三种基于神经网络的洪水实时预报方案的比较研究

在总结神经网络应用的基础上,归纳了3种基于神经网络的洪水实时预报方案。第一种是神经网络水文模型的模拟模式加模拟误差的自回归校正模型,第二种是权重系数固定的神经网络实时预报方案,第三种是权重系数自动更新的神经网络实时预报方案。采用10个不同流域的日流量资料对这3种方案进行率定和校核。比较这3种方案的实时预报精度。结果发现,第三种方案不仅预报精度要高于其他两种方案,而且比第一种方案少了一个自回归校正模型,结构简洁。本文建议采用第三种洪水实时预报方案。

Study of Three Real-time Flood Forecasting Schemes Based on the Neural Network

From the review of applications for the artificial neural network(ANN)in hydrology,three real-time flood-forecasting schemes based on the ANN are generalized in this paper.The first scheme is the ANN model in the simulation model plus an AR model to forecast the simulation errors of the ANN model.The second one is the ANN model in the real-time mode with all the weights of the ANN being fixed.The third one is the ANN model in the real-time mode whose weights are continuingly updated by the back-propagation method.The daily data from the ten different watersheds are selected to test these three schemes in terms of their efficiency in real-time flood forecasting.It is found that the third scheme on average performs best in flood forecasting.Moreover,when compared to the first scheme,the third scheme is more parsimonious since it does not need any additional"correction mod el".It is recom-mended that the third real-time scheme be used in the flow forecasting practice.