Dynamic output-feedback controllers for discrete-time linear systems with markovian jumping parameters, imperfect mode observation and additive noise perturbations

• Ano de defesa: 2020
• Autor(a) principal: Mercado, Pablo Ivan Zamora
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Controle Estocástico; Controle Otimo; Markov jump linear systems; Optimal Control; Optimization; Otimização; Sistemas Lineares com saltos Markovianos; Stochastic Control
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-10092020-161726/

A class of stationary dynamic output-feedback controllers for discrete-time Markovian Jumping Linear Systems(MJLS) considering the minimization of a long run average cost is studied. The Markov chain that governs the parameters is not required to be ergodic, and it is allowed to be periodic and contain transient states / non-communicating classes, which enlarges the class of system, e.g. now comprising periodic systems as a sub-class. A compact optmization formulation is obtained for the mode-independent/detector-based controller of partial/full order, allowing one to explore the complexity and consequently obtain the best implementable performance for an application. We develop a two stage feasibility - optimization algorithm using the operator-based approach. We present a set of numerical examples in the context of random non-trivial systems that are subject to jumps in order to represent our results and compare the performance with a classical genetic algorithm, resulting in a clear advantage for our proposed algorithm.