To improve the maintenance capability of key traction power supply equipment, a traction transformer service health management system suitable for railways has been designed. The advantages of MATLAB in scientific calculation and C# in system interface development were complemented. The basic data ledger was established by using Mysql database and cloud server, and the state evaluation model of main transformer mechanism and experience was constructed by driving the historical test data of transformer in traction substation. The mechanism model adopts the analytic hierarchy process (AHP)-entropy weight method to determine the subjective and objective weights, chooses the cloud model to determine the membership degree of each index, and finally carries out fuzzy operation to get the appropriate comments. Principal component analysis-whale optimization algorithm-support vector machines (PCA-WOA-SVM) was selected for joint classification evaluation, and synthetic minority over-sampling technique (SMOTE) was to oversampling the dataset to balance the integrity of the sample. The results show that the recognition accuracy after dataset equalization is 96.67%, which is 5% higher than processing before. After verification, the hierarchical joint evaluation provides a new approach for conducting condition assessment and maintenance strategy formulation of traction transformers, reflecting the reliability and accuracy of the model and system.