基于非参数控制图的数据质量监控方法改进研究  

Improvement on Data Quality Monitoring Methods Based on Nonparametric Control Charts

作　　者：吴纯杰[1] 李泓玮 WU Chun-jie;LI Hong-wei(School of Statistics and Management,Shanghai University of Finance and Economics,Shanghai 200433,China)

机构地区：[1]上海财经大学统计与管理学院,上海200433

出　　处：《数理统计与管理》2024年第1期1-19,共19页Journal of Applied Statistics and Management

基　　金：国家自然科学基金面上项目(11871324);全国统计科学研究项目重大项目(2020LD03)。

摘　　要：为监控多条多变量数据流的数据质量,本文提出了几种监控方法。本文首先介绍了统计过程控制领域的背景知识。之后本文采用闵式距离准则,基于3类常用的闵式距离,提出了3种EWMA类型的非参数控制图,此外,本文基于向量内积提出了第4种EWMA类型的非参数控制图。首先通过统计模拟,比较了4种控制图在不同的相关性水平下,对单条数据流过程均值漂移的监控效果。随后本文研究了平滑系数λ的取值对监控效果的影响。之后,本文将上述4种思路拓展到了大样本的情形,提出了5种可以监控多条数据流质量的方法。将这5种方法和现有的一种典型方法做了比较,比较了它们对不同幅度漂移的监控效果。最后使用山体滑坡的监控数据,比较了它们对实际数据的监控效果,显示了本文提出方法的优势。In order to monitor the data quality of multivariate data streams,we propose several monitoring methods.We first introduce some background information of statistical process control,which is necessary to understand this paper.We adopt the Minkowski distance standard,propose 3 EWMA nonparametric control charts based on three kinds of common Minkowski distance.We also propose a fourth EWMA nonparametric control chart based on vector product.In the simulation part,first we compare the performance of the 4 charts when mean shift occur in one data stream under different correlation levels.Then we study how can the smoothing parameter influence the performance of the charts.Later,we apply the 4 charts in the large sample case and propose 5 methods to monitor the data quality of several data streams.The performance of the 5 methods are compared with an existing typical method,in cases when different scales of shift occur.Finally,the advantage of the methods we propose is showed when they are used to monitor real data in a landslide example.

关 键 词：数据质量 EWMA控制图 非参数控制图 闵式距离 

分 类 号：O212[理学—概率论与数理统计] O213[理学—数学]