In view of the heavy rainfall disaster, theinterpretation forecast model was established uses the fuzzy neural network (FNN) method based on the ECMWF in this study. At the same time, the information gain method is introduced in the processing of the prediction model input. The heavy rainfall forecasting factors obtained from the primary selection are calculated, and the forecasting factor is obtained according to the feature weighting condition.Here,the forecast information is condensed, and a few unrelated forecasting information variables with large information volume are extracted to optimize the network structure of the forecasting model. Furthermore, the modeled samples were reconstructed by setting thresholds, and the modular forecasting models of heavy rainfall and non-rainstorm were established. The actual forecast results of the experimental prediction of the 24h aging of the large-scale heavy rainfall in Guangxi from 2012 to 2016 showed that the new scheme has better forecast results and stable forecasting effect, which has certain universal applicability.Further comparison with the forecast results of the ECMWF shows that the new method after the interpretation of the ECMWF has more accurate forecasting ability than the numerical interpolation model. The forecasting techniques are all positive techniques. Thus the better interpretation effect can improve the forecasting ability of ECMWF forecasting heavy rainfall to a certain extent, and has better reference value for business forecasters.