面向质量设计的Kriging组合建模技术  

Ensemble Kriging modeling technique for quality design

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 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：《计算机集成制造系统》2021年第7期2023-2034,共12页Computer Integrated Manufacturing Systems

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资助项目(KYCX19_0350)。

摘　　要：Kriging代理模型广泛用于替代计算昂贵的工程仿真模型。而现有单个核函数Kriging建模技术的预测性能依赖于具体问题,往往在不同情况下表现差异较大,缺乏普适性和稳健性。针对该问题,研究了具有多个核函数的Kriging组合建模,提出同时考虑核函数选择和多组权重因子。首先基于随机搜索变量选择法选择显著核函数组合,其次借助K均值聚类法获得多组权重因子,最后结合选择的显著核函数和多组权重因子构建Kriging组合模型。仿真算例和工业实例的比较结果表明,所提方法不仅产生更为精确和稳健的预测,而且能为质量设计提供可靠的优化参数组合。Kriging surrogate model is widely used to replace computationally expensive engineering simulation.However,prediction performance of existing Kriging modeling techniques with stand-alone kernel function is problem dependent,which often varies greatly under different conditions,lacking generality and robustness.To solve this problem,ensemble Kriging modeling with multiple kernel functions was investigated.Meanwhile,the kernel function selection and multi-group weight factors were considered.Significant kernel functions were chosen based on stochastic search variable selection method.Multi-group weight factors were derived by K-means clustering method.Furthermore,ensemble Kriging model combined with selected kernel functions and multi-group weight factors was constructed.The comparison results from simulation functions and practical industrial example showed that the proposed method could not only generate more accurate and robust prediction,but also provide a reliable optimized parameters scheme for quality design.

关 键 词：KRIGING模型 组合建模 核函数选择 多组权重因子 质量设计 

分 类 号：TH122[机械工程—机械设计及理论]