年径流变化的BP神经网络预报模型研究

针对现有基于线性方法的年径流预报模型预报精度不高的问题,利用乌江洪家渡1963~2016年径流系列资料,以5~10月月平均流量作为预报影响因子,构建以年径流量为预报对象的BP神经网络模型,形成6-11-1的网络结构,并选取泛化能力强的贝叶斯规则法TRAINBR为训练函数。模拟结果表明,模型预报效果良好,对于年径流预报具有实用价值;BP神经网络模型相比逐步线性回归方法能更精确表达年径流预报因子与预报对象的映射关系;采用的训练函数TRAINBR能有效改善模型的泛化能力。研究成果可为径流预报提供参考。

Study on BP Neural Network Forecast Model for Annual Runoff Change

For the annual runoff prediction, prediction precision of the existing models based linear method is not high enough, so the article used the annual runoff series of Hongjiadu during 1963-2016, the monthly average flow from May to October as the input factors, the annual runoff as prediction object based the BP neural network model was established, which formed the 6-11-1network structure. TRAINBR, the Bayesian method of good generalization ability, was selected as training function. The simulation results show that the model has good prediction effect and practical value for annual runoff prediction. The BP neural network model can express the mapping relationship between annual runoff prediction factors and the prediction object more accurately than the stepwise linear regression method. The TRAINBR training function can effectively improve the generalization ability of the model. The research can provide reference for runoff forecasting.