A feedback linearization technique by using neural networks: Application to bioprocess control¹

Article type: Research Article

Authors: Boutalis, Y.S.^{; *} | Kosmidou, O.I.

Affiliations: Automatic Control Systems Laboratory, Department of Electrical & Computer Engineering, Democritus University of Thrace, GR-67100 Xanthi, Greece

Correspondence: [*] Corresponding author. Tel.: +30 25410 79504; Fax: +30 25410 79512; E-mail: ybout@ee.duth.gr.

Note: [1] This work was partially supported under grant PRENED 99, KE 881.

Abstract: In this paper an approach for the problem of feedback linearization of an uncertain system using Artificial Neural Networks (ANN) is proposed. A bioprocess system is considered as an example of a highly nonlinear and time varying uncertain system. The neural network is used as a compensator to the effect of the model uncertainties, which appear in the linearizing control law. The updating of the neural weights is carried out on-line using a conventional back propagation scheme. The stability of the tracking error system as well as the decay of the tracking error to zero are guaranteed by defining an appropriate error signal which has to be minimized by the trained ANN. It is also experimentally shown that the proposed neural network compensator performs well even in the case where the full state vector of the system is not available for measurement.

DOI: 10.3233/JCM-2006-65-602

Journal: Journal of Computational Methods in Sciences and Engineering, vol. 6, no. 5-6, pp. 269-282, 2006