Localização de faltas em redes de distribuição de energia elétrica: uma metodologia utilizando redes neurais artificiais aliada a dados de reclamação de clientes de uma concessionária

• Ano de defesa: 2018
• Autor(a) principal: Souza, Fabrício Augusto de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Engenharia Elétrica; UTFPR
• Palavras-chave em Português: Falhas de energia elétrica; Redes neurais (Computação); Energia elétrica - Distribuição; Electric power failures; Neural networks (Computer science); Electric power distribution; CNPQ::ENGENHARIAS; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/4543

The regulation of the electricity sector in recent decades, coupled with an increase in the level of information provided by consumers in the sector, has demanded from electricity distribution concessionaires greater efficiency of their systems and higher levels of quality in the supply of energy to their customers. Therefore, it has become a priority for the utilitys the aspects related to the duration of interruptions in the energy supply. In this scenario, in order to improve the failure localization efficiency in the electric distribution system, this works presents a method that proposes the localization of these power outage in rural areas from current and voltage data of the feeder output in which the failure happened, using Artificial Neural Networks, allied to data regarding calls of customer complaints from the affected region. The methodology was developed through simulations in the ATPDraw software, to generate the input data of the Artificial Neural Networks, developed in Matlab® software in two stages, being the first one is responsible for the classification of faults and the second one by its locations in the feeder of simulated distribution system. The use of two types of Artificial Neural Networks, the Multilayered Perceptron and the Kohonen maps, one for each step, assure a differential for the technique. The results presented point out the potential of the proposed methodology.