Fault Location Method of Complex Urban Distribution Network Based on MLP-GCN

To overcome the shortcomings of some traditional fault location methods which can't fully consider the topology of complex urban distribution networks, a novel MLP-GCN method which integrates Multi-Layer Perceptron (MLP) and Graph Convolutional Network (GCN) is proposed to locate the fault. By considering the overcurrent, overvoltage, circuit breaker status, and other information of each node and line in the distribution network, the physical properties of the feeder and load nodes are extracted as fault data feature vectors, and then the node and edge features of the feeder are converted by a multi-layer perceptron, and then input to the graph convolutional network to form a mapping between the fault data vector and the fault branch, and forms the fault location model to predict the fault point. Finally, the model is simulated in a S-SCDN (supply region of substation centralized distribution network) complex feeder cluster, and its effectiveness is verified.