Forecasting Total Hourly Electricity Consumption in Brazil Through Complex Seasonality Methods

Accurate electricity demand forecasting, especially in the short run, is critical for the optimal management of power systems. However, high-frequency time series usually contain unique stylized facts, such as multiple seasonal patterns, which imposes an extra challenge in accurate estimation and forecasting. Aiming to contribute to the reliable operation and planning of the Brazilian National Interlinked System (SIN), this work compares the accuracy of several statistical methods when forecasting multiple series of hourly electricity consumption in Brazil up to 7 days (168 h) ahead. Both benchmark methods and complex seasonality models, i.e., methods that consider multiple seasonal patterns of the underlying series, are assessed. We provide robust evidence towards the adherence of two selected complex seasonality models for the Brazilian total electricity consumption in the short term. Results and implications in terms of decision-making are further discussed.