Forecasting of Wind Power Generation Cristalândia wind farm using Tree-Based Machine Learning Approaches

Abstract

Wind power has gained increasing attention as a rapidly growing source of sustainable electricity generation. As variable renewable energy, however, its reliability, stability, and efficiency depend on factors such as wind speeds, air density, and turbine characteristics. As a result, an effective energy management strategy requires the accurate forecasting of wind power generation. Machine learning approaches have been applied to forecasting wind power generation, but their proper fine-tuning is still not fully understood. In this work, we trained using 5-fold cross-validation and fine-tuned using a GridSearch tree-based machine learning models, namely, Extreme Gradient Boosting (XGBoost) and Random Forest, for the forecasting of wind power generation. We evaluated XGBoost and Random Forest using data from the Cristalândia wind farm in BrumadoBA. The results suggest that tree-based models can accurately forecast wind power generation. Since they are relatively simple and easy to train when compared to machine learning models based on neural architectures, tree-based models are competitive approaches to forecast wind power generation.