人工神经网络模型在渭河下游洪水预报中的应用

通过对渭河下游站点的时间序列及空间分布的分析,确定出影响华县站流量的时间和空间信息,并将其引入神经网络模型;采用典型的BP神经网络,重点对网络的隐含层节点数、训练次数和学习率进行分析,构建了渭河下游华县断面流量预报的人工神经网络模型;并采用RMSE、NSC和相关系数 R作为模型效果评定标准,将其与传统多元统计回归模型进行了对比。结果表明,所建的BP神经网络模型较多元统计回归模型的预报效果有显著的提高。

Application of ANN Models in Flood Forecasting for the Lower Reach of the Weihe River

Artificial neural networks (ANNs) have been proven to be very successful in dealing with complicated non-linear problems. In this paper, ANNs are adopted to forecast daily flow in the lower reach of the Weihe River. The primary objective of this study is to investigate the possibility to integrate more temporal and spatial information in daily flow forecasting models, which is not easily attained in the traditional time-series models. In order to achieve this objective, correlation analysis is firstly made in this paper. Furthermore, several issues with ANN model application, e.g. the determination of hidden layer nodes and training iterations, are discussed. Model performance is analyzed and some results are presented. Model calibration and verification show that the precision of the ANN model is obviously higher than those of the traditional statistical models.