Topology reconstruction and fault location method of distribution network based on graph neural network

This article proposes a fault section localization method for distribution networks that combines one-dimensional convolutional neural networks (1D-CNN) and graph neural networks (GNN), aiming to improve localization precision and robustness in topology changing scenarios. The rapid and accurate localization of faults in the distribution network is crucial for ensuring the safe and stable operation of the system. Existing artificial intelligence (AI) based localization models often experience significant performance degradation when the topology of the power grid changes. To address the aforementioned challenges, this paper proposes a novel hybrid neural network architecture. Firstly, an attention based spatiotemporal graph convolutional network (ASTGCN) is adopted to extract fault features from telemetry data from multi-scale spatiotemporal dimensions, and combined with graph attention network (GAT) to effectively fuse multi-source telemetry information; Secondly, 1D-CNN is introduced to adjust the dimensionality and compress the features at the node level, achieving the mapping from node state to branch fault state; Finally, the final fault section localization result is output through a fully connected network (FCN). The results show that the proposed model can maintain high localization precision under different topological structures, significantly better than existing methods, and has good generalization ability and engineering application prospects.