Desacoplamento de Sistemas Multivariáveis via Aprendizado Participativo Evolutivo
Authors
Lincoln C. Santos
Programa de Pós-Graduação em Engenharia Elétrica (PPGEL), Associação ampla entre CEFET-MG e UFSJ
Lucas S. Oliveira
Departamento de Engenharia Mecatrônica, CEFET-MG, Divinópolis/MG
Valter J. S. Leite
Programa de Pós-Graduação em Engenharia Elétrica (PPGEL), Associação ampla entre CEFET-MG e UFSJ; Departamento de Engenharia Mecatrônica, CEFET-MG, Divinópolis/MG
Decoupling by learning, multivariable systems, evolving systems
Abstract
This work presents a new methodology for decoupling multivariable systems, based on the evolutive participative algorithm. Decoupling is achieved by adding compensation signals to the control signals produced by controllers designed to operate in single-input single-output loops. Unlike the analytical approaches available in the literature, the model of the controlled system is dispensable, with its dynamics learned during operation. As a result, the approach can compensate for parametric variations or even structural changes in the controlled systems. These characteristics are not addressed by techniques in the literature that rely on knowledge of the process model. The proposed methodology is applied to a boiler model with 4 inputs and 4 outputs, yielding superior results compared to other evaluated methodologies, particularly excelling in scenarios where uncertainties (parametric or structural) affect the controlled system.
