基于GMM-Ada-LASSO模型的高维过程统计质量监控方法  

Statistical Quality Monitoring Method Based on GM

作　　者：张帅 杨剑锋[2] 薛丽[3] Zhang Shuai;Yang Jianfeng;Xue Li(School of Management Engineering,Henan University of Engineering,Zhengzhou 451191,China;School of Business,Zhengzhou University,Zhengzhou 450001,China;School of Management Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450046,China)

机构地区：[1]河南工程学院管理工程学院,郑州451191 [2]郑州大学商学院,郑州450001 [3]郑州航空工业管理学院管理工程学院,郑州450046

出　　处：《统计与决策》2024年第17期47-52,共6页Statistics & Decision

基　　金：国家自然科学基金资助项目(U1904211;71672182);国家社会科学基金资助项目(20BTJ059);河南省重点研发专项(241111212000;241111222700)。

摘　　要：针对高维数据往往不服从正态分布导致统计监控模型识别精度低、监控效率差的问题,文章提出一种基于高斯混合模型的变量选择控制图方法。首先,利用高斯混合模型将高维过程分解成若干个服从正态分布的子分布;然后,运用Adaptive LASSO算法识别潜在异常变量;最后,构建多元EWMA控制图实现高维过程统计质量监控。通过仿真实验,在六种不同情形下对所提方法的监控性能进行测试。结果表明,与传统MEW⁃MA和VS-MEWMA控制图相比,所提监控方法对非正态数据具有较强的稳健性,对高维过程具有良好的监控性能。High dimensional data often do not obey normal distribution,which leads to low accuracy and poor monitoring efficiency of statistical monitoring model.In view of this problem,the paper proposes a variable selection control chart method based on Gaussian mixture model.Firstly,the Gaussian mixture model is used to decompose the high-dimensional process into several sub-distributions which obey the normal distribution.Then,Adaptive LASSO algorithm is used to identify potential abnormal variables.Finally,a multivariate EWMA control chart is constructed to realize the quality control of high dimensional process statistics.Through simulation experiments,the monitoring performance of the proposed method is tested in six different situations.The results show that compared with the traditional MEWMA and VS-MEWMA control charts,the proposed monitoring method is more robust for non-normal data and has better monitoring performance for high dimensional processes.

关 键 词：高维数据 非正态过程 高斯混合模型 变量选择控制图 

分 类 号：F224.7[经济管理—国民经济]