Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration
Abstract The inherent variability of wind and solar energy introduces fluctuations in power generation, making accurate forecasting essential for maintaining the grid’s stability. This study addresses key research gaps in wind energy forecasting, including the inability of traditional statistical mo...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-07-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-05250-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849403036791734272 |
|---|---|
| author | T. A. Rajaperumal C. Christopher Columbus |
| author_facet | T. A. Rajaperumal C. Christopher Columbus |
| author_sort | T. A. Rajaperumal |
| collection | DOAJ |
| description | Abstract The inherent variability of wind and solar energy introduces fluctuations in power generation, making accurate forecasting essential for maintaining the grid’s stability. This study addresses key research gaps in wind energy forecasting, including the inability of traditional statistical models to capture complex, nonlinear temporal patterns, the underutilization of real-time, location-specific data, the lack of comparative analyses across diverse models and datasets, and the absence of systematic model selection strategies for future forecasting. To overcome these limitations, this study applies advanced machine learning (ML) and deep learning (DL) techniques with systematic hyperparameter tuning to enhance predictive performance. Three scenarios were examined: Case 1 used a Kaggle wind turbine SCADA dataset; Case 2 employed real-time wind data from Aralvaimozhi, Tamil Nadu, India; and Case 3 focused on future forecasting using the best performing models from the earlier cases. A wide range of ML models—Random Forest (RF), Decision Trees, Linear Regression, K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), and Gradient Boosting—alongside DL models such as Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) were evaluated. Weather features, particularly wind speed, were incorporated to improve the prediction accuracy. A Stacking Ensemble model was also constructed from the top-performing models to boost robustness and forecast reliability. The performance was evaluated using the Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R2) metrics. The results showed that Random Forest excelled in Case 1, while Case 2 saw outstanding performance from Random Forest, XGBoost, and the Stacking Ensemble, achieving R2 values of 0.995, 0.997, and 0.998; MAE of 0.027, 0.035, and 0.014; MSE of 0.026, 0.014, and 0.0016; and RMSE of 0.16, 0.119, and 0.04, respectively. By directly addressing the forecasting challenges of wind energy, this study supports improved resource management, grid reliability, and operational planning. The findings highlight the effectiveness of hyperparameter-tuned ensemble models, particularly stacking ensembles, in enhancing renewable energy forecasting and advancing global sustainability goals in the future. |
| format | Article |
| id | doaj-art-124b95264d9f4a52a8d6baa2199c1af2 |
| institution | Kabale University |
| issn | 2045-2322 |
| language | English |
| publishDate | 2025-07-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-124b95264d9f4a52a8d6baa2199c1af22025-08-20T03:37:22ZengNature PortfolioScientific Reports2045-23222025-07-0115112210.1038/s41598-025-05250-3Enhanced wind power forecasting using machine learning, deep learning models and ensemble integrationT. A. Rajaperumal0C. Christopher Columbus1School of Electrical Engineering, Vellore Institute of TechnologySchool of Computer Science and Engineering, Vellore Institute of TechnologyAbstract The inherent variability of wind and solar energy introduces fluctuations in power generation, making accurate forecasting essential for maintaining the grid’s stability. This study addresses key research gaps in wind energy forecasting, including the inability of traditional statistical models to capture complex, nonlinear temporal patterns, the underutilization of real-time, location-specific data, the lack of comparative analyses across diverse models and datasets, and the absence of systematic model selection strategies for future forecasting. To overcome these limitations, this study applies advanced machine learning (ML) and deep learning (DL) techniques with systematic hyperparameter tuning to enhance predictive performance. Three scenarios were examined: Case 1 used a Kaggle wind turbine SCADA dataset; Case 2 employed real-time wind data from Aralvaimozhi, Tamil Nadu, India; and Case 3 focused on future forecasting using the best performing models from the earlier cases. A wide range of ML models—Random Forest (RF), Decision Trees, Linear Regression, K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), and Gradient Boosting—alongside DL models such as Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) were evaluated. Weather features, particularly wind speed, were incorporated to improve the prediction accuracy. A Stacking Ensemble model was also constructed from the top-performing models to boost robustness and forecast reliability. The performance was evaluated using the Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R2) metrics. The results showed that Random Forest excelled in Case 1, while Case 2 saw outstanding performance from Random Forest, XGBoost, and the Stacking Ensemble, achieving R2 values of 0.995, 0.997, and 0.998; MAE of 0.027, 0.035, and 0.014; MSE of 0.026, 0.014, and 0.0016; and RMSE of 0.16, 0.119, and 0.04, respectively. By directly addressing the forecasting challenges of wind energy, this study supports improved resource management, grid reliability, and operational planning. The findings highlight the effectiveness of hyperparameter-tuned ensemble models, particularly stacking ensembles, in enhancing renewable energy forecasting and advancing global sustainability goals in the future.https://doi.org/10.1038/s41598-025-05250-3Energy forecastRenewable energyMachine learningDeep learningSustainability |
| spellingShingle | T. A. Rajaperumal C. Christopher Columbus Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration Scientific Reports Energy forecast Renewable energy Machine learning Deep learning Sustainability |
| title | Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration |
| title_full | Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration |
| title_fullStr | Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration |
| title_full_unstemmed | Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration |
| title_short | Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration |
| title_sort | enhanced wind power forecasting using machine learning deep learning models and ensemble integration |
| topic | Energy forecast Renewable energy Machine learning Deep learning Sustainability |
| url | https://doi.org/10.1038/s41598-025-05250-3 |
| work_keys_str_mv | AT tarajaperumal enhancedwindpowerforecastingusingmachinelearningdeeplearningmodelsandensembleintegration AT cchristophercolumbus enhancedwindpowerforecastingusingmachinelearningdeeplearningmodelsandensembleintegration |