Feasibility of constrained generalized predictive control within invariant sets framework

Abstract

Feasibility issues of generalized predictive control structures are analysed in case of constraints on the input, output or other auxiliary variables with linear dependence on the system variables. The constraints are formulated as sets of linear equality or inequality relations on the input variables that would represent the optimisation domain for a quadratic cost function. Looking as the main goal at the prediction of the feasibility of this optimisation problem, sufficient conditions are presented in this direction. The main contribution is the description of the optimisation domain as an invariant set with respect to the predictive control law dynamics. The main advantage is that all the results are based on off-line calculations offering qualitative information prior to the effective implementation.