Abstract: This paper takes reservoir real-time flood forecasting as the research object, aiming at the problem that the forecast lead time for flood prediction of small and medium-sized river basins is short and the natural forecast period of river basins is difficult to meet the production requirements of hydraulic project, studies the reservoir real-time flood forecasting method, and builds a reservoir real-time flood forecasting model based on meteorological and hydrological coupling. Firstly, a 24-hour precipitation forecast model is established by using short-time proximity extrapolation technique and STMAS-WAF mesoscale weather model. Then the results of meteorological forecast are coupled to hydrological forecast to achieve the spatio-temporal scale matching of meteorological and hydrological. Finally, a precipitation forecast correction method based on the feedback of incoming water forecast results is established, and the precipitation forecast results in the future 24 h are revised on a rolling basis according to previous rules. The revised precipitation forecast is input into the hydrological model to predict the incoming water from the reservoir in real time, so that the hydrological model can be accurately input from the source, and the real-time flood forecast of the reservoir with high precision and long forecast time can be realized. The problem of insufficient forecast period of traditional hydrological model is solved. The Xin’anjiang Reservoir is taken as an example to carry out a case study of the model, and the coupled meteorological and hydrological forecast scheme of Xin’anjiang Reservoir is formulated to carry out real-time flood forecast of the reservoir, and the influence of different rainfall input forms on the forecast accuracy is discussed. The results show that: the model takes into account the bidirectional feedback of meteorology and hydrology, and improves the real-time flood forecast period of the Xin’anjiang Reservoir from about 6 hours to 18 hours, ensuring that the forecast accuracy within 18 hours is above 80%. Compared with two traditional methods that do not consider precipitation forecast and use unmodified precipitation forecast, the effective forecast period of the new method is increased by 200% and 29% respectively, effectively extending the real-time flood forecast period of reservoirs, improving the flood forecast service ability, and providing an effective method for real-time flood forecast of reservoirs.