Performance improvement in MPC with time-varying horizon via switching

A crucial design parameter in model predictive control is the choice of the prediction horizon. Procedures to estimate this parameter are off-line and computationally intensive. Moreover, a single choice is perhaps not the best option at all time instants. We resolve this issue in this paper by a time-varying horizon achieved by switching betweenmultiple model predictive controllers. The stability of such an overall controller is discussed and the improvement in performance demonstrated.