Research on fault location model of low-voltage distribution system based on deep learning and graph neural network

In this study, a DG-MFAN (Dynamic Graph-Multimodal Fusion Attention Network) is proposed, which combines deep learning with graph neural network (GNN). Aiming at the complex characteristics of low-voltage distribution network, such as dynamic topology change and multi-source heterogeneous data, the framework captures the evolution of spatial characteristics brought about by topology change in real time through dynamic graph construction and adaptive learning. Using DynaGNN and lstm (Long Short-Term Memory) branches in the spatio-temporal feature extraction layer, the spatial features of dynamic topological structure and the temporal features of electrical data are extracted respectively; With the help of cross-modal Transformer encoder, the deep interaction between electrical time series and topological structure data at the feature level is realized, and the ability to identify complex fault features is improved. Finally, the multi-task learning framework is adopted to identify the fault section and accurately locate the fault point at the same time. The experimental results show that DG-MFAN is superior to many comparison models in the accuracy of section location (93.5%), relative position error (3.8%) and robustness score (8.6), which shows better fault location accuracy and dynamic adaptability, and provides a new and effective method for fault location of low-voltage distribution system.