Data-driven Nonlinear MPC using Dynamic Response Surface Methodology

Abstract

Abstract For many complex processes, it is desirable to use a nonlinear model in the MPC design, and the recently proposed Dynamic Response Surface Methodology (DRSM) is capable of accurately modeling nonlinear continuous processes over semi-infinite time horizons. We exploit the DRSM to identify nonlinear data-driven dynamic models that are used in an NMPC. We demonstrate the ability and effectiveness of the DRSM data-driven model to be used as the prediction model for a nonlinear MPC regulator. This DRSM model is efficiently used to solve a non-equally-spaced finite-horizon optimal control problem so that the number of decision variables is reduced. The proposed DRSM-based NMPC is tested on a representative nonlinear process, an isothermal CSTR in which a second-order irreversible reaction is taking place. It is shown that the obtained quadratic data-driven model accurately represents the open-loop process dynamics and that DRSM-based NMPC is an effective data-driven implementation of nonlinear MPC.