基于独立分量相异性分析的动态过程监控研究  

Research on dynamic process monitoring based on dissimilarity analysis of independent component analysis

作　　者：张成[1] 李明业 潘立志 李元 ZHANG Cheng;LI Mingye;PAN Lizhi;LI Yuan(School of Science,Shenyang University of Chemical Technology,Shenyang 110142,China;School of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学理学院,辽宁沈阳110142 [2]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《应用科技》2023年第5期17-24,共8页Applied Science and Technology

基　　金：国家自然科学基金项目(62273242);辽宁省教育厅基础科研项目(LJKMZ20220792,LNYJG2022177).

摘　　要：针对动态独立分量分析(dynamic independent component analysis,DICA)在过程中捕获的独立分量存在自相关性和故障漏报的问题,提出基于独立分量相异性分析(DICA dissimilarity analysis,DICA-DISSIM)的动态过程监控方法。首先,利用DICA从原始数据中捕获独立分量;其次,在独立分量子空间中引入滑动窗口并进行相异性分析得到一个新的统计指标来监控过程的当前状态;最后,利用变量贡献图方法分析过程异常原因。与传统的DICA相比,所提方法能够有效地降低DICA捕获的独立分量的自相关性,降低了过程动态特征对故障检测的影响,最后解决了DICA统计量中存在的故障漏报问题。通过对动态数值例子和田纳西−伊斯曼过程进行仿真,仿真结果表明,与独立分量分析(independent component analysis,ICA)、DICA和动态主成分分析(dynamic principal component analysis,DPCA)相比,该方法有效地提高了动态过程监控性能。Aiming at the problems of autocorrelation and failure omission of the independent components captured in the process of dynamic independent component analysis(DICA),a dynamic process monitoring method based on independent component dissimilarity analysis is proposed in this paper.Firstly of all,DICA is used to capture independent components from the original data;secondly,the sliding window is introduced into the independent component subspace and the dissimilarity analysis is carried out to obtain a new statistical index to monitor the status of the current dynamic process;in the end,the contribution graph of variables is used to analyze the cause of abnormal process.Compared with traditional DICA,the proposed method can effectively reduce the autocorrelation of the independent component captured by DICA,reduce the influence of process dynamic characteristics on fault detection and finally solve the problem of missing fault report in DICA statistics.The simulation results of dynamic numerical examples and Tennessee-Eastman processes show that the proposed method can effectively improve the performance of dynamic process monitoring compared with independent component analysis(ICA),DICA and dynamic principal component analysis(DPCA).

关 键 词：动态独立分量分析 独立分量子空间 自相关性 相异指数 故障检测 动态过程 故障诊断 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]