Invariant sliding domains for constrained linear receding horizon tracking control

Abstract

A classical approach for guaranteeing persistent feasibility of model predictive controllers during setpoint changes adds an artificial reference variable, whereby allowing for reference offset at a cost specified by an additional term in the cost function. Typically, the classical approach employs a linear quadratic regulator parameterized by the artificial reference as a terminal control law and hence requires invariant set computations in an augmented state/reference space. This paper develops a receding horizon sliding control technique for constrained linear setpoint tracking. By exploiting the flatness property of sliding hyperplanes, the artificial reference can be eliminated from the control scheme and the terminal invariant set is contained in the original dimensions of the state space only. The proposed dual mode receding horizon control design approach is proven to maintain persistent feasibility and stability.