Fault Prediction and Classification of Large-Scale Substation Equipment Based on Neural Network

In the power system, transformer equipment is one of the important components, responsible for the transmission to the substation of high voltage current down and distribution to the user. Because of its complexity and high-quality requirements, substation equipment failure is inevitable. However, the existing fault prediction methods are mostly based on traditional methods, which have shortcomings in accuracy and reliability. The application of machine learning technology to fault prediction of substation equipment can greatly improve its reliability and security. Based on this, the random walk graph sampling algorithm is firstly adopted in this paper to form multiple batches of sub-graphs in the way of random walk, so as to solve the training complexity problem of large-scale equipment structures and improve the generalization of model training. On this basis, the classification algorithm based on stochastic graph convolution sampling neural network (GT-GCNN) is proposed. By aggregating the features of higher-order nodes, the semantic correlation information between long-distance graph nodes is fused, the overall perception ability of graph structure database is improved by multi-batch subgraph training, and the effectiveness of GT-GCNN algorithm in Inductive Learning scenario is verified in five public databases.