Nonlinear model predictive control of LDPE autoclave reactors

Abstract

Polymerization reactors are characterized by highly nonlinear dynamics, multiple operating regions, and significant interaction among the process variables, and are therefore, usually difficult to control efficiently using conventional linear process control strategies. It is generally accepted that nonlinear control strategies are required to adequately handle such processes. In this work, we develop, implement, and evaluate via simulation a nonlinear model predictive control (NMPC) formulation for the control of two classes of commercially relevant low-density polyethylene (LDPE) autoclave reactors, namely, the single, and multi-zone multi-feed LDPE autoclave reactors. Mathematical models based on rigorous, first-principles mechanistic modeling of the underlying reaction kinetics, previously developed by our research group, were extended to describe the dynamic behaviour of the two LDPE autoclave reactors. Unscented Kalman filtering (UKF) based state estimation, not commonly used in chemical engineering applications, was implemented and found to perform quite well. The performance of the proposed NMPC formulation was investigated through a select number of simulation cases on the mathematical ‘plant’ models. The resulting closed-loop NMPC performance was compared with performance obtained with conventional linear model predictive control (LMPC) and proportional-integral-derivative (PID) controllers. The results of the present study indicate that the closed-loop disturbance rejection and tracking performance delivered by the NMPC algorithm is a significant improvement over the aforementioned controllers.