Predição de Indicadores de Qualidade de Energia Utilizando Técnicas de Processamento de Dados e Redes Neurais

Abstract

In Brazil, one way to evaluate the service performance of electric power distribution utility is through the monitoring of electricity indices of Power Quality: SAIDI (System Average Interruption Duration Index) and SAIFI (System Average Interruption Frequency Index). They are regulated by the Brazilian National Electric Energy Agency (ANEEL), which is responsible for the application of financial penalties in case of non-compliance with regulatory limits. The estimation of these indicators provides insight into the future outlook of the company, enabling the identification of areas that have a tendency to worsen quality over time. In this work, we present a methodology for forecasting both indicators using data preprocessing, recurrent neural networks, and LSTM networks. This article uses real data from an electric utility and aims to estimate the daily SAIDI and SAIFI indicators. The results indicated that there was an improvement in the forecasting of the SAIFI, but there was no apparent benefit for the SAIDI. However, improvement in prediction allows a more appropriate allocation of maintenance teams.