In response to the issues of parameter calibration difficulties, susceptibility to local optima, and poor applicability in traditional hydrological models, a Bayesian optimization algorithm-based long short-term memory (BOLSTM) model was constructed based on machine learning principles, and it was applied to runoff simulation in the Han River Basin based on the simulation results, the SHAP method was used to attribute the influencing factors in the rainfallrunoff process, and the timelag analysis method was used to quantify the impact of the Middle Route of the SouthtoNorth Water Diversion Project on the runoff process of the watershed. The results show that the runoff simulation effect of the BOLSTM model is better; the construction of the Middle Route of the SouthtoNorth Water Diversion Project has delayed the impact of rainfall on the flow at the outlet of the Han River Basin, with the flow at the outlet of the basin increasing after 6 days of a rainfall; while before the construction of the project, the occurrence of a rainfall would lead to changes in the flow at outlet of the basin after 5 days, and the impact of rainfall in the Han River Basin on the flow at the outlet of the basin was greater before that of the project.