Abstract: In order to obtain the accurate wear conditions of metro wheel tread with the running mileage, the gray metabolism model, quadratic exponential smoothing model and unary linear regression model are selected as the single prediction model, the minimum absolute value of combined prediction error at the sample points is taken as the optimal criterion, an optimal non-negative variable weight combination forecasting model is established for the wheel tread wear changing trend. The prediction performance of the variable weight combination model and that of each individual model are tested by practical cases, the results show that this model can effectively overcome the defects of single prediction models, the accuracy and stability of the variable weight combination prediction model are significantly better. It provides a feasible solution for the accurate prediction of wheel tread wear trend for metro trains in actual operation.