%0 Journal Article
%T 基于机器学习模型的短期降雨多模式集成预报
%A 彭 勇
%A 李福威
%A 王子茹
%A 疏杏胜
%J 南水北调与水利科技
%D 2020
%X 短期降雨预报对洪水预报和水库调度极为重要, 提高短期降雨预报精度有着重要的意义。以 TIGGE 资料中 心的 ECMWF、CM A 及 NCEP 三个集合预报中心发布的桓仁水库流域预报降雨数据为基础, 利用 ANN、ELM 以及 SVM 模型对桓仁水库流域未来 1~ 3 d 降雨进行多模式集成预报, 以期提高预报精度, 并从绝对平均误差、均方根 误差、相对误差、纳什系数、预报准确率等多个方面分析了集成预报的效果。试验结果表明, 基于 SVM 和 ELM 的 多模式集成预报模型预报效果均优于单一模式, 基于 ANN 的集成预报模型在输入因子选择合适的情况下, 其预报 效果也优于单一模式, 三种模型中, SVM 模型对降雨预报精度改善最为明显。说明基于机器学习模型的多模式降 雨集成预报方法可行且能够提高短期预报降雨精度。  H igh accuracy o f sho rt2term rainfall for ecast ing is of g reat impo rtance fo r flood for ecasting and reserv oir operat ion. It can not only improv e the accuracy o f floo d for ecasting but also make t he r eser voir operation mo re scient ific and reasonable. Based o n the pr edicted rainfall of the Huanren reservo ir basin using ECMWF, CMA and NCEP in the T IGGE datasets, the artif icial neural netw or k ( ANN) , support v ect or machine ( SVM) and ext reme lear ning machine ( ELM) mo dels w ere dev eloped to simulate and forecast the rainfall o f Huanren reservo ir basin in the nex t 1 to 3 day s, and the effect o f the forecasting results wer e analyzed fr om the aspects of mean absolute err or ( MAE) , roo t mean square error ( RMSE) , Bias, Nash2Sutcliffe efficiency co efficient ( NSE) and predictio n accur acy. Results showed that the int eg rat ed forecasting mo dels based o n SVM and ELM wer e better than the sing le mo dels, and the integ rated models based on ANN wer e better t han the sing le mo dels w hen the input facto rs w ere selected pr operly. Amo ng the thr ee integr ated models, SVM model had the mo st o bv ious impr ovement in ra infall forecasting accuracy, which indicated that the multi2mo del rainfall int eg rat ed fo recast ing method based o n machine learning model was feasible and co uld impr ov e the accuracy o f sho rt2term rainfall fo recast ing .  国家重点研发计划( 2017YFC0406005)
%K 多模式
%K 机器学习模型
%K 短期降雨
%K 集成预报
%K multi-mode
%K machine learning model
%K rainfall
%K integrated forecasting
%U http://www.nsbdqk.net/nsbdyslkj/article/abstract/20200106?st=article_issue