Comparison of Random Forest and SARIMA Methods Optimized with Genetic Algorithm for Wind Forecasting

In this work, the Random Forest and SARIMA methods were used, both optimized through genetic algorithms, with the purpose of forecasting wind speed and, consequently, energy production. The data used were collected in $\mathbf{1 0}$ minute intervals over 55 hours and 40 minutes of measurements carried out by LIDAR campaigns in the São Marcos Bay, specifically in the Mearim region. The database was divided into $\mathbf{9 0 \%}$ for training and $\mathbf{1 0 \%}$ for testing, corresponding to 5 hours and 30 minutes of forecasting, allowing the evaluation of model performance. In the experiments, Random Forest presented a mean squared error (MSE) of 0.08, while SARIMA obtained an MSE of 1.16. Random Forest demonstrated the lowest error and highest precision, thus being recommended for forecasting data in the analyzed region.