Scheduling quasi-min-max model predictive control.

摘要

Most chemical operations are nonlinear with input/output constraints and generally experience varying operating conditions such as batch operations and start-up and shut-down processes in continuous operations. Thus, the research goal of this thesis was to develop an advanced model predictive control strategy for these variable features of chemical operations. Model predictive control strategy was used because of its ability to handle nonlinearities and input/output constraints and its inherent optimization property.; Linear parameter varying systems were first studied because of its potential to handle time varying dynamics. It is assumed that scheduling parameters can be measured or estimated in real time; thus, the current linear model is known without uncertainties. However, in the future, the linear models are uncertain but known to belong to a polytope constructed by a family of linear models. In the designed MPC algorithm, infinite horizon predictions are made by calculating an infinite number of control actions. The first stage prediction plus the upper bound of the future predictions (“quasi worst case”) are minimized; therefore, the algorithm is referred to as “quasi-min-max”. The algorithm is also referred to as “scheduling” because the calculated control actions depend on the current linear model. Closed-loop stability is guaranteed when the algorithm is implemented in a receding horizon fashion.; After linear parameter varying systems are addressed, nonlinear chemical operations were studied. Nonlinear systems are approximated by a combination of a current linear model and a linear parameter varying model. Current nonlinear dynamics are approximated by the current linear model, while the future nonlinear behaviors are unknown but vary with a range covered by the linear parameter varying model. This approximation technique considers the nonlinear dynamics and can handle nonlinear transition processes with different operating conditions as well. From the application of the scheduling quasi-min-max algorithm on a jacketed styrene polymerization reactor, it was found that combining the current linear model and the linear parameter varying model reduces conservatism regarding input constraints. Furthermore, constructing the linear parameter varying model by choosing proper operating conditions improves control performance, feasibility, and computational properties.