A systematic method to handle unfeasibility in constrained predictive control

A new systematic method is proposed in this paper to handle infeasibility problem appearing in constrained model predictive control (MPC). When there are constraints on both the inputs and outputs, it is possible that the constrained MPC is unable to find a feasible solution to satisfy all of the constraints when a disturbance pushes the process outside the usual operating region. To recover the feasibility, a new linear programming based algorithm is presented to relax the constraints on outputs and to minimize the change of constraint limits. It has been successfully applied to the Shell benchmark problem