Stabilizing predictive control of NARX models

In this paper a predictive control algorithm for nonlinear discrete-time systems is presented. Starting from a state-space model, conditions for the asymptotic tracking of constant reference signals in a neighbourhood of a given equilibrium are first derived. Then it is shown that the system under control can be locally described in terms of a suitable NARX (Nonlinear ARX) model, which can be identified by means of well established techniques. For the NARX model, a receding-horizon predictive control algorithm is proposed which guarantees local stability and robust asymptotic tracking in the neighbourhood of the equilibrium.