Neural Network Implementation of Nonlinear Receding-Horizon Control

Abstract

The Receding Horizon (RH) approach is an effective way to derive control algorithms for nonlinear systems with stabilising properties also in the presence of state and control constraints. However, RH methods imply a heavy computational burden for on-line optimisation, therefore they are not suitable for the control of ‘fast’ systems, for example mechanical ones, which call for the use of short sampling periods. The aim of this paper is to show through an experimental study how a Nonlinear RH (NRH) control law can be computed off-line, and subsequently approximated by means of a neural net-work, which is effectively used for the on-line implementation. The proposed design procedure is applied to synthesise a neural NRH controller for a seesaw equipment. The experimental results reported here demonstrate the feasibility of the method.