Research On The Application Of Flood Forecast On Neural Network

Influenced by the sichuan basin and the north qinling mountains, there are manyrivers that flooding frequently in DaZhou. Every time, floods bring huge economic lossand heavy casualties to the government and people. In fact, if we don’t take effectivemeasures to prevent floods, the lives and property of the people will be faced withenormous threat, economic losses will be growing.Flood forecasting is a very important non-engineering flood control measures,directly impact on the flood control, rational utilization of water resources and waterconservancy project management etc. At present, all of the flood forecast model is setup based on the measured data, such as statistical model, the deterministic predictionmodel, can only approximate to simulate the actual flood rules. But this flood forecastmodel was limited by many conditions in practical application, it difficult to be used todealing with the internal relations between the flood and the elements. Based on theresearch results of domestic and overseas, this paper makes a study on flood forecastinguse artificial neural network application.It introduced the basic theory of artificial neural network and BP neural network indetail, putted forward attention about establish BP network, analyzed the defects of BPneural network, and introduced some modified algorithm. Used the BP neural networkmodel that improved by LM algorithm to predict the flood flow of donglin hydrometricstation on upstream of zhou river in DaZhou, sichuan province, analyzed the forecastresults, found out some shortcomings of BP neural network model.This thesis focused on how applies genetic algorithm (GA) to artificial neuralnetwork (ANN). Through study of GA algorithm, Aim at problems in BP networkduring the learning process, optimized the BP neural network and designed GA-BPneural network model, this model can avoid obtain the local optimal solution in thelearning process of BP neural network model. Using the hydrological data of zhou riveras research samples, the GA-BP flood forecasting model was realized byMATLAB(2009a) neural network toolbox as development environment, obtained theforecast results through repeated training, compared with BP neural network model that only use LM algorithm.The results show that the prediction result of GA-BP neural network model isreasonable, small relative error, fast convergence rate and high prediction accuracy. Themodel has preferable practicability, can get better application in flood forecasting field,also can provide some reference opinions to flood control work in DaZhou.