Adaptive Robust MPC: Combining Robustness with Online Performance Enhancement

Abstract

Model Predictive Control (MPC) is a well-known control technique, which uses a system model to perform an explicit numerical optimization of future performance subject to constraints on system states and control inputs. In this context an inaccurate model can result in unreliable predictions and controller performance that is far from optimal, and consequently up to 80% of the overall effort of implementing MPC is spent on obtaining an adequate model [8]. Moreover, although model uncertainty can be accounted for using robust MPC techniques, the degree of uncertainty in the system model and unknown disturbances crucially affect the bounds on the achievable performance of a MPC strategy [3].