Power Transformer Fault Diagnosis Method based on Spatio-Temporal Multiscale Graph Convolutional Neural Network
Chengzhen Li, Xuebin Lv, Haitao Dai, Chenhao Li, Dong Qiang Gao, Jiayuan Guo, Cunrui Xu, Faye Zhang · 2025
Aiming at the problems of insufficient fault feature extraction and insufficient spatio-temporal correlation mining among multiple sensors in traditional transformer fault diagnosis methods under complex environmental conditions, a power transformer fault diagnosis method based on spatio-temporal multiscale graph convolutional neural network (ST-MSGCN) was proposed. Firstly, a spatio-temporal graph containing the node topological relationship is constructed based on the correlation between the sensor node data to effectively fuse the feature information of multiple sensor nodes. Secondly, a spatio-temporal multiscale graph convolution neural network is designed to mine multiscale fault features in the spatio-temporal graph from two different dimensions, time and space, to achieve accurate identification of fault types. Finally, relevant experiments were carried out on the transformer fault dataset, and the average diagnostic accuracy reached $97.42 \%$, which proved the effectiveness of the proposed method and provided a new idea of spatio-temporal feature fusion for intelligent diagnosis of power equipment.