In recent years, artificial intelligence methods predicted the hydrologic flood forecasting universally. This study employed a Back-Propagation Network (BPN) as the main structure in flood forecasting to learn the characteristic of catchment system. When the modeling, it could obtain the infinite set of parameters to satisfy the efficiency coefficient with stochastic initiate parameters. This paper attempted to combine Self-Organizing Map (SOM) to cluster representative weights and biases. A SOM network with classification ability was applied to classify the BPN parameter rules and to obtain the winning parameters. Finally, it will provide the point prediction, interval prediction and probability prediction. It can see that the artificial neural network has a high accuracy of the flood forecasting.