Condition Monitoring of Wind Turbine Systems by Explainable Artificial Intelligence Techniques

Area of Science:

• Renewable Energy Engineering
• Artificial Intelligence in Energy
• Machine Learning for Condition Monitoring

Background:

• Wind turbine performance evaluation traditionally uses univariate power curves, often failing to capture complex operational dynamics.
• Multivariate models are needed to account for factors like working parameters and ambient conditions influencing power output.
• Existing methods may overlook crucial variables affecting wind turbine efficiency and health.

Purpose of the Study:

• To develop and validate a data-driven methodology using explainable artificial intelligence (XAI) for constructing multivariate wind turbine power curves.
• To establish a reproducible workflow for identifying the most influential input variables for condition monitoring.
• To enhance anomaly detection by uncovering previously unexplored explanatory variables.

Main Methods:

• Application of explainable artificial intelligence (XAI) for building multivariate power curve models.
• Sequential feature selection to minimize root-mean-square error and identify optimal input variables.
• Computation of Shapley coefficients to quantify the contribution of each selected variable to model error.

Main Results:

• The proposed methodology effectively detects hidden anomalies in wind turbine performance.
• A novel set of highly explanatory variables related to rotor and blade pitch control was identified.
• The approach demonstrated superior performance compared to conventional univariate models.

Conclusions:

• XAI-powered multivariate power curves offer significant advantages for wind turbine condition monitoring.
• The methodology provides novel insights into operational factors affecting turbine performance and enables more robust anomaly detection.
• This approach paves the way for improved predictive maintenance and operational efficiency in wind energy.