Development of a Wind Speed Forecasting Model Using Observed Data and Machine Learning Approaches

Considering the growing potential of artificial intelligence (AI), its application has become increasingly relevant in climate-related studies and energy assessments. In this study, the Random Forest algorithm was applied to impute missing values in time series of air temperature, wind speed, atmospheric pressure, and wind direction. The performance of the data imputation was evaluated using RMSE, MSE, and MAE metrics, as well as the Kolmogorov–Smirnov (KS) test, which supported the selection of the most appropriate exogenous variable. Subsequently, short-term wind speed forecasting was performed using the SARIMAX model, and monthly energy generation was estimated for the V80/2000, SWT-2.3-101, and S95/2100 wind turbine models. The proposed methodology was applied to data from 50 conventional meteorological stations of the National Institute of Meteorology (INMET) located in Northeast Brazil. The results indicate that the gap-filling procedure was effective, particularly for wind speed and mean air temperature. Moreover, the SARIMAX model demonstrated good forecasting performance at most of the analyzed stations. Overall, the findings suggest that the majority of the locations analyzed present favorable conditions for wind-based electricity generation.