摘要
为对数据驱动的径流预报模型的预报因子进行方案优选,利用相关系数法(CA)对松花江流域上游3个水文代表站(大赉站、扶余站、哈尔滨站)的月径流、佳木斯站的月降水和74项大气环流指数等潜在预报因子进行初筛,然后用主成分分析法(PCA)、互信息法(MI)进一步优选因子,最后基于支持向量机回归模型(SVR)结合选取的潜在预报因子(共9种组合方式)预报松花江流域佳木斯站的月径流,并采用均方根误差(R_(MSE))、平均相对误差(M_(RE))、决定性系数(R^(2))、合格率(Q_(R))评价不同预报因子组合间的SVR模型的预报性能。结果表明,SVR结合7因子的PCA模型与其他因子组合的SVR模型相比,更适用于松花江流域的月径流预报,其R_(MSE)、M_(RE)、R^(2)、Q_(R)分别为856.68m^(3)/s、31.4%、0.8094、73.3%,各项评价指标均优于其他因子组合方案。
The objective of this paper is to select the optimal predictors for the data-driven streamflow forecasting model.The method of correlation coefficient(CA)was preliminarily utilized to filtrate potential predictors that including the monthly runoff of three hydrological stations(Dalai,Fuyu and Harbin)in the upstream of the Songhua River Basin,the monthly precipitation in Jiamusi station and 74 atmospheric circulation indexes.The principal component analysis(PCA)and mutual information(MI)were applied to further select optimal potential prediction factors.Finally,the monthly streamflow of Jiamusi station located in the Songhua River Basin was predicted by combining the selected optimal predictors(a total of 9 situations)based on the support vector regression model(SVR).Moreover,the R_(MSE),M_(RE),R^(2)and QR were used to evaluate the performance of SVR under the combination of different potential predictors.The results show that the SVR-PCA 7 model was more suitable for monthly streamflow forecasting in Songhua River Basin,and its R_(MSE),M_(RE),R^(2)and Q_(R)were 856.68m^(3)/s,31.4%,0.8094 and 73.3%,respectively.All evaluation metrics of this model were better than other schemes of combination factors.
作者
朱春苗
吴海江
宋小燕
宋松柏
ZHU Chun-miao;WU Hai-jiang;SONG Xiao-yan;SONG Song-bai(College of Water Resources and Architectural Engineering,Northwest A&F University,Yangling 712100,China)
出处
《水电能源科学》
北大核心
2021年第6期12-15,41,共5页
Water Resources and Power
基金
国家自然科学基金项目(41501022)
国家科技基础资源调查专项(2017FY100904)
中央高校基本科研业务费(2452020167)。
关键词
相关系数法
主成分分析法
互信息法
SVR模型
松花江流域
correlation analysis method
principal component analysis
mutual information method
support vector regression model
Songhua River Basin
