We consider the problem of reducing interaction in the closed-loop response of model predictive control (MPC). Interaction in MPC may be caused by diagonal weighting of inputs in the MPC cost function that are not diagonally related to the outputs. If instead of weighting the plant inputs a suitable decoupled input signal is used in the MPC cost function, then a significant reduction in cross coupling can occur. In the case where the plant has a static interaction matrix, complete decoupling occurs. Simulation examples show that the procedure can be implemented via a simple modification to standard MPC algorithms, and is applicable to ill-conditioned and non-minimum phase plants.
17th World Congress of the International Federation of Automatic Control. Proceedings of the 17th World Congress of the International Federation of Automatic Control (Seoul, Korea 6-11 July, 2008) p. 9940-9945