Forecasting electricity consumption in the southeast region: Application of ARIMA and LSTM models

Abstract

The objective of this study is to forecast electricity consumption in the Southeast region of Brazil from April 2023 to March 2024 using ARIMA models and LSTM neural networks. Using monthly data from 2002 to 2023, the research compares the models based on the error metrics RMSE, EAM, and MAPE. The ARIMA model captures seasonal and linear patterns in the short term, while the LSTM model excels in predicting nonlinear and long-term trends. The combination of the two approaches has shown promise in improving forecasting accuracy, suggesting that policymakers can have reasonable expectations about future projections. This research contributes methodologically by exploring complementary approaches, and a practical contribution to efficient energy planning, based on more assertive short-term forecasts, which allow for the safe operation of the electricity system, reducing the risk of overloads and interruptions in energy supply.