In order to improve the accuracy of transformer fault diagnosis, a transformer fault diagnosis method based on an improved SSA optimized MDS-SVM is proposed. Firstly, a multi-dimensional scaling(MDS) method is used to extract features from 20 dimensional transformer fault feature data to reduce the sparsity and multicollinearity of high-dimensional data. Then, the paper introduces a salp swarm algorithm(SSA) and improves the algorithm by adding trust mechanism and mutation to improve the convergence speed and ability of the algorithm. By comparing with the original SSA, PSO, GWO, and beta-GWO, the improved SSA algorithm is tested to verify its superiority. Finally, the improved algorithm is used to reduce the dimension of the MDS and optimize the parameters of a support vector machine(SVM) to build a new fault diagnosis model. The fault diagnosis accuracy is analyzed and compared with that of the SVM fault diagnosis model optimized by common algorithms, the BP neural network(BPNN), the $ K $-nearest neighbor(KNN) and random forest(RF) fault diagnosis models. The results show that the accuracy of the MDS-SVM transformer fault diagnosis model based on the improved SSA is higher than that of other algorithm models, and the generalization ability is stronger.