Randomized MPC with Model Uncertainties Based on LSCR Approach

This paper deals with a probabilistic approach for online system identification in MPC models with parameter uncertainties. The problem is to construct robust control in probabilistic sense for all possible values from some predefined set. In that case it is a good way to shrink the set over time in order to get better sets. The proposed algorithm is based on the modified LSCR (Leave-out Sign-dominant Correlation Regions) and applied to state space model used in MPC. In the paper we address a convergence of the algorithm over time, which required significant modifications comparing to our previous work. Practical part demonstrates in more detail with Matlab and CVX how to obtain smaller control for a nonminimal-phase second order plant with two unknown parameters and investigates convergence of confidence regions over time.