Fault diagnosis of power transformers based on dissolved gas analysis and multi-kernel graph convolution network integrated with dual-channel classifiers

A power transformer fault diagnosis method based on dissolved gas analysis and multi-kernel graph convolution network integrated with dual-channel classifiers (DM-DC) is proposed to address the problems of insufficient accuracy and large deviation in recognition when dealing with imbalanced data. Firstly, construct multi-dimensional supplementary feature vectors adopting multi-feature dissolved gas ratio analysis to enrich the characterization features of transformers. Secondly, extract and model sample features deeply adopting graph generation network and multi-kernel graph convolution network to further explore the relationship between representation features and fault samples. Finally, a dual-channel classification network composed of binary classifier and multi-class classifier is introduced to alleviate the training bias towards the majority class and improve the model’s ability to handle imbalanced data through sample level reweighting. The various experiments shows that the proposed DM-DC can effectively solve the problem of low accuracy of minority class samples, with superior overall diagnosis accuracy performance, and is suitable for power transformer fault identification.