Abstract: The fluctuation of river water level is a complex nonlinear process. Traditional neural network prediction is of slow convergence and poor stability with large error. To effectively predict river water level, a prediction model based on Elman neural network optimized by genetic algorithm (GA) is proposed. The effective combination of GA and Elman network solves the deficiencies of Elman neural network. The water level at Yongding river monitoring station is predicted by the proposed model and validated according to measured hydrological data, and the prediction results are compared with those obtained by Elman neural network and BP neural network. Results imply that the GA-Elman water level prediction model is of fast convergence and high precision. According to the prediction results, reservoirs and river barrages can be operated rationally for an effective allocation of water resources to meet the demands of irrigation, power generation and flood control.

Key words: river water level, prediction model, GA algorithm, Elman network, BP network, effective allocation of river water resources