Previsão de demanda de energia elétrica no Brasil utilizando ARIMA e redes neurais

Abstract

Accurate forecasting of electricity demand is a crucial element for the efficient and sustainable planning of the energy sector. This study focuses on a comparative analysis between the ARIMA (Autoregressive Integrated Moving Average) model and Recurrent Neural Networks (RNNs) for forecasting electricity demand in Brazil, utilizing advanced time series modeling techniques. The importance of this study is based on the growing need to optimize the use of energy resources, effectively balancing supply and demand. Accurate forecasting significantly contributes to preventing overloads and waste, which are essential factors for cost reduction and ensuring a stable supply. Furthermore, the rational use of electricity has a direct impact on environmental preservation, enabling the reduction of greenhouse gas emissions and supporting sustainability goals. In the Brazilian context, the relevance of this study is even more pronounced, considering that the country's energy matrix is one of the cleanest in the world, with a significant share of renewable sources. According to recent data, renewable sources will represent a substantial portion of Brazil’s installed electricity generation capacity in the coming years. Accurate demand forecasting is fundamental to the success of government initiatives and long-term planning in the sector, contributing to a more sustainable, economical, and equitable energy development for the Brazilian population. The primary objective of this study was to forecast electricity demand in Brazil by using and comparing two robust methods: the ARIMA model and RNNs. The specific objectives included an exploratory analysis of energy demand data, the implementation and tuning of ARIMA and RNN models, a performance comparison using the Mean Absolute Percentage Error (MAPE) metric, and the identification of the most effective method considering accuracy, complexity, and applicability to real-world data. The results obtained in this study demonstrated the effectiveness of both models in forecasting electricity demand in Brazil. The ARIMA model exhibited slightly superior accuracy, while the RNN model stood out for its computational efficiency. The comparative analysis revealed that, although ARIMA achieved higher accuracy in predictions, RNNs offer significant advantages in terms of processing time and scalability. These findings have important implications for the Brazilian energy sector, providing insights for optimizing energy resource planning and management, aligning with sustainable development goals and the country's energy transition targets.