A reasonable maintenance schedule for railway traction substations is an important part of railway electric transports. It should take the reliability and the maintenance costs of the system into account. The fault tree analysis is applied to qualitative analysis and the GO method is used to quantitative analysis. Formula derivation of the railway substations’ reliability is got. Simulating the model through the survey data from the literature and improving the periodic maintenance means, establishing an optimized maintenance model of the railway traction substations based on reliability and the cost. Through two aspects of single equipment reliability or overall reliability, the model is solved with improved particle swarm algorithm. The result indicates that the model is objective and effective, that the model can reflect that the reliability of the system is higher, the maintenance costs are higher. And the minimum reliability of the system can be selected according to different maintenance cost.