In this paper, a neural network medium-long term hydrological forecasting model coupling LM algorithm with self-adaptive algorithm is established in combining statistical analysis with fuzzy analysis, choosing predictors such as rainfall and atmospheric circulation in previous stage that affect the monthly discharge at the Yichang Station of the Yangtze River, comparing the advantage and disadvantage of several modified BP algorithms, discussing several problems in the modeling process. The results of calculation show that the model is highly effective.