A Joint Model-driven and Data-driven Distribution Network Line Fault Prediction Method under Typhoon

In recent years, extreme weather events such as typhoons have severely affected the safety and stability of the grid. In order to accurately assess line fault risks under typhoon, this paper proposes a joint model-driven and data-driven method integrating weather forecasts, geographic data and grid parameters. First, fault mechanisms are analyzed to identify key factors associated with faults. Then, plenty of pseudo-labeled samples are generated using a logistic regression to solve the lack of line fault data. Finally, a one-dimensional convolutional neural network and graph sampling aggregation algorithm (1D CNN-GraphSAGE) is trained to predict weather lines are faulty or not and some labeled samples are used to evaluate the performance of the method. It has been proven that the method can effectively capture spatial and topological features and accurately predict line faults through case studies, it can also adapt to topology changes with transferability and offer better interpretability than traditional data-driven models.