Dynamic controlled pattern extraction and pattern-based model predictive control. (January 2022)

• Record Type: Journal Article
• Authors: Zheng, Niannian; Luan, Xiaoli; Liu, Fei
• Is Part Of: Journal of process control. Volume 109(2022)
• Page Start: 32
• Page End: 43
• Publication Date: 2022-01
• Subjects: Dynamic controlled principal component analysis -- Autoregressive with exogenous model -- Dynamic causality -- Process pattern -- Model predictive control; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2021.11.010
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: Many latent variable modeling methods have been developed to extract the running pattern of industrial process, but the dynamic causality between control inputs and pattern is unmodeled or implicit, and thus the direct pattern control is impracticable. To overcome this limitation, the dynamic controlled pattern extraction and pattern-based model predictive control (MPC) are investigated in this paper. Firstly, a novel dynamic controlled principal component analysis (DCPCA) is proposed to extract the pattern of the industrial process from measured variables. Specially, the autoregressive with exogenous (ARX) model is introduced to characterize the dynamic relationships of the process. By maximizing the covariance of the ARX prediction and the spatial projection, the process running information can be captured by the pattern maximally with the minimum dimensions, and also benefiting from this way, both the free motions and the dynamic causality between the control inputs and pattern is established explicitly. Then, a well-designed robust tube-based MPC is implemented for optimal pattern tracking. Finally, case studies illustrate the effectiveness and advantages of the proposed DCPCA algorithm and pattern-based MPC strategy. Highlights: The novel scheme of dynamic controlled pattern extraction and pattern-based process control is proposed. This scheme models, monitors and controls the process through pattern which concentrates the most process running information, and fully captures the process dynamic. A DCPCA method is originally proposed to extract the desired pattern from measured process variables. DCPCA explicitly establishes the dynamic causality between the control inputs and the pattern for the first time, which overcomes the common limitation of the existing latent variable modeling methods and provides the practicability for pattern control. The process dynamics are described by the vector ARX model about pattern motion, based on which the robust tube-based MPC is used to realize the optimal pattern tracking. The pattern-based process monitoring is implemented successfully in the benchmark of Tennessee Eastman process.