高效的因子分类筛选方法及仿真应用

针对序贯分支（SB）方法难以对位置效应和散度效应同时筛选的情形，基于多重序贯分支的基本思想，同时考虑位置和散度效应，将显著因子分类别筛选出来。首先，通过“预分支”步骤，将因子按照其位置和散度效应符号分组；然后，采用序贯分支方法将所有子组同时进行位置和散度效应显著性检验，在结合序贯概率比检验（SPRT）控制筛选过程中的第一类和第二类错误的基础上，实现因子的分类筛选，以便后续阶段的建模及优化；最后，通过仿真试验说明所提方法在解决因子分类筛选问题上的有效性、高效性及稳健性。

An Efficient Factor Screening and Classification Procedure Based on Multi-Response Sequential Bifurcation and Applications

Sequential bifurcation(SB)is the most effective and efficient method of factor screening for many input problems.However,SB usually ignores the factors which are significant for dispersion effects.Besides,it could not be used to simultaneously screen the location and dispersion effects due to its basic assumptions.'Therefore,in this paper a novel screening procedure based on multi-response sequential bifurcation was proposed to screen and classify factors simultaneously.First,the factors were pre classified into two categories according to the signs of location and dispersion effects.Next,for any category,a new procedure combined with the sequential probability ratio test(SPRT)which included stopping rules and importance tests to control the two types of error was proposed to screen and classify all the significant factors into three classes.Finally,a simulation model was employed to verify the effectiveness and robustness of the proposed screening procedure.The results show that considering both the location and the dispersion simultaneously has more economic efficiencies than considering them separately.