Fault Location Method of Distribution Network Based on Improved Graph Neural Network

Distribution network fault location is the key technology to ensure the stable operation of power system, and it is very important for the reliability and security of the system. In this paper, a fault location method of distribution network based on improved graph neural network GraphSAGE-GAT is proposed, inspired by the advantage of graph neural network to capture topological features. In this method, the topology of distribution network was transformed into an adjacency matrix, and the electrical features of fault lines were superimposed into the model. The representation of graph nodes was learned through sampling and aggregating of neighbor node information by GraphSAGE, and the multi-head attention mechanism in GAT was introduced to improve the feature characterization, and then the fault location problem was transformed into a node classification problem. Finally, the method was verified in the IEEE09 node distribution network model. The experimental results showed that GraphSAGE-GAT model has better accuracy and timeliness of fault location compared with common graph neural network models.