Wednesday 09 April 2025
A novel approach to detecting icing on wind turbine blades has been developed, using a combination of machine learning and data privacy techniques. The new method, known as FedHPb, is designed to tackle the problem of class imbalance in wind turbine blade icing detection, where there are far more non-icing events than icing events.
Wind turbines rely heavily on accurate weather forecasts to ensure optimal operation. However, icing events can significantly reduce power output and even cause damage to blades. Current methods for detecting icing use a combination of sensors and machine learning algorithms, but these approaches have limitations. For example, they may not perform well in scenarios where there is limited data or when the icing event is rare.
FedHPb addresses this issue by using a federated learning approach, which enables multiple wind turbines to share their data while maintaining privacy. The method combines this with a prototype-based model that learns to recognize patterns in the data and adapt to new situations. This allows FedHPb to improve its performance over time, even when faced with imbalanced data.
In addition to its ability to handle class imbalance, FedHPb also offers improved data privacy. By only transmitting aggregated data, rather than individual turbine readings, the method reduces the risk of sensitive information being exposed. This is particularly important for wind farm operators, who need to protect their data to maintain a competitive advantage.
The performance of FedHPb was tested on real-world data from two wind farms in China and compared to five other machine learning models. The results showed that FedHPb outperformed the other methods, achieving an average improvement of 19.64% in its Fβ score and 5.73% in its balanced accuracy.
The development of FedHPb has significant implications for the wind energy industry. By enabling more accurate detection of icing events, it could help reduce the risk of damage to blades and improve overall power output. The method’s focus on data privacy also ensures that sensitive information is protected, maintaining trust between wind farm operators and their customers.
The next step in the development of FedHPb is to integrate it with existing weather forecasting systems. This would enable wind turbines to receive more accurate warnings about icing events, allowing them to take proactive measures to reduce damage. As the method continues to evolve, it has the potential to make a significant impact on the efficiency and reliability of wind energy production worldwide.
Cite this article: “Federated Learning for Wind Turbine Icing Detection: A Novel Class-Imbalanced Heterogeneous Approach”, The Science Archive, 2025.
Wind Turbine, Icing Detection, Machine Learning, Data Privacy, Federated Learning, Class Imbalance, Weather Forecasting, Power Output, Wind Farm, Energy Production
