摘要
将模型作为函数,模型的输入输出参数为变量,给出变量真值的定义。在此基础上,给出单一模型误差e,利用误差e求出模型输出变量真值的期望μ,用μ作为修正值以提高计算结果的精确度。通过给出模型集成误差λ的定义,详细推导了λ的计算方法,利用λ计算模型集成输出变量真值y的期望μy、方差σy2、σy2/vy2,μy作为最终计算结果的修正值,而σy2/vy2可作为模型集成选择的依据,对同一输出变量而言,σy2/vy2小的模型集成是较优的.
The paper considers one model as a function wherein the inputs and output of one model are regarded as variables and the definition of variable true values is also presented. Then it puts the definition of the error e of a single model forward whereby the expected valueμ,μof the output variable true value of one model can be calculated and considered as corrected value to enhance the precision of calculation. In the following, the definition of the error λof model integration and its detail calculation are also describeded. By using λ,the expected valueμy, σy^2、σy^2/υy^2,μycan be gotten.μy may be regarded as corrected value to the final result of model integration. At the same time, we can do model integration selection according toσy^2/υy^2. In all model integrations the least concomitant σy^2/υy^2 is the best selection for the same output variable.
出处
《运筹学学报》
CSCD
北大核心
2006年第4期89-98,共10页
Operations Research Transactions
基金
国家十五攻关计划项目[2002BA218C]
湖南省自然科学基金(06JJ20075)
关键词
运筹学
模型集成
误差
优化
Operations research, model integration, error, optimization