Abstract: A modified GWO-SVM transformer fault diagnosis method based on Borderline-SMOTE-IHT mixed sampling is proposed to address the issue of low accuracy in transformer fault diagnosis caused by imbalanced transformer fault data. Firstly,the Borderline SMOTE algorithm is used to select the most representative boundary samples to generate new minority class samples,and the IHT algorithm is used to remove noise or edge samples from the majority class,increasing the differences in features between classes.Secondly,based on the idea of differential evolution,the dynamic convergence factor and probability mutation mechanism are introduced into the gray wolf algorithm to optimize the penalty factor and kernel parameters in the SVM model for improving the algorithm’s global search ability and convergence accuracy. Finally,the effectiveness of the proposed method is demonstrated through experimental comparative analysis.