Fuzzy Clustering Neural Network as Flood Forecasting Model

Abstract
Flood forecasting is always a challenge in Taiwan, which has a subtropical climate and high mountains. This paper develops a fuzzy clustering neural network (FCNN), and implements this novel structure and reasoning process for flood forecasting. The FCNN has a hybrid learning scheme; the unsupervised learning scheme employs fuzzy min-max clustering to extract information from the input data. The supervised learning scheme uses linear regression to determine the weights of FCNN. The network, which learns from examples, is a hydrological processes theory-free estimator. Most of the parameters, weights of the network, are adjusted automatically during the network training. Only one parameter needs to be justified during constructing the flood forecasting models. The one-hour-ahead floods of the Lanyoung River during tropical storms are forecasted by the constructed models. Our results show that the simple but reliable model is capable of real time flood forecasting.