Model Predictive Control of Interconnected Linear and Nonlinear Processes

Abstract

A plant-wide control strategy based on integration of linear model predictive control (LMPC) and nonlinear model predictive control (NMPC) is proposed. The design method is applicable to plants that can be decomposed according to the nonlinearity properties of the individual unit operations. The basic idea is to apply LMPC and NMPC controllers to the linear and nonlinear subsystems, respectively. A systematic procedure for performing the plant decomposition given nonlinearity information is presented. Because the subsystems are coupled via material and energy flows, a sequential solution procedure that aims to minimize the amount of unknown information in the MPC designs is developed. The plant decomposition and sequential MPC solution algorithms are applied to a large-scale styrene production flowsheet. Three controller coordination strategies are developed to handle the information-exchange problems caused by sequential MPC solution. The methods are shown to be nominally stabilizing for nonlinear plants with a certain triangular structure. A multi-rate extension for plants with time-scale separations is presented. A reaction/separation process with recycle is used to compare the different hybrid MPC approaches.