郭金玉 王霞

摘 要：為了解决工业数据的序列相关性以及数据的全局和局部结构在某些异常状态下的变化问题，通过“时滞偏移”方法将动态行为纳入多流形投影（multi-manifold projections，MMP）模型，提出一种动态多流形投影算法（dynamic multi-manifold projections，DMMP）在统计过程监测中的应用方案。首先，在原始采样数据的基础上加入时滞变量使之具有动态特性;其次，通过分别求解全局图最大值和局部图最小值，获得全局和局部结构信息;然后，构建统一框架，即全局图最大值和局部图最小值，提取高维动态数据有意义的低维描述;最后，通过对比统计量与控制限进行故障检测，将基于DMMP的监控方案通过田纳西-伊斯曼过程验证其可行性和有效性。仿真结果表明，DMMP的整体性能优于一些传统的保持全局或局部特征的算法。新算法解决了传统算法中具有时间相关性的数据信息获取不全面的问题，为提高传统算法在动态工业过程故障检测中的性能提供了参考。

关键词：自动控制其他学科;统计过程监测;全局图;局部图;时滞偏移;动态多流形投影

中图分类号：TP277   文献标识码：A

DOI：10.7535/hbkd.2022yx01002

收稿日期：2021-11-22;修回日期：2021-12-23;责任编辑：王淑霞

基金项目：国家自然科学基金（61673279）;辽宁省教育厅科学研究项目（LJ2019007）

第一作者简介：郭金玉（1975—），女，山东高唐人，副教授，博士，主要从事故障诊断、生物特征识别算法及应用方面的研究。

通讯作者：郭金玉副教授。E-mail：969554959@qq.com

Application of dynamic multi-manifold projections algorithmin statistical process monitoring

GUO Jinyu，WANG Xia

（College of Information Engineering，Shenyang University of Chemical Technology，Shenyang，Liaoning 110142，China）

Abstract：To solve the problem of serial correlation of industrial data and the changes of global and local structure of data in some abnormal states，the "time lag migration" method was used to incorporate dynamic behavior into the multi-manifold projections（MMP） model，and an application scheme of dynamic multi-manifold projections（DMMP） algorithm in statistical process monitoring was proposed.Firstly，time-lag variables were added to the original sample data to make it dynamic.Secondly，the global and local structure information was obtained by solving the global graph maximum and local graph minimum separately.A unified framework，i.e.global graph maximum and local graph minimum，was constructed to extract meaningful low-dimensional representations for high-dimensional dynamic data.Finally，the fault detection was performed by comparing the statistics with control limits.The feasibility and effectiveness of the monitoring scheme based on DMMP was verified by the Tennessee-Eastman process.The simulation results show that the overall performance of DMMP is better than those of some traditional preserving global or local feature algorithms.The new algorithm solves the problem of incomplete acquisition of time-dependent data information in traditional algorithms，and provides a reference for improving the performance of traditional algorithms in fault detection of dynamic industrial process.

Keywords：

other disciplines of automatic control technology;statistical process monitoring;global graph;local graph;time lag migration;dynamic multi-manifold projections

随着中国向高质量方向发展的迈进，现代工业生产过程日趋精细化、复杂化，因此必须保证生产操作系统不出现任何故障。……