Data-Driven Control of Highly Interactive Systems using 3DoF Model-On-Demand MPC: Application to a MIMO CSTR

This paper presents a Three-Degree-of-Freedom Model Predictive Control (3DoF MPC) framework based on Multi-Input-Multi-Output (MIMO) “Model-on-Demand” (MoD) estimation. MoD is a data-centric weighted regression algorithm that generates local models over adaptively varying neighborhoods of changing operating conditions. The 3DoF formulation enables individualized tuning of parameters relating to setpoint tracking and measured and unmeasured disturbance rejection. Online estimation of system dynamics using MIMO MoD and augmentation with the 3DoF MPC structure allows the generation of control laws based on efficient locally linear approximations of system nonlinearities. This paper evaluates the framework through a case study involving a nonlinear MIMO Continuous Stirred Tank Reactor (CSTR) model. The MIMO CSTR system is highly interactive, making data-driven estimation and control notably more challenging than its SISO counterpart. The generation of an informative database using modified “zippered” multisines is presented. The paper concludes with a case study demonstrating the effectiveness of 3DoF MoD MPC in achieving constrained MIMO control of reactor concentration and temperature in the presence of disturbances through a flexible and intuitive approach.