Enhanced wind power forecasting using machine learning, deep learning models and ensemble integration

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Grid Stability and Management

Background:

• Wind energy variability necessitates accurate forecasting for grid stability.
• Traditional statistical models struggle with complex, nonlinear temporal patterns in wind data.
• Gaps exist in utilizing real-time data, comparative model analysis, and systematic model selection for wind energy forecasting.

Purpose of the Study:

• To enhance wind energy forecasting performance using advanced machine learning (ML) and deep learning (DL) techniques.
• To address limitations of traditional models by incorporating real-time data and systematic hyperparameter tuning.
• To conduct comparative analyses of diverse ML/DL models and develop robust forecasting strategies.

Main Methods:

• Evaluated various ML models (Random Forest, XGBoost, etc.) and DL models (LSTM, MLP).
• Utilized SCADA and real-time wind data, incorporating weather features like wind speed.
• Developed a Stacking Ensemble model from top-performing individual models for improved reliability.

Main Results:

• Random Forest (RF) excelled in initial datasets; RF, XGBoost, and Stacking Ensemble showed high performance in real-time data.
• Stacking Ensemble achieved R-squared values up to 0.998, with low MAE, MSE, and RMSE.
• Hyperparameter tuning and ensemble methods significantly boosted predictive accuracy and forecast reliability.

Conclusions:

• Advanced ML/DL techniques, particularly hyperparameter-tuned stacking ensembles, are highly effective for wind energy forecasting.
• Improved forecasting supports better resource management, grid reliability, and operational planning for renewable energy.
• The study advances renewable energy forecasting, contributing to global sustainability goals.