Construction of the FFWS Using Supervised and Unsupervised Performance in the Small Catchment

The radial basis function neural networks model (RBFNNM), a kind of hybrid neural networks model, is developed and applied to construct the flood forecasting and warning system (FFWS) in small catchment, South Korea. The RBFNNM consists of supervised and unsupervised training performance in the networks system. The parameters of the RBFNNM such as centers, widths, connection weights, and biases are determined by the RBFNNM training performance. The RBFNNM validation performance is also carried out using these parameters. The results of the RBFNNM training and validation performance are compared with those of the two multilayer perceptron neural networks model (MLPNNM). The two MLPNNM are embedded with the one step secant backpropagation (OSSBP) and resilient backpropagation (RBP) algorithm respectively. The RBFNNM shows better results than the MLPNNM-OSSBP and MLPNNM-RBP slightly on the basis of the statistical analysis and flood hydrograph respectively. The constructed FFWS using the RBFNNM spent less time for the training performance and can be easily used by the hydrologists with little background knowledge of the RBFNNM. This system, furthermore, can enable many local officers and engineers to protect large flood disasters and save the private life and wealth in the Wi-stream catchment, South Korea.