摘要
针对一类与时间相关的质量特性存在定时删失情形的稳健参数设计问题,提出了一种结合期望最大化算法和改进的随机森林算法的变量选择与过程优化方法.首先,采用期望最大化算法计算不同水平组合下的位置与散度估计;其次,利用改进的随机森林算法选择重要的因子效用;然后,将响应的置信区间引入优化策略中以构建更保守的约束条件,降低模型不确定对优化结果的影响;最后,通过一个实际的工程案例验证所提方法的有效性.实例分析的结果表明,所提方法能有效降低信息损失及模型不确定性对建模和优化结果的影响,能获得更加可靠的可控因子最优设计值.
A variable selection and process optimization method,which combines the expectation maximization algorithm with the modified random forest algorithm,is proposed to solve a class of robust parameter design problems with time-censored quality characteristic.Firstly,the expectation maximization algorithm is used to estimate the position and dispersion under different levels of combination.Secondly,the modified random forest algorithm is used to select important factors.Thirdly,the confidence interval of the response is introduced into the optimization strategy to construct a more conservative constraint,thus reducing the impact of model uncertainty on the optimization results.Finally,a real engineering case verifies the effectiveness of the proposed method.The results of case study show that the proposed method can effectively reduce the impact of information loss and model uncertainty on modeling and optimization results,and can obtain more reliable optimal settings of controllable factors.
作者
杨世娟
汪建均
马义中
马妍
翟翠红
YANG Shijuan;WANG Jianjun;MA Yizhong;MA Yan;ZHAI Cuihong(School of Economics and Management,Nanjing University of Science and Technology,Nanjing 210094,China)
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2021年第9期2392-2403,共12页
Systems Engineering-Theory & Practice
基金
国家自然科学基金(71771121,71931006)
江苏省研究生科研与实践创新计划项目(KYCX200344)。
关键词
试验设计
参数设计
期望最大化算法
变量选择
位置与散度
experiment design
parameter design
expectation-maximum algorithm
variable selection
location and dispersion