An Ensemble Network for High-Accuracy and Long-Term Forecasting of Icing on Wind Turbines

Area of Science:

• Renewable Energy
• Artificial Intelligence
• Meteorology

Background:

• Wind turbine icing significantly reduces energy generation efficiency.
• Predictive maintenance is essential for optimizing wind farm operations.
• SCADA data offers valuable insights for forecasting environmental impacts.

Purpose of the Study:

• To develop a high-accuracy, long-term icing prediction model for wind turbine blades.
• To leverage SCADA sensor data for advanced wind energy forecasting.
• To mitigate power loss caused by wind turbine icing.

Main Methods:

• A novel deep learning network, PCTG (Parallel CNN-TCN GRU), was developed.
• The model integrates Convolutional Neural Networks (CNN), Temporal Convolutional Networks (TCN), and Gated Recurrent Units (GRU).
• SCADA time-series data from multiple wind turbines was utilized for training and validation.

Main Results:

• The PCTG model achieved an average prediction accuracy of approximately 97% for horizons up to 2 days.
• Maintained over 95% accuracy for long-term predictions up to 22 days ahead.
• Demonstrated over 99% accuracy for 10-day ahead icing prediction on a separate dataset.

Conclusions:

• The PCTG model offers a robust solution for accurate and long-term wind turbine icing prediction.
• This deep learning approach effectively captures temporal dependencies and variable interactions in SCADA data.
• The findings support proactive measures to prevent power loss and enhance wind energy reliability.