Closed-loop identification and model predictive control of extractive dividing-wall column

Abstract

Extractive dividing-wall column (EDWC) is a promising thermally-coupled system for separating multiple azeotrope or close-boiling mixtures; however, its intensified structure with relatively small physical space and strong interactions makes model predictive control (MPC) an attractive scheme for EDWC operation. The performance of MPC strongly depends on the accuracy of the dynamic model. The presence of recycled entrainer stream in EDWC makes it impossible to identify the dynamic model of EDWC using open-loop experiments. In the present study, a proportional-integral (PI) control strategy based on temperature difference was first studied for EDWC operation. Subsequently, a systematic closed-loop identification experiment was implemented to identify a dynamic model, and an MPC was designed using the identified model. The results indicate that the accuracy of the identified model is sufficient to design MPC controller for EDWC operation. Both direct and indirect manipulation of vapor split ratio were considered in MPC. The transient responses of product purities indicate that MPC is better than PI for EDWC operation for rejecting disturbances in feed flow rate and composition, even though MPC shows slow dynamic responses compared to PI control for rejecting feed composition disturbances.