Flood forecasting using the radial basis function neural network with fuzzy min-max clustering

摘要

In this study, a radial basis function (RBFNN) is employed to develop a rainfall-runoff model for flood forecasting. In essence, the nonlinear relation of rainfall and runoff can be considered a linear combination of some nonlinear functions. RBFNNs employ a hybrid two-stage learning scheme, unsupervised and supervised learning. In the first stage, fuzzy min-max clustering is proposed for measuring the similarity of the input data. During the second stage, the weights from the hidden layer to output layer are determined by multivariate linear regression method. The modified RBFNN is a model-free estimator with only two parameters that must be determined. Recently, powerful earthquake-induced landslides blocked the Chingshui River, and a new reservoir was born. Flood forecasting is the top priority for establishing a warning system. Several rainfall and runoff events data collected during typhoons are used to construct the rainfall-runoff model. Our results show that the RBFNN can be applied successfully to build rainfall-runoff models and provide high accuracy of flood forecasting.