基于人工神经网络的长期径流预报模型研究

Long-term runoff forecasting model based on artificial neural network

调水工程供水效益的发挥主要取决于引水过程与受水区本地径流过程的有机结合。为了预测水库未来径流信息,指导供水工程调度及实现水资源高效配置,以鄂北地区水资源配置工程受水区的封江口水库为研究对象,利用BP神经网络对非线性混合回归模型进行年、月尺度的长期径流预报。结果表明:利用该方法得到的年尺度长期径流预报确定性系数为0.790,相对均方根误差为20.7%,预报合格率为66.7%;与降雨-径流相关曲线法预报结果相比,该方法确定性系数提高了9.6%,相对均方根误差降低了18.8%,预报合格率提高了11.9%。月尺度长期径流预报确定性系数为0.714,相对均方根误差为77.6%,预报合格率为52.8%,亦优于降雨-径流相关曲线法预报结果。利用人工神经网络进行水库长期径流预报可提高精度。

The performance of water transfer project s water supply efficiency depends largely on the organic combination of water diversion and local runoff process in water demand area.In order to grasp the future runoff information of the reservoir to direct reservoir operation and realize efficient allocation of water resources,BP artificial neural network was adopted to forecast the long-term runoff of Fengjiangkou Reservoir in the demand area of North Hubei Water Transfer Project in this study.The results showed that the certainty coefficient of long-term runoff forecast in annual period was 0.790,the relative root mean square error was 20.7%,and the qualified rate was 66.7%.Compared with the results of rainfall-runoff correlation curve method,the certainty coefficient was increased by 0.069,the relative root mean square error was reduced by 4.8%,and the qualified rate of the forecast was increased by 7.1%.The certainty coefficient,relative root mean square error,and qualified rate of the monthly runoff forecasting were 0.714,77.6%and 52.8%,respectively,which were better than the results of rainfall-runoff correlation curve method.Long-term reservoir runoff forecasting by using artificial neural networks can improve its accuracy.