Explorando as tendências climáticas da velocidade do vento no Brasil (1961-2020) e propondo um modelo híbrido para previsões

Abstract

This study aimed to analyze daily wind speed data in Brazil collected from 1961 to 2020 to assess changes in two 30-year climatological normals, referred to as periods P1 and P2. The analysis was conducted through 54 meteorological stations managed by INMET, utilizing missing data filled with the bootstrap expectation maximization algorithm. The study included the application of statistical tests, such as the MannKendall test to detect linear trends and the Sen slope test to quantify trend intensity. These findings have the potential to enhance the optimization of renewable wind energy generation in Brazil. Additionally, two databases, onshore and offshore, were used to evaluate wind potential. The onshore database consisted of hourly wind speed information collected at automatic stations in 2019, located near wind parks. The offshore database was obtained from a buoy offshore, Fortaleza, located on the coast of Ceará, near an offshore wind farm projected by the company Totalenergies Petróleo & Gás Brasil. The Monte Carlo method was used to build prediction confidence intervals. During the P2 period, positive trends were observed in the Amazon basin and the semi-arid region, while negative trends were highlighted along the coast and in the Southeast region. Average wind speeds during P1 were higher than those in P2, revealing variations over the years. Comparison of the climatological normals P1 and P2 revealed significant positive trends in some regions during P2, with an increase in the average monthly variation of up to 0.85 m/s. The average slopes of significant trends were -1 and 1 ms-1 decade-1 for all analyzed periods. These results indicated high wind magnitudes, with Maximum Power Density (PD) of 778.08 Wm-2 for the (NEB) Northeast Region, 170.90 Wm-2 for the (SUB) South Region, and 401.67 Wm-2 for the Southeast Region (SUB). The hybrid approach (SARIMA+GRU) proved superior to individual algorithms in generating wind speed and offshore wind potential forecasts. The prediction gain percentage reached up to 95% compared to individual models. The studied region has a total estimated technical potential of 2766 GW, and the analysis revealed that a single park with on the NEB coast 200 turbines can meet the entire NEB region's demand, equivalent to 229% more than the region's demand. These results highlight the significant wind potential in the region, contributing to future renewable energy projects. Imputing 10% of the data was identified as the most reliable, showing low Root Mean Square Error (RMSE) values, values close to 1 for DP, and r values exceeding 0.93, indicating the effectiveness of this method in reconstructing historical wind speed data in Brazil.