To enhance the learning ability of a transformer fault diagnosis model for unbalanced samples and improve the recognition accuracy of minority fault samples, a transformer fault diagnosis technology based on sample expansion and feature optimization and support vector machine (SVM) optimized by improved grey wolf optimizer (GWO) with multi-strategy (IGWO) is proposed. First, the mixed oversampling technique based on K-nearest neighbor oversampling approach and kernel based adaptive synthetic algorithm is used to expand the minority samples to obtain the balanced datasets,, and analysis of variance (ANOVA) is used to select the transformer candidate ratio features. Then, by improving the initialization strategy and update formulas of parameters and positions of the GWO and introducing a differential evolution strategy to adjust populations, an improved GWO with multi-strategy is proposed. Finally, a transformer fault diagnosis model based on mixed oversampling technology and SVM optimized by IGWO is constructed, and experimental results show the method can enhance the recognition accuracy of the model for minority fault samples and improve the overall classification performance of the model effectively.