Abstract: The rapid calibration of hydrological model parameters is one of the important research contents in flood forecasting and early warning of flash floods and riverine floods in middle and small-sized rivers. The automatic parameter optimization function can greatly improve the efficiency of hydrological models. The random search algorithm is the basis of most global parameter optimization algorithms, but it is seldom used because of time-consuming. Brooks' adaptive random search method as the object and. Net parallel object are used for CPU parallel transformation, and NVIDIA CUDA object is used for GPU+CPU parallel transformation. The Xin'anjiang Model on the Qipei River sub watershed in Myanmar is taken as the optimization object, and the optimization effect and calculation efficiency are compared. The research results show that the optimization results of ARS algorithm and SCE-UA are equivalent, and the computing efficiency of ARS algorithm after parallel transformation is significantly improved. The research results have important reference value for the comparison and selection of parameter optimization algorithms in the application of hydrological models.