Modelagem do consumo de energia elétrica na região Sudeste com séries temporais e redes neurais artificiais

• Autor(a) principal: Silva, Mateus de Freitas
• Data de Publicação: 2022
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/36042
• topic: Sarima; Autoregressive; Autorregressivo; Forecast; LSTM; RMSE; Previsão; CNPQ::CIENCIAS EXATAS E DA TERRA

The electricity issue, with its complex production, distribution and consumption process, has always had a lot of visibility, for environmental issues, planning failures, blackouts and economic viability. In Brazil, the Southeast region is the main consumer of energy, and has a diversified production matrix. For this reason, this research approaches the monthly time series of electricity consumption in the region, in order to find a model capable of making predictions for this variable, and absorb its behavior over time, its trend and its seasonality. Time series techniques were used using the Box-Jenkins methodology, with autorregressive seasonal models of type SARIMA, and models of Recurrent Neural Networks of type LSTM (Long Short Term Memory). The criterion used to make an analytical decision on which model is best was the Root Mean Squared Error, called by RMSE, so that the chosen model, in all cases, was the one with the lowest value for the RMSE. Based on this, it was found that, for short-term predictions, with data feedback, the Neural Networks model proved to be quite effective. On the other hand, if the intention is to develop long-term forecasts, projecting consumption for the next 12 months, at once, the SARIMA model is the most appropriate, avoiding very sharp errors in certain months.