Abstract: Based on the feedforward neural network(FNN) and the method of optimal approximation, the forecasting system for rainfall in Deyang City is trained with refinement intelligent grid forecasting, real-time rainfall monitoring data, ECMWF numerical forecasting data and CMPA Precipitation data. The forecasting results are verified and assessed with CMPA Precipitation data. The results show that the improved system can not only effectively intelligently optimize the precipitation of input layer factors and the forecast of its falling area, but also make the forecast results of the distribution of heavy precipitation centers and extreme precipitation in the process of rainstorm weather more close to the reality. Therefore, it can provide useful reference for local precipitation forecast.