRPD) is commonly employed in the partial discharge pattern detection of gas insulated metal encased switchgear (GIS). Convolutional neural networks may automatically extract features, however because of the vastness of the data set, the model training effect is subpar. The traditional feature extraction approach relies on expert knowledge and experience. Therefore, a GIS partial dis-charge pattern detection approach based on CNN-SVM is proposed in this study. By including a multi classification support vector machine, this technique enhances the structure of the conventional convolutional neural network (CNN) model. The characteristics retrieved by the CNN are then input into the multi classification support vector machine for classification. The findings demonstrate that this technique may significantly increase the accuracy of GIS partial discharge pattern recognition, with the hybrid resnet18-SVM model developed by Alexnet-SVM, Googlenet-SVM, and Resnet18-SVM having the highest recognition accuracy at 99.3% .