Dissolved gas in oil is an important basis for transformer fault diagnosis. In order to fuse and expand the characteristic properties of dissolved gas content in transformer oil and improve the accuracy of transformer fault diagnosis, this paper proposes a support vector machine transformer fault diagnosis method based on Improved BP neural network. Firstly, the five-dimensional gas feature attributes are fused and extended to 128 dimensions by Improving BP neural network. Then, using the extracted feature vectors of each layer as the input of the support vector machine to diagnose the transformer fault and increase the Improved nerve The weight of the feature vector with higher diagnostic accuracy in the network; finally, the feature vector with the largest cumulative weight is selected as the input, and the support vector machine is used for fault diagnosis of the transformer. The method is mapped by multi-layer neural network to make the extracted gas feature information merge and expand to have more obvious feature differences, which can effectively improve the diagnostic accuracy of the support vector machine. The simulation results show that the proposed algorithm has a higher diagnostic accuracy than the neural network and support vector machine transformer fault diagnosis method. At the same time, with the increase of training data samples, the diagnostic accuracy of the model has a greater improvement.