TIME SERIES-BASED APPROACHES FOR IMPROVING WIND POWER GENERATION FORECAST ACCURACY

This study provides a detailed analysis and prediction of power generation at wind farms in Germany using Lasso, LightGBM, and CatBoost machine learning models. Feature Engineering was used on the data, which allowed the extraction of more detailed data, which was used to improve the quality of the models. Through Extensive Data Analysis (EDA), the authors identify and develop lagged and moving features from the energy production time series, under the assumption that accurate predictions can significantly improve the stability of energy systems, especially in the context of increasing dependence on renewable energy sources. The performance of each model is evaluated based on the Mean Absolute Error(MAE), Mean Squared Error(MSE), and Root Mean Squared Error(RMSE) metrics, with CatBoost exhibiting the highest accuracy. In conclude, pointing to opportunities for further research aimed at optimizing these models and adapting them to other regions, emphasizing the comprehensive and long-term potential of this study in the context of energy field.