Estimação fasorial para relés de proteção numéricos usando redes neurais artificiais e o algoritmo de Levenberg-Marquardt

• Ano de defesa: 2018
• Autor(a) principal: Silva, Chrystian Dalla Lana da
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/26339/001300000v64w
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Elétrica; UFSM; Programa de Pós-Graduação em Engenharia Elétrica; Centro de Tecnologia
• Palavras-chave em Português: Algoritmo de Levenberg-Marquardt; Estimação fasorial; Redes neurais artificiais; Relés de proteção numéricos; Artificial neural networks; Levenberg-Marquardt algorithm; Numerical protective relays; Phasor estimation; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/15077

In this work, two methodologies for phasor estimation in numerical protective relays in Electrical Power Systems will be presented: one based on Artificial Neural Networks, and another based on the Levenberg-Marquardt algorithm. Firstly, a brief introduction to the mathematical concepts of some phasor estimation algorithms used for comparison with the proposed methods is presented. These algorithms consist of some Discrete Fourier Transform based methods, integral based methods, the Mimic filter, the cosine filter, the Kalman filter, and a method based on curve-fitting which was proposed by the author in another work. Next, the two proposed methods are presented in detail where all of their aspects are shown, from the mathematical equations that govern them, to the application of these equations in a structure with the purpose of real-time phasor estimation. The comparative analysis between the methods is done through the simulation of signals with varied parameters, including the components that may impair signal reading by the relay. These components are the unidirectional exponentially decaying component, also known as DC component or continuous component, harmonic components, and noises. From the simulations, six Performance Indexes are used, which quantitatively evaluate each algorithm on the criteria of response oscillation, overshoot, and convergence time, whence the best relative performances can be numerically obtained. The simulations graphical results are also shown, with the goal of providing visual aid and serve as complementation and verification for the numerical results. After the simulated cases are evaluated, an analysis of the results and characteristics of each method is made, highlighting their strengths and weaknesses.