A Lightweight Graph-based Convolutional Neural Network for Fault Correlation Analysis in Power Systems

With the continuous expansion of the scale of the power grid, the number and types of equipment in the power grid is increasing, the power grid as a large complex system, the probability of failure and the severity of the failure is higher than the simple system and determine the system. Transformer is the key equipment in the power grid, and its failure often causes serious consequences. In order to solve the problem of poor effect of traditional data mining methods in concurrent fault identification, this paper proposes a concurrent transformer fault mining method based on knowledge graph. The method adopts a combination of top-down and bottom-up approaches to construct a knowledge graph of transformer faults, and then uses an improved lightweight graph convolutional neural network to mine the concurrent faults in the knowledge graph. The experimental results show that the proposed method can effectively identify concurrent faults in the knowledge graph of transformer faults, and the average precision and recall rates are better than those of the traditional graph convolutional network, which verifies the effectiveness of the method.