A Two-phase Objective Function for a Constrained MPC and Its Application to a Grinding Plant

In the paper a two-phase MPC controller is described that uses a non-quadratic terminal cost. The terminal cost is computed with respect to a predefined control strategy, and it can therefore be treated as the second phase of the method. Because the problem of optimizing the MPC objective is not a quadratic programming problem, medium-scale, gradient-based optimization methods are used for the MPC implementation. The method is efficient for problems which have a lot of soft constraints, because it avoids the introduction of slack variables. The stability of the MPC is established in the paper, and the method is compared with an unconstrained MPC in the grinding plant optimization problem.