文献类型：期刊文献

英文题名：Novel Monitoring Strategy Combining the Advantages of the Multiple Modeling Strategy and Gaussian Mixture Model for Multimode Processes

作者：Zhang, Shumei[1];Wang, Fuli[1,2,3];Tan, Shuai[4];Wang, Shu[1,2,3];Chang, Yuqing[1,2,3]

机构：[1]Northeastern Univ, Coll Informat Sci & Engn, Shenyang 110819, Liaoning, Peoples R China;[2]Northeastern Univ, State Key Lab Integrated Automat Proc Ind Technol, Shenyang 110819, Liaoning, Peoples R China;[3]Northeastern Univ, Res Ctr Natl Met Automat, Shenyang 110819, Liaoning, Peoples R China;[4]E China Univ Sci & Technol, Key Lab Adv Control & Optimizat Chem Proc, Minist Educ, Shanghai 200237, Peoples R China

年份：2015

卷号：54

期号：47

起止页码：11866

外文期刊名：INDUSTRIAL & ENGINEERING CHEMISTRY RESEARCH

收录：;EI(收录号：20154901642622);WOS:【SCI-EXPANDED(收录号:WOS:000365930800012)】；

基金：The authors gratefully acknowledge support from the following foundations: National Natural Science Foundation of China (61533007, 61374146, and 61403072), State Key Laboratory of Synthetical Automation for Process Industries Fundamental Research Funds (2013ZCX02-04), and the Fundamental Research Funds for East China University of Science and Technology (22A201514050).

语种：英文

外文关键词：Gaussian distribution

摘要：The multiple modeling strategy and Gaussian mixture model (GMM) have been widely used to monitor multimode processes. On the basis of a deterministic view, multiple modeling strategy builds the specific model for each mode, which can extract more accurate information for monitoring. However, multiple modeling strategy is unable to deal with the situation in which the online mode information cannot be determined, and this condition easily leads to a severe error when an inappropriate model is used for monitoring. GMM builds a mixture model for the whole process from a probabilistic view. It unites all the models probabilistically for monitoring without having to identify the mode information. However, it may perform badly for some specific modes because some irrelevant models of other modes are introduced by GMM. Besides, it may not efficiently capture the local features especially for complex processes with transitional modes. In this paper, a novel monitoring strategy, which combines the advantages of multiple modeling strategies and GMM, is proposed for multimode processes. All possible models are probabilistically united for monitoring when the mode cannot be identified for sure. If the mode can be determined completely, the corresponding model is deterministically used for monitoring. To evaluate the feasibility and efficiency of the proposed method, the Tennessee Eastman challenge is demonstrated to compare the proposed method with multi-PCA and traditional GMM.