Abstract: Support vector machine (SVM) is one of the best theoretical algorithm to solve the problem of small sample classification. Kernel parameter selection is very important, which directly affects the accuracy of fault diagnosis. In this paper, the particle swarm optimization (PSO) was used to optimize the parameters of SVM, the PSO-SVM model was proposed which was applied to fault diagnosis of track circuit. By comparing the MATLAB simulation results, it was concluded that the parameters obtained by PSO were better than the random parameters, and the fault diagnosis accuracy of the established PSO-SVM model was higher than that of the ordinary SVM model.