Introducing robustness in model predictive control with multiple modelsand switching

Model predictive control is model-based. Therefore, the procedure is inherently not robust to modelling uncertainties.Further, a crucial design parameter is the prediction horizon. Only offline procedures to estimate an upper bound of the optimalvalue of this parameter are available. These procedures are computationally intensive andmodel-based. Besides, a single choice ofthis horizon is perhaps not the best option at all time instants. This is especially true when the control objective is to track desiredtrajectories. In this paper,we resolve the issue by a time-varying horizon achieved by switching betweenmultiplemodel-predictivecontrollers. The stability of the overall system is discussed. In addition, an introduction of multiple models to handle modellinguncertainties makes the overall system robust. The improvement in performance is demonstrated through several examples.