摘要
长期径流预报对于掌握未来径流信息,实现水资源的高效利用具有重要意义。当前长期径流预报可利用模型众多,且各模型在不同预报条件下表现各有优劣,为实现多模型间的相互协调和性能互补,以桓仁流域长期径流预报为研究对象,耦合相关性分析和向前搜索包裹法从众多水文-气象因子中筛选影响桓仁水库入库径流的关键预报因子;采用统计分析法和机器学习法共6种径流预报方法,分别建立桓仁流域年径流预报模型和汛期月径流预报模型,对比分析多模型在该流域长期径流预报的适用性。结果表明:大气环流因子与预报对象的相关性明显高于水文因子,其在流域长期径流预报中起关键作用;基于主成分分析的人工神经网络(Artificial Neural Network Model based on Principal Compo⁃nent Analysis,PCA-BP-ANN)、支持向量机(Support Vector Machine,SVM)等机器学习模型的年径流预报效果优于传统统计模型;汛期各月径流预报中,各模型预报精度有所差异,基于主成分分析的人工神经网络模型(PCA-BP-ANN)在5月份和8月份的预报合格率最高,且相对人工神经网络(Back Propagation-Artificial Neural Network,BP-ANN)模型提升了10%左右,但在6、7月份的预报效果不如其他模型,而门限回归(Threshold Regression,TR)模型在7月份表现最佳,合格率达94%;选择汛期各月表现最优的预报模型,给出综合多模型预报方案,在最优预报方案下,桓仁流域年径流预报以及汛期相对重要的7、8月径流预报的合格率均能达到90%以上。
Long-term runoff forecast is of great significance for mastering future runoff information and realizing efficient utilization of water resources.At present,there are many models for long-term runoff forecasting,and each model has its own advantages and disadvantages un⁃der different forecasting conditions.In order to achieve mutual coordination and performance complementarity among multiple models,this paper selects the key predictors affecting the runoff in Huanren River Basin from the hydrological and meteorological factors by coupling the correlation analysis and forward-searching parcel method.Six runoff forecasting methods,including the statistical analysis and machine learning method are applied to forecast the annual and monthly runoff.And the applicability of the multi-model in the long-term runoff fore⁃cast of the basin is compared and analyzed.The results show that the correlations between the atmospheric circulation factors and the forecast⁃ed runoff are higher.The annual runoff forecast accuracy of the machine learning models,such as the artificial neural network model based on principal component analysis(PCA-BP-ANN)and the support vector machine(SVM)is higher than that of traditional statistical models.For the monthly runoff forecast in the flood season,the forecast accuracy of each model varies from month to month.The artificial neural net⁃work model based on principal component analysis(PCA-BP-ANN)has the highest forecast qualified rate in May and August and is about 10%higher than the artificial neural network model(BP-ANN),but the forecast effect is not as good as other models in June and July.While the threshold regression model(TR)performs best in July,with a qualified rate of 94%.The mode is selected with the best perfor⁃mance in each month of the flood season to give a comprehensive multi-model forecast scheme.Under the optimal forecast scheme,the quali⁃fied rate of annual runoff forecast and the monthly runoff in July and August in the flood season are both higher than 90%.
作者
李福威
包爱美
疏杏胜
丁伟
LI Fu-wei;BAO Ai-mei;SHU Xing-sheng;DING Wei(Heyu Hydropower Development Company,Guodian Electric Power Development Co.,Ltd.,Benxi 117201,Liaoning Province,China;School of Hydraulic Engineering,Dalian University of Technology,Dalian116024,Liaoning Province,China)
出处
《中国农村水利水电》
北大核心
2022年第11期6-12,共7页
China Rural Water and Hydropower
基金
国家重点研发计划项目(2021YFC3000205)。
关键词
长期径流预报
多模型预报
预报因子
机器学习
桓仁流域
long-term runoff forecast
multiple models
predictors
machine learning
Huanren River Basin
