Adaptive expert fusion model for online wind power prediction

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy

Background:

• Wind power prediction is vital for grid stability due to inherent variability.
• Existing methods struggle with real-time adaptation to changing weather and data distributions.

Purpose of the Study:

• To introduce a novel Adaptive Expert Fusion Model (EFM+) for accurate online wind power prediction.
• To enhance the adaptability and robustness of wind power forecasting models.

Main Methods:

• Developed an ensemble model (EFM+) integrating XGBoost and self-attention LSTM with dynamic weights.
• Implemented adaptive weight updates based on recent sample performance and error.
• Enabled Bayesian inference for real-time uncertainty quantification.

Main Results:

• EFM+ demonstrated superior prediction accuracy and reduced error compared to existing models.
• The model exhibited high robustness and stability across diverse operational scenarios.
• Sensitivity and ablation analyses confirmed the effectiveness of EFM+ components.

Conclusions:

• EFM+ offers a promising solution for online wind power prediction, effectively addressing nonstationarity and uncertainty.
• The adaptive fusion approach enhances forecasting reliability for power system operations.