Learning-based predictive control for MIMO systems

A learning-based approach for robust predictive control design for multi-input multi-output (MIMO) linear systems is presented. The identification stage allows to obtain multi-step ahead prediction models and to derive tight uncertainty bounds. The identified models are then used by a robust model predictive controller, that is designed for the tracking problem with stabilizing properties. The proposed algorithm is used to control the nonlinear model of a quadruple-tank process using data gathered from it. The resulting controller, suitably modified to account for the nonlinear system gain matrix, results in remarkable tracking performances.