412235 Design of Decoupling Model-Predictive Control for Multivariable Systems

Monday, November 9, 2015
Exhibit Hall 1 (Salt Palace Convention Center)
Jyh-Cheng Jeng, Department of Chemical Engineering and Biotechnology, National Taipei University of Technology, Taipei, Taiwan

Most industrial processes are multivariable systems which consist of several measurement and control signals. Multivariable systems often present complicated couplings between controlled and manipulated variables so that process interactions occur. The degree of interactions increases as a result of the high demand on product quality, the attempt on process intensification and the required energy integration. A multivariable system with strong interactions between the channels can be much more difficult to control, and the decentralized control often cannot provide satisfactory decoupling performance. In this case, the decoupling control by designing a decoupler to minimize the process interactions is advised. However, the ideal decouplers are not always physically realizable. Furthermore, the main problem of this decoupling approach is the fact that the complexity of decoupler elements and apparent decoupled processes increases for high-dimensional systems, which requires model reductions.

Model predictive control (MPC) has become the method of choice in several process industries to solve difficult multivariable control problems that include inequality constraints. However, the conventional MPC design does not explicitly address the process interactions so that satisfactory decoupling performance is not necessarily attained. This study propose a novel MPC design approach that enables achieving the condition of decoupling control without adding any decouplers. To accomplish this, each control signal is conceptually decomposed into several constituent components for MPC calculation. One of the components is used to drive a controlled variable to its set-point, and the other components are used to eliminate the effects from other manipulated variables on this controlled variable. An additional term is then included in the MPC cost function to penalize the incomplete compensation of process interactions. By tuning the weighting matrices, it is possible to deal with the trade-off among the decoupling performance, the tracking performance, and the control efforts. When a large weight is given to the incomplete decoupling, the controller provides effective decoupling so that a set-point change for one controlled variable has no (or minimal) effect on the other controlled variables.

Extended Abstract: File Not Uploaded
See more of this Session: Interactive Session: Systems and Process Control
See more of this Group/Topical: Computing and Systems Technology Division