Rede neural NARX aplicada ao amortecimento de oscilações de potência

• Ano de defesa: 2020
• Autor(a) principal: Carbonera, Luis Felipe Bianchi
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• 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: Regulador de tensão; Estabilizadores de sistemas de potência; Redes neurais; Algoritmo genético; Automatic voltage regulator; Power system stabilizers; Neural network; Genetic algorithm; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/22693

The dynamics of power systems with time-varying and non-linear elements is a complex matter since it involves continuous adjustments among every part for a suitable and efficient performance. Small disturbances in the load variation procedure also routinely occur. Consequently, the controller parameters must be adjusted to the variable conditions. The non-linear autoregressive model with exogenous input (NARX) neural network (NN) has been used in many non-linear dynamic systems. This paper explores the NARX combined with a multiobjective optimization by using genetic algorithms (GAs) to damp local and interarea oscillation modes. The NN model is trained by using a historical database determined by the GA for several load levels. Subsequently, the model can change the stabilizer parameters in real time after the learning phase. This study is used to tune a power system stabilizer (PSS) in a two-area four-machine system. This study is to adjust for the IEEE PSS4b in a system of four machines and two areas. The results of the simulations in the case study demonstrated that the GA-NARX-PSS had the best effects in dampening the oscillations resulting from the disturbances compared to other stabilizer adjustment techniques found in the literature. It was observed that the GA-NARX-PSS damped as oscillations caused by a short circuit with a longer duration than that supported by the PSS tuned with both AG optimization and the whale optimization algorithm (WOA).