Aiming at the problems of high failure rate and low maintenance quality of railway turnout, this paper took the power curve of S700K switch machine as the research object, and a Compensation Distance Evaluation Technique (CDET) combined with an Modified Particle Swarm Optimization (MPSO) optimized Support Vector Machine (SVM) intelligent turnout fault diagnosis method is proposed. Firstly, the action mechanism of S700K switch machine is analyzed, and the power curve is divided into five stages: start, unlock, transfer, lock and structured representation, then the corresponding feature sets of switch power curve in each stage are extracted respectively. Secondly, the CDET is used to reduce the dimension of the extract feature candidate and select the sensitive feature. Finally, the disturbance and momentum terms are introduced to modify the particle swarm optimization algorithm, which is used to optimize the SVM parameters. Then SVM as a classifier to predict the turnout faults, and is compared with PSO-SVM, SVM and other algorithms. The simulations show that the method diagnosis accuracy is more than 97%, which can accurately and effectively identify the type of turnout fault.