Modelagem da geração de energia hidrelétrica no Brasil com séries temporais

Abstract

The electric power sector in Brazil represents one of the fundamental pillars of the national economy and daily life, with hydroelectric generation being its main source. This study aimed to analyze, model, and forecast the monthly hydroelectric energy generation in Brazil using time series techniques, with a focus on SARIMA-class models, in order to contribute to national energy planning. Monthly data from January 1996 to December 2024 were analyzed, obtained from the Brazilian National System Operator (ONS) via the Ipeadata platform. After verifying the presence of trend and seasonality using the Cox-Stuart and Fisher tests, first-order regular differencing was applied, rendering the series stationary. Based on the analysis of autocorrelation (ACF) and partial autocorrelation (PACF) functions, candidate models were proposed. Using information criteria (AIC and BIC) and predictive metrics (MAPE and RMSE), the SARIMA(2,1,1)(0,1,1)12 model was selected. The model residuals behaved as white noise, and the forecasts for the year 2024 showed good adherence to the observed values, confirming the model’s ability to capture the temporal structure of the series. The results highlight the potential of statistical modeling to support energy planning and suggest that future approaches also consider machine learning techniques for additional accuracy gains.