FIR Neural Network And Its Applications In Flood Forecasting

Artificial neural network (ANN) has very strong learning ability, real-time potential, good robust and nonlinear mapping ability, etc, is now being more and more applied to problems of many fields. Much success has been acquired in hydrological forecasting. Firstly, this paper discusses the study of ANN and its application, especially, BP and FIR neural network. Then, the development and the major research interests of the generalization ability of ANN, hydrological forecasting in ANN and flood routing are analyzed. Based on these, new methods are raised and research results are listed as follows:1) Considering the generalization ability of ANN, learning time and convergence depend on training samples, samples are selected by clustering analysis in this article. Thus, amounts of sample are reduced and they are improved. Samples data, just like flood discharge, are smoothed by inserting in dynamic ANN (such as FIR neural network) models building so that better generalization ability are gained. The contrast experiment shows the measures are feasible and have good results.2) ANN applied to hydrological forecasting, are mostly BP neural network, static, and conflicting with hydrological series which is dynamic system. In order to avoid effectively flood disaster, higher accuracy flood forecasting is importance. Dynamic FIR neural network is adopted to flood forecasting and has better performance than BP neural network and extends extension onto it.3) Up to now, flood routing is mostly modeled in hydrodynamic and hydrology models which unfeasible and need much information. FIR neural network can overcome the defects and identify availably flood routing rule. The results is just that. Moreover, a new method in data pretreatment is given and good outcome is obtained.Finally, a summary is given and some problems to be farther studied are discussed.