偏最小二乘法与人工神经网络耦合的小流域产沙模型  

Sediment yield model for small watersheds based on coupling of partial least square regression and artificial neural network

作　　者：李世欣[1,2] 温建[3] 邵孝侯[1] 王晓亚 王玉英 

机构地区：[1]河海大学水利水电学院,江苏南京210098 [2]河南农业大学机电工程学院,河南郑州450002 [3]河南农业大学信息与管理科学学院,河南郑州450002 [4]河南省南召县水土保持科学研究站,河南南召474650

出　　处：《河海大学学报（自然科学版）》2010年第2期149-153,共5页Journal of Hohai University(Natural Sciences)

基　　金：国家自然科学基金(40801102)

摘　　要：针对小流域侵蚀产沙的复杂性,将偏最小二乘回归与人工神经网络耦合,建立了小流域降雨侵蚀产沙检验模型,并应用于小流域降雨侵蚀产沙预报.采用偏最小二乘法对多维自变量中的信息进行组合和提取,从而得到对因变量解释能力最强并可很好概括自变量信息的主成分,有效克服了变量之间的多重相关问题,实现了对高维数据的降维处理.把提取的主成分作为神经网络的输入,提高了网络的学习效率和稳健性.应用结果表明,偏最小二乘神经网络耦合模型的拟合和检验精度均优于偏最小二乘回归模型和人工神经网络模型精度.With regard to the complexity of sediment yield in small watersheds, a coupling partial least square regression and artificial neural network model was proposed by combining the partial least square regression with the artificial neural network. The proposed model was applied to the prediction of sediment yield owing to precipitation in small watersheds. The information of multi-dimensional independent variables was combined and extracted by use of the partial least square regression so as to find the most important components with strong interpretation capacity for dependent variables and satisfactory depiction for independent variables. Accordingly, the problem of multi-correlations among variables could be solved, and the amount of input dimensions could be reduced. The extracted components were regarded as the input of the artificial neural network. The leaming efficiency and robustness of the network were improved. The results show that the proposed model is of higher fitting and prediction precisions than the models based on the partial least square regression or the artificial neural network. Key words： small watershed; sediment yield; partial least square regression; artificial neural network

关 键 词：小流域 产沙 偏最小二乘法 人工神经网络 

分 类 号：P333.4[天文地球—水文科学]