Comparison of Wind Generation Prediction Models Using Ensemble Learning and Deep Learning Techniques

The prediction of wind energy generation is crucial for optimizing the integration of renewable sources into the electrical grid, improving planning, and reducing uncertainty in energy production. While statistical methods for wind energy prediction have been traditionally used, artificial intelligence techniques offer new opportunities to enhance accuracy. This work focuses on comparing wind energy generation prediction models using various artificial intelligence techniques to identify the most efficient and accurate model for predicting energy generated by wind farms. Four models have been used: XGBoost, Random Forest, Gradient Boosting Regressor, and Long Short-Term Memory (LSTM). The methodology includes the preprocessing of meteorological and energy production data, the implementation of the models, and the evaluation of their performance using metrics such as MAE, RMSE and $$R^2$$. The results show that the Random Forest model exhibits the best accuracy and lowest error compared to the other three models. In terms of MAE, Random Forest achieved 48 417.7, while XGBoost and Gradient Boosting Regressor reached 50 185.2 and 49 856.8, respectively. The LSTM model achieved lower performance, with an MAE of 51 920.4, highlighting the limitations of deep learning approaches under relatively small dataset conditions. This superiority can be attributed to its ability to handle data variability and capture peaks and troughs more effectively. These conclusions provide valuable insights for improving the integration of wind energy into the electrical grid, contributing to the efficiency and sustainability of the energy sector.