Seleção aleatória da estrutura de modelos com auxílio da taxa de redução do erro e herança genética

• Ano de defesa: 2022
• Autor(a) principal: Fagundes, Luis Paulo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Elétrica
• Palavras-chave em Português: Identificação de sistemas; Seleção da estrutura aleatória; Taxa de redução do erro; Seleção da estrutura; NARX; MIMO; System Identification; Randomized model structure selection; Error reduction rate; Structure Selection; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA::ELETRONICA INDUSTRIAL, SISTEMAS E CONTROLES ELETRONICOS::CONTROLE DE PROCESSOS ELETRONICOS, RETROALIMENTACAO; Engenharia Elétrica; Análise estrutural (Engenharia) - Modelos matemáticos; Análise estrutural (Engenharia) - Programas de computador; Simulação por computador
• Link de acesso: https://repositorio.ufu.br/handle/123456789/35342; http://doi.org/10.14393/ufu.te.2022.391

The systems present in industry and science are commonly non-linear, which has made methods for selecting the structure of this type of system widely studied over the last thirty years. There are several methods in literature to deal with the model structure selection in system identification, although these methods have their specific benefits, they face some difficulties in selecting a parsimonious model structure. In this work, two methods based on the Randomized Model Structure Selection (RaMSS), are introduced in order to deal with the model structure selection problem. The first one, named Randomized Model Structure Selection with Error Reduction Ratio (RaMSS-ERR), uses the error reduction rate as a pre-filter in the terms analysis, improving the convergence, the second one, Randomized Model Structure Selection with Genetic Inheritance (RaMSS-EGI) uses a genetic inheritance in order to get faster convergence. The methods were applied to reference models commonly used in the literature of single input and single output, and in models with multiple input and multiple outputs identifying it with more parsimony than some methods present in the literature. They were also applied in the identification of systems with a large candidate regressor set, in the identification of a continuous stirred-tank reactor, and in a C3/C4 column system. The results show that the proposed methods may be used to identify both linear and nonlinear model structures with a reduced number of iterations, computational time, and number of explored models.