Spatiotemporal correlation graph neural network fault location method for large-scale distribution network

With the continuous expansion of distribution network scale, the existing distribution network fault localization method can only realize regional judgment and cannot effectively locate faulty line segments when there is no additional equipment installed for fault change observation. To solve this problem, A spatiotemporal correlation graph neural network (scGNN) is proposed for fault location in distribution network. First, considering the influence of fault type and distributed power supply, a low-observability node selection method is proposed by fault imbalance current, which completes the global observation of the system fault information by selecting specific node data changes. Second, a graph convolution spatiotemporal attention module considering fault information back propagation and attenuation variations is proposed for differentiated feature extraction for different types of faults. Further, a spatiotemporal correlation graph neural network fault localization method is proposed for distribution network fault localization with limited observation information. Its monitoring of basic quantities such as overvoltage and current integrates system topology and data changes. Based on the above, the proposed method can significantly improve the localization accuracy of different types of faulty line segments without additional equipment installation. Finally, the feasibility and effectiveness of the proposed method are verified by the IEEE 123 node distribution grid and the Portugal rural grid.