Self-learning model predictive control based on the sequence of controllable sets

This paper presents a stabilizing model predictive control (MPC) algorithm based on the off-line computation of sequence of 1-step controllable sets for linear parameter varying systems and a parameter learning technique that improves its performance. The presented MPC algorithm guarantees non-monotone convergence towards a suitably chosen terminal set regardless of the system parameters, while the parameter learning technique is used to improve performance. The benefits of such an approach are shown in application of the proposed algorithm on a laboratory rotary inverted pendulum system.