Sintonia QR do Regulador Linear Quadrático LQR Discreto e Programação Dinâmica Aproximada baseada em Ação-Estado para Aplicações Online do Projeto de Sistemas de Controle Ótimo
DOI:
https://doi.org/10.5540/tcam.2024.025.e01686Keywords:
Programação Dinâmica, Controle Ótimo, HDP, Q-Function, ADHDP, Sistemas Multivariáveis, Convergência, DLQR.Abstract
Due to increasing technological development and the consequent industrial applications, new methods for control design and Reinforcement Learning have been developed, not only to solve new control problems, but also to improve the performance of controllers already implemented in real-world systems. Reinforcement Learning and Discrete Linear Quadratic Regulator approaches are connected by Adaptive Dynamic Programming methods. These paradigms are oriented towards the design of optimal controllers in multivariable systems. For the case of the Discrete Linear Quadratic Regulator, AD-HDP, Reinforcement Learning, Iteration Policy and Iteration Value, a method and an algorithm are developed and implemented for online control design. Based on the selection of the Q
and R weighting matrices , a method to tune Discrete Linear Quadratic Regulator controllers is also presented, this method provides guidelines for constructing heuristics for the selection of weighting matrices, aspects of convergence related to the weighting matrices variations are investigated. For a third-order multivariable dynamic system, the proposed algorithm and tuning heuristics are evaluated by the ability to establish the optimal control policy.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2024 Trends in Computational and Applied Mathematics
This work is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.
Copyright
Authors of articles published in the journal Trends in Computational and Applied Mathematics retain the copyright of their work. The journal uses Creative Commons Attribution (CC-BY) in published articles. The authors grant the TCAM journal the right to first publish the article.
Intellectual Property and Terms of Use
The content of the articles is the exclusive responsibility of the authors. The journal uses Creative Commons Attribution (CC-BY) in published articles. This license allows published articles to be reused without permission for any purpose as long as the original work is correctly cited.
The journal encourages Authors to self-archive their accepted manuscripts, publishing them on personal blogs, institutional repositories, and social media, as long as the full citation is included in the journal's website version.