Controlador preditivo não linear embarcado para aeropêndulo utilizando rede neural artificial.

Abstract

This work proposes a synthesis procedure of an NMPC (Nonlinear Model Predictive Control) using neural networks for an aero-pendulum. The data, obtained by simulating the closed-loop response of the system and an NMPC controller for a range of desired values to the system output, is used to train an artificial neural network. The emulated controller is used to control the plant and achieves similar reference-tracking performance. This approach significantly reduces the computational burden involved in the NMPC original approach, making it possible to run a synthetized NMPC controller on low processing capability hardware.