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

Ensuring the secure and stable operation of distribution networks hinges on precise fault location. However, in the application of data-driven fault location techniques, the number of fault categories often significantly pales in comparison to normal categories. This disparity in sample distribution poses challenges to existing data-driven methods. To tackle this issue, our paper introduces an improved approach for fault location based on a graph convolutional neural network (GCN). Voltage and current signals from the distribution network are collected, and the GCNSMOTE model is employed to address the class imbalance problem, ultimately leading to accurate classification. The scope of our research encompasses the IEEE33 and IEEE123 test systems. Experimental results affirm that our proposed model demonstrates heightened accuracy and the ability to accurately classify data even in the presence of substantial noise. These outcomes validate the effectiveness and robustness of our proposed method.