基于Elman网络和实时递归学习的洪水预报研究

Research on Flood Forecasting Based on Elman Neural Network and Real-time Recurrent Learning

流域洪水形成过程异常复杂且动态变化,一般难以用传统的静态神经网络进行描述。然而,具有反馈连接的动态神经网络能够很好地反映洪水过程这种动态变化特征。为此,研究基于Elman动态网络构建流域洪水预报模型,采用具有在线学习功能的实时递归学习算法进行模型训练,并将所建模型和算法运用于淮河水系响洪甸水库的入库洪水实时预报中。结果表明,所建模型预报精度高,实时性强,能够为流域的防洪决策提供支持。

It is usually difficult to describe complex and dynamic flood process by traditional static neural networks.However,the dynamic neural networks with feedback connection can reflect this dynamic feature of flooding.The Elman neural network is used to build the flood forecasting model.This model is trained by the real-time recurrent learning algorithm which is suitable for on-line learning.The model and algorithm are applied to forecast the real-time flood flow into Xianghongdian Reservoir in Huai River.The results show that the Elman network with real-time recurrent learning can be utilized with high accuracy to the study of real-time flood forecasting.