Chinese Electric Power Equipment Fault Recognition Based on Graph Convolutional Networks

Electric power equipment fault recognition is an important problem in power system. Traditional methods focus on the various indicators produced by power equipment automatically while the textual information artificially generated is rarely paid attention to. In this paper, we transform this problem into text-based power fault recognition task. This task usually faces two difficulties: lack of data and complex grammar in the electric power documents. To overcome the difficulties, proposed a method based on graph convolutional network for the task, i.e., C-TextGCN. Our C-TextGCN is built based on point-wise mutual information and term frequency between words and documents. To evaluate our C-TextGCN, we collected a new dataset, i.e., CPTF. The experimental results on the CPTF dataset show that our method has significant advantages and outperforms state-of-the-art methods.