基于混沌径向基神经网络模型的洪水预测研究

Flood forecasting research based on the chaotic RBF neural network model

【目的】建立更为理想的洪峰大流量预测模型。【方法】针对目前大部分预测模型对洪峰大流量数值预测结果不太理想的问题,根据自然界中普遍存在的混沌现象,在对洪水系统进行混沌识别的基础上,建立了基于混沌理论和径向基神经网络(RBF)的预测模型,将实测洪水序列进行相空间重构得到训练样本,利用MATLAB 7.0工具箱确定网络结构。【结果】将所建立的RBF模型用于汾河石滩水文站2004年最大洪水的预测,结果表明,该模型预测结果的合格率、平均相对误差、相关系数(R)、均方根误差(RMSE)和Nash-Sutcliffe系数(NSC)分别为100%,4.69%,0.9793,4.2260和0.9552,而传统Volterra级数自适应预测模型的分别为93.75%,8.97%,0.9540,10.2632和0.7358,可见RBF模型的预测结果较好,并且对预测洪峰大流量数值取得了较理想的预测效果。【结论】将混沌理论和径向基神经网络结合建立预测模型,作为提高洪水预报精度的一种新尝试,对以后进行洪水预测研究具有一定的参考价值。

【Objective】 The study was to establish a better flow of the flood forecasting model.【Method】 At present,most of the forecasting models of the great flood peak flow numerical prediction are not ideal.In the chaos of the flood systems on the basis of identification,forecasting model was estaldished based on chaos theory and RBF neural network to measure flood sequence of space reconstruction by training samples,and the network structure was determined by using MATLAB 7.0 toolbox.【Result】 The RBF forecast model was used by Fenhe Shitan Hydrometric Station in 2004 to measure the largest flood forecasts,and the results showed the pass rate,with an average relative error,correlation coefficient （R）, root mean square error （RMSE） and Nash -Sutcliffe coefficient （NSC） were 100% ,4.69% ,0. 979 3,4.226 0 and 0. 955 2, and those of the traditional Volterra adaptive prediction model were 93. 75%, 8. 97%, 0. 954 0, 10. 263 2 and 0. 735 8. RBF model can have better predication results and has made better numerical prediction of large flow flood peak. [Conclusion] To build predictive models based on Chaos Theory and the RBF neural network can be a new attempt in improving flood forecasting accuracy, which has some reference value on flood forecasting in the future.