摘要
通过对比相关系数法、逐步回归法及综合方法3个预报因子筛选方案的模拟结果,确定优选预报因子的最佳方法,采用BP人工神经网络模型对大通、屏山和汉口3个站点进行枯水期(当年11月-次年4月)径流预报研究.结果显示,采用相关系数法初选及逐步回归法优选所筛选出的预报因子集合可以得到更好的预报效果;该模型在枯水期月尺度径流预报中,检验期的平均合格率为56.44%,达不到实际预报的需求.而采用旬尺度模拟计算月径流的预报效果要远远高于月尺度径流模拟,检验期平均相对误差与合格率分别为12.27%和71.63%,有较好的预报精度.可以为长江流域水文预报工作提供一定的参考.
This study determined the best approach to select optimum forecast factors by comparing the simulation results of three selecting schemes, including correlation coefficient method, stepwise regression method and synthesis method, and introduced BP neural network model to predict seasonal low-flow in the Yangtze River basin for the period from November to next April. The results show that the factors select- ed at first by correlation coefficient method and then screened with stepwise regression analysis, can a- chieve better forecast results. And the monthly-scale runoff forecast based on BP neural network model can not meet the actual prediction requirements because the average qualified rate during the test period is only 56.44%. While the average relative error and qualified rate of monthly runoff calculated by ten-day-scale are 12.27% and 71.63% respectively, so as to provide a preferable reference for the work of the Yangtze River hydrological forecasting.
出处
《武汉大学学报（工学版）》
CAS
CSCD
北大核心
2015年第6期758-763,共6页
Engineering Journal of Wuhan University
基金
国家自然科学基金项目(编号:51339004
51279139)
国家水体污染控制与治理科技重大专项(编号:2014ZX07104-005)
中央高校基本科研业务费专项资金(编号:2042014kf1012
2042014kf0033)
关键词
人工神经网络
多方法优选因子
枯水期径流预报
长江流域
artificial neural network
multimethod optimized selection factors
low-flow runoff forecas-ting
Yangtze River basin
