Previsão de consumo de energia elétrica no brasil: uma comparação entre métodos

Abstract

The prediction of electricity consumption plays a crucial role in the effective management of energy resources. This study investigates electricity consumption forecasting methods applied in the Brazilian context. Four widely recognized forecasting techniques were explored: Exponential Smoothing, ARIMA (Autoregressive Integrated Moving Average), Support Vector Machines (SVM), and Artificial Neural Networks (ANN). The analysis began with the decomposition of the time series, allowing for the identification of trends, seasonality, and random variations in the data. The Holt-Winters Exponential Smoothing and the ARIMA model were used to model the underlying temporal aspects. The Support Vector Machines (SVM) technique was enhanced through parameter optimization, resulting in highly accurate and reliable forecasts. Although it required significant computational efforts, SVM demonstrated remarkable capability in capturing complex relationships. ANN was investigated and emerged as the most effective approach, providing exceptionally precise and robust predictions. This method stood out in capturing complex and sequential patterns, making it the preferred choice for forecasting electricity consumption in Brazil. However, the final choice of the model should consider not only evaluation metrics such as MSE (Mean Squared Error), MAE (Mean Absolute Error), and RMSE (Root Mean-Squared Error) but also adaptation to specific data and computational demands.