Explorando modelos híbridos para a previsão de vento: uma análise multirregional

Abstract

Wind energy is one of the leading low-carbon technologies for electricity generation, and Brazil has great potential for its development, especially in the Northeast and South regions. Recent studies show an interesting complementarity between wind and hydropower, as during drought periods, wind generation tends to be higher due to increased wind speeds, and vice versa. In 2022, wind energy became the second most used source in the Brazilian electricity matrix, accounting for 11.8% (or 25.6 GW) of the installed capacity, behind only hydropower, which represents 61.9%. The National Energy Plan estimates that by 2050, Brazil will have an installed wind energy capacity of 110 GW, further consolidating the relevance of this source. However, the intermittent nature of wind energy presents challenges for its efficient integration into the national electricity grid. To address these challenges, the ability to predict and analyze generation trends is crucial, allowing greater precision in the operation of the National Interconnected System (SIN) and enabling more informed decisions regarding energy supply and reservoir management. Among the main factors that influence the generation of a wind farm are wind direction and speed. This work aims to study different models to predict wind energy generation in the state of Ceará and compare them with other Brazilian states, such as Bahia, Piauí, Rio Grande do Norte, Rio Grande do Sul, and Santa Catarina. Models such as ARIMA, CNN, LSTM, CNN-LSTM, CNN-ALSTM, ARIMA-CNN, and ARIMA-LSTM were evaluated, and the results highlighted the superiority of the parallel hybrid models, especially ARIMA-CNN and ARIMA-LSTM. The forecasts made by the ARIMA-CNN model for the state of Rio Grande do Sul were used to estimate the energy production of the Honda Energy Xangri-lá wind farm, using two approaches: one based on power intervals and the other dependent on wind direction. Both approaches proved to be promising, particularly when working with average production over intervals longer than a week.