关键词: 低压电流互感器, 在线检定, 卷积神经网络, 故障诊断

Abstract:

Low-voltage current transformer is a key equipment in the power system. In the on-line verification of low-voltage current transformer, it is very important to classify and identify fault samples. In this paper, an improved global average pooling 1DCNN-SVM fault diagnosis model is proposed for on-line verification of low voltage current transformer. The proposed method improves the structure of the traditional CNN model, introduces global average pooling to replace the fully connected network structure, and uses SVM to replace Softmax function in the test stage. Through experiments, the proposed method is compared with the traditional CNN. The experimental results show that the proposed method is better than the traditional CNN structure model in terms of training time, testing time and test accuracy of the model.

Key words: low-voltage current transformer, on-line verification, Convolutional Neural Network（CNN）, fault diagnosis