基于组合建模技术的串行多阶段参数优化  

SerialMultistage Parameter Optimization Based on Ensemble Modeling Technique

作　　者：肖甜丽 马义中[1] 林成龙 XIAO Tianli;MA Yizhong;LIN Chenglong(School of Economics and Management,Nanjing University of Science and Technology,Nanjing,Jiangsu 210094,China)

机构地区：[1]南京理工大学经济管理学院,江苏南京210094

出　　处：《工业工程与管理》2021年第5期82-90,共9页Industrial Engineering and Management

基　　金：国家自然科学基金资助项目(71931006,71871119);江苏省研究生科研创新计划项目(KYCX19_0350,KYCX20_0284)。

摘　　要：针对串行生产系统中模型形式不确定导致模型误差累积问题,提出结合组合建模技术和最大最小满意度函数的多阶段多响应优化方法。首先,利用单个元模型自身绝对预测精度及其在全体元模型中相对预测精度,改进一般组合建模技术;其次,针对不同阶段所有响应,利用改进组合建模技术进行组合响应面的构建;然后,采用最大最小满意度法集成串行系统的多个阶段;最后,通过蚁群优化算法获得各阶段的最佳参数组合。仿真案例结果表明:与单一元建模和一般组合建模方法相比,所提方法的模型预测精度更高,优化结果更可靠。A multistage-multiresponse optimization method combined ensemble modeling technique and“maximin”desirability method was presented for the problem of model error accumulation caused by model-form uncertainty in a serial manufacturing system.Firstly,absolute prediction accuracy of the single model itself and relative prediction accuracy among all single models were used to improve general ensemble model.Then,the modified ensemble modeling technique was utilized to build response surface for all the responses in different stages.Furthermore,multiple stages of the serial system were integrated by the“maximin”desirability method,and then optimal parameter combination of each stage was obtained by ant colony optimization.The results of simulation case illustrate that the proposed method has higher model prediction accuracy and more reliable optimization results,compared with single modeling technique and general ensemble modeling technique.

关 键 词：串行生产系统 响应曲面法 模型形式不确定 组合建模 最大最小满意度法 

分 类 号：F273.2[经济管理—企业管理]