Identificação de sistemas não-lineares com modelo Volterra-Kautz

• Ano de defesa: 2022
• Autor(a) principal: Serafin, Higor de Souza
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial; UTFPR
• Palavras-chave em Português: Sistemas não-lineares; Funções ortogonais; Estimativa de parâmetros; Filtros digitais (Matemática); Otimização matemática; Algoritmos; Nonlinear systems; Functions, Orthogonal; Parameter estimation; Digital filters (Mathematics); Mathematical optimization; Algorithms; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/29968

This work presents the problem of estimating the Kautz poles in the kernel expansion with Volterra-Kautz models. The optimal solution for the parameters that model the poles is still open in the literature. Thus, this work brings two optimization approaches, one using the Levenberg-Marquardt algorithm and the other using Bayesian Optimization to find the Kautz parameters. Function bases for kernels are built through a structure of digital filters. To validate the implemented algorithms, data collected from an electrically coupled system was used and, on the MATLAB® software, several experiments were performed to study the impacts of the different parameters of each method used, it was also observed how the increase of the functions in the base influence the model. As a result, the generated models reached a mean square error of the order of 10−4, while works in the literature, which used the same dataset for modeling, reached a mean square error in the order between 10−2 and 10−3.