Flood Forecasting Model Based On Copula Function And Neural Network

The frequency of floods has been at a high level caused by many external factors since a long time ago.Therefore,researchers around the world have taken flood forecasting as a research subject.This thesis studies the flood forecasting technology based on Copula function and neural network model.Firstly,we discuss the application of Copula function and neural network model in flood forecasting and then improve the two models by combining them with the popular methods of mixing multiple models.In the thesis,our method can be concluded as follows: 1.We first make use of the EM algorithm and the genetic algorithm to calculate the parameters of the mixed Copulas function model.2.Then we choose one of the parameter set with better fitting performance.3.After that,the mixed Copulas function model is built up according to the relationship between flood peak and the period of flood volume.4.We make use of the relationship between the model and the edge distribution of flood peak to estimate the period of flood volume.5.The result is compared with the prediction achieved by the single Copulas function model.The simulation results show the following facts that: 1.The genetic algorithm can better converge to the optimal value than the EM algorithm.2.The mixed Copulas is more accurate than the single Copula function.3.The mixed Copulas has the characteristics of flexibility and can fit the correlation of two variables in a better way.Secondly,in view of the complexity of water level time series and the lag of change,we introduce an extended wavelet neural network model.The input number of the extended wavelet network model is determined according to the correlation analysis and model simulation.The experiment results show that the extended wavelet neural network contains more accurate information than the typical wavelet neural network and BP neural network can improve the prediction accuracy of the neural network.However,as the number of entries with weaker correlations may result in lower prediction accuracy,the number of entries should be chosen appropriately.Finally,improper choice of inputs to the model can cause over-fitting and computationally intensive problems.Therefore,Gamma Test algorithm is proposed to extend the selection of input data of wavelet neural network.In this way,we can ensure the high accuracy,high efficiency and scalability of the subsequent prediction model.