Based On The Improved Artificial Neural Network Evaluation And Forecasting Of Hydrological Elements

The development, utilization and management of water resources, such as the water resource system's optimization, decision, evaluation, prediction and other aspects, are the rational activity of human exerting on natural water system. Artificial neural network is a complex network system, which is made of a number of simple elements and is used to simulate brain behaviors, to recognize fuzzy information and to deal with complex nonlinear relationship. Artificial neural network has the ability to do parallel process, distributed store and information reasoning, and it also shows some obvious characteristics in settling fault tolerance, nonlinear, local limitations, and convexity, etc. In recent years, the application potentialities of neural network technology has been revealed in many scientific fields, especially in hydrology and water resources community, which reflects the important theoretical value and broad application prospect.The paper studies the basic theory of artificial neural network and its application in hydrology factors assessment and prediction based on the synthetic analysis about the domestic and international application of artificial neural network in water resource system analysis and hydrological forecast. Combining water domain knowledge with project cases, the application of improved artificial neural network in groundwater quality evaluation, flood and runoff forecasting is discussed. The sample training, forecast scheme verification and error analysis are carried on.Based on dynamic and randomized analysis of the regional groundwater, groundwater quality system evaluation model is established and used to assess groundwater quality by artificial neural network. Results show that the application of improved BP neural network in groundwater quality synthetic evaluation is feasible, and the network model, once trained, can be applied widely in environment problems evaluation.According to the importance and complexity of runoff forecast, the steepest descent-conjugate gradient method of fuzzy pattern recognition neural network is described. By learning recognition runoff conditions and runoff mechanism, the modeling principle and method of rainfall runoff forecast based on artificial neural network are discussed. Also, the sample training, forecast scheme verification and error analysis are carried on. Result shows that the fuzzy pattern recognition neural network runoff forecast model is feasible.According to the complexity of natural river system, the river flood forecast could be divided into single channel and river system flood forecast. The neural network flood forecast model is constructed. The improved BP algorithm of flood forecasting peak recognition theory is used to establish the river flood evaluation scheme. The prominent characteristics of flood forecast by artificial neural network technology are that the simulation of flood evolution has relatively high precision, easy implement, speedy and flexible operation.