ABSTRACT

This chapter looks at intuitive and relatively simple ways of improving robustness. An MPC problem is deemed feasible if it is possible, with the d.o.f. available, to satisfy all the constraints. One of the difficulties with establishing robustness results when constraints are present is that the answer is directly linked to feasibility. Predictive control gives a nonlinear control law when constraints are active and this implies that traditional linear robustness analysis/design cannot be implied. The benefits of the Youla parameterisation and formal robust design can still be applied, to some extent, during constraint handling and moreover it is a relatively simple approach to implement. However, there are no guarantees of recursive feasibility. Membership of an appropriate invariant set is equivalent to testing for constraint satisfaction of the closed-loop predictions over an infinite horizon.