基于极限学习机模型的中国西北地区参考作物蒸散量预报  

Prediction of Reference Crop Evapotranspiration in Northwest China Based on Extreme Learning Machine Model

作　　者：魏俊 崔宁博[1,2] 陈雨霖 张青雯 冯禹 龚道枝[4] 王明田 WEI Jun;CUI Ning-bo;CHEN Yu-lin;ZHANG Qing-wen;FENG Yu;GONG Dao-zhi;WANG Ming-tian(State Key Laboratory of Hydraulics and Mountain River Engineering and College of Water Resources and Hydropower,Sichuan University,Chengdu 610065,Sichuan Province,China;Provincial Key Laboratory of Water-saving Agriculture in Hilly Area of South China,Chengdu 610066,Sichuan Province,China;Key Laboratory of Agricultural Soil and Water Engineering in Arid and Semiarid Areas,Ministry of Education,Northwest Agricultural and Forestry University,Yangling 712100,Shaanxi Province,China;State Engineering Laboratory for Efficient Water Use and Disaster Loss Reduction of Crops,Institute of Environment and Sustainable Development in Agriculture,Chinese Academy of Agricultural Sciences,Beijing 100081,China;Agricultural Meteorology Center of Sichuan,Chengdu 610071,Sichuan Province,China)

机构地区：[1]四川大学水力学与山区河流开发保护国家重点实验室,四川成都610065 [2]南方丘区节水农业研究四川省重点实验室,四川成都610066 [3]西北农林科技大学旱区农业水土工程教育部重点实验室,陕西杨凌712100 [4]中国农业科学院农业环境与可持续发展研究所作物高效用水与抗灾减损国家工程实验室,北京100081 [5]四川省农业气象中心,四川成都610071

出　　处：《中国农村水利水电》2018年第8期35-39,共5页China Rural Water and Hydropower

基　　金："十三五"国家重点研发计划课题(2016YFC0400206);国家自然科学基金项目(51779161);"十二五"国家科技支撑计划课题(2015BAD24B01);2017年中央高校基本科研业务费专项资金资助

摘　　要：为有效提高西北地区参考作物蒸散量(ET_0)预报精度,在西北地区选择6个代表性气象站点,以P-M模型计算的ET_0作为标准值,利用1993-2016年逐日气象资料构建10种极限学习机(extreme learning machine,ELM)ET_0预报模型,用k-折交叉验证估计模型泛化误差,并将其与Hargreaves-Samani、Chen、EI-Sebail和Bristow等4种在西北地区计算精度较高的模型进行比较。结果表明:ELM_1(输入T_(max)、T_(min)、RH、n和u_2)、ELM_2(输入T_(max)、T_(min)、n和u_2)、ELM_4(输入T_(max)、T_(min)、RH和u_2)及ELM_7(输入T_(max)、T_(min)和u^2)模型均具有较高模拟精度,其MAE分别为0.199、0.209、0.250、0.273 mm/d,RMSE分别为0.270、0.285、0.341、0.422 mm/d,NSE分别为0.983、0.981、0.973、0.987,R^2分别为0.984、0.982、0.975、0.960,整体评价指标(global performance indicator,GPI)排名分别为1、2、3、4;模型可移植性分析表明,ELM模型具有较强的泛化能力,除了ELM_7在喀什站、敦煌站的模拟精度相对较低之外,其余ELM模型在西北地区各站点模拟结果的MAE均在0.40 mm/d以下、RMSE均在0.49以下、NSE均在0.95以上、R^2均在0.96以上;在相同输入的情况下ELM模型模拟精度均高于HargreavesSamani、Chen、EI-Sebail和Bristow。因此,在气象资料缺乏情景下ELM模型可作为西北地区ET_0计算的推荐模型。In order to effectively improve the forecast accuracy of reference crop evapotranspiration（ ET0） in Northwest China,six representative meteorological stations are selected in the northwest region,and the ET0 calculated by using the PM model is used as the standard value,the daily meteorological data from 1993 to 2016. Ten extreme machine learning ET0 forecasting models are used to estimate the model generalization error by K-fold cross-validation and calculated with the four types of Hargreaves-Samani,Chen,EI-Sebail and Bristow in the northwestern region. Higher accuracy models are compared. Results show： ELM1（ input T（max）,T（min）,RH,n,and u^2）,ELM2（ input T（max）,T（min）,n and u^2）,ELM4（ input T（max）,T（min）RH and u2） and ELM7（ input T（max）,T（min）and u2） All the models have high simulation accuracy with MAE of 0.199,0.209,0.250,0.273 mm/d,RMSE of 0. 270,0. 285,0. 341,0. 422 mm/d,NSE of 0. 983,0. 981,0. 973,0.987,and R^2 of 0. 984,0. 982,0. 975 and 0. 960. The global performance indicator rankings are 1,2,3,and 4 respectively. Model portability analysis shows that the ELM model has strong generalization ability,in addition to the relatively low simulation accuracy of ELM7 between Dunhuang and Kashgar rivers,the MAE of the remaining models at different sites in the Northwest China is 0. 40 mm/d. Below,RMSE is all below 0.49 mm/d,NSE is above 0.95,and R^2 is all above 0.96. the ELM model has strong generalization ability. In the case of the same input,the simulation accuracy of the lower ELM model is higher than Hargreaves-Samani,Chen,EI-Sebail and Bristow. Therefore,in the absence of meteorological data,the ELM model can be used as a recommended model for the calculation of ET0 in the Northwest China.

关 键 词：蒸散量 预报模型 极限学习机 K-折交叉验证 西北地区 可移植性 

分 类 号：TV93[水利工程—水利水电工程]