Predição de Indicadores de Qualidade de Energia Utilizando Decomposição em Frequência e Redes Neurais Profundas

• Ano de defesa: 2022
• Autor(a) principal: Mattedi, Bruno Stinghel
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Aprendizado de Máquina; Aprendizado Profundo; Indicadores de Qualidade; Séries Temporais; Redes Neurais Convolucionais; subject.br-rjbn; Engenharia Elétrica
• Link de acesso: http://repositorio.ufes.br/handle/10/16494

With the evolution of industrial processes, the market has become increasingly competitive, WHERE any advantage can be the differential for a company to be successful or not. This precept also applies to the captive energy market, even though the consumer does not have the option of choosing his energy distributor. This is because the National Electric Energy Agency (ANEEL), with its regulatory power, has established limits for some service quality indicators, which, when exceeded by the distributors, result in fines being paid. Thus, the control of these indicators provides a great strategic advantage, so that investments can be optimized to avoid the limits being violated. In this context, this work presents an approach based on the use of neural networks for the prediction of indicators of an electric energy distributor. More specifically, the collective indicators of service quality continuity will be predicted: the System Average Interruption Duration Index (SAIDI) and the System Average Interruption Frequency Index (SAIFI). The focus is on creating predictions that perform well, enabling better management of expenses by reducing amounts paid in fines. In this work, three types of neural networks were used, namely: a shallow neural network, a Long Short-Term Memory (LSTM), and a Convolutional Neural Network (CNN) combined with LSTM (CNN+LSTM). In addition, seeking to reduce the complexity of the data, two pre-processing techniques based on the frequency decomposition of the time series were used: one based on the Wavelet Transform and the other based on the empirical decomposition of the time series. The proposed approach based on neural networks and decomposition of time series was applied to a real dataset composed of SAIDI and SAIFI indicators of an electricity distribution company. Comparing the results of the proposed methods, it was possible to observe that the decomposition based on the Wavelet Transform combined with the LSTM and CNN+LSTM networks presented better performances for the prediction of the SAIDI and SAIFI indicators, respectively.