Adaptive Tube-Enhanced Multi-Stage Nonlinear Model Predictive Control

Abstract A robust adaptive controller for nonlinear plants with parametric uncertainties, additive disturbances, and state estimation errors based on the tube-enhanced multi-stage (TEMS) nonlinear model predictive (NMPC) framework is proposed. In TEMS NMPC, primary multi-stage NMPC is used to achieve robustness against the uncertainties which have a large effect on the evolution of the state of the plant, and ancillary multi-stage NMPC is used to track the predictions of the primary controller to counteract the effect of the small uncertainties. We propose updating, at each time step, the uncertainty set considered by the scenario trees of the primary and ancillary controllers with a tighter non-falsified uncertainty set, which results from solving a guaranteed parameter estimation (GPE) optimization problem. This produces significant performance improvements over the non-adaptive implementation as will be shown on the Williams-Otto continuous stirred tank reactor (CSTR) case study.