Low computational efficiency and different parameters with the same effect (local optimization) are the main problems in the study of parameter optimization of distributed hydrological models. Based on RAGA (Accelerated Genetic Algorithm Based on Real Number Coding), this paper proposes a distributed hydrological model parameter classification optimization method, i.e., the parameters needed to be calibrated are divided into several categories according to their physical meanings, and optimized category by category. This method can reduce the dimensions of the parameters to be optimized. On the one hand, it can increase the speed of optimization calculation, on the other hand, it can approach the global optimum to a certain extent, reducing the problem of different parameters with the same effect. This paper adopts the distributed hydrological model WEP-L (Water and Energy Processes in Large Scale Basin) model to simulate and calibrate the monthly discharge process of the area above the Maqu hydrological station in the Yellow River Basin from 1997 to 2000, and verify the series from 2006 to 2016. Comparing the parameter optimization methods without classification, it is found that the speed of WEP-L model parameter optimization has increased by about 37%, and the Nash efficiency coefficient (NSE) has increased from 0.739 to 0.829 after using the parameter classification optimization method. It shows that the parameter classification optimization method can not only save time, but also ensure the globality of the optimization algorithm and improve the accuracy of simulation