Methodology for Designing and Comparing Robust Linear versus Gain-Scheduled Model Predictive Controllers

Abstract

A methodology is proposed to design a robust Gain-Scheduled Model Predictive Control (MPC) strategy and to quantify the relative advantages of this controller versus a Linear MPC strategy. For the purpose of analysis and controller design, the process is represented by a nonlinear state-affine model identified from input−output data. This model can be split in linear and nonlinear terms where the linear part is used for controller design and the nonlinear part is accounted for as model uncertainty. Then, robust stability and robust performance tests are formulated based on linear matrix inequalities where the manipulated variables weight of the controllers is tuned to maximize performance. The uncertainty bounds used for the robustness tests are obtained in an iterative fashion by using the frequency response of the manipulated variable with respect to the feedback error. The control strategy performance is quantified by the ratio between the error norm and the disturbance norm. Finally, a case study involving a multiple-input−multiple-output bioreactor is presented. The study is able to predict for which range of operation the Gain-Scheduling MPC surpasses the performance of the Linear MPC.