Fault Location Method for Active Distribution Network Based on Graph Kernels Attention Network

The effectiveness of the artificial intelligence based fault location technology for power grids relies highly on training data. Once the topology structure of the distribution network changes, the accuracy of the fault location model will significantly decrease. To address this issue, in the paper we propose a distribution network fault location method based on graph kernels attention network. The vertices and edges of the graph in graph attention network correspond to the electrical nodes and lines of the distribution network. The attention coefficient is calculated based on the similarity of fault features between adjacent vertices, and the graph kernels attention network is constructed based on the connection relationship between nodes and their neighboring nodes. The status of each node is calculated to determine the fault location. The proposed method integrates the correlation between vertex features into the fault location model in a better way, thereby improving adaptability of the fault location model to the changes in distribution network topology. Finally, an IEEE33 node distribution network system is built for verification, and simulation results demonstrate that the fault location model proposed in this paper has the advantages of high positioning accuracy and good robustness. Moreover, in the event of changes in the topology structure of the distribution network, the model still maintains a high level of accuracy in fault location .