The worldwide growth of wind power capacity underscores the need for more extensive data utilization in critical areas such as Operation & Maintenance, condition monitoring, and forecasting. However, the widespread use of Machine Learning in wind power carries the risk of excessive reliance on complex black-box models. To address this, our study focuses on developing an eXplainable Artificial Intelligence (XAI) framework for multivariate wind power problems. The workflow involves a Sequential Features Selection algorithm to identify suitable variables for regression. Shapley coefficients are then computed to estimate each feature's impact on the output, revealing hidden patterns and relationships. The framework effectiveness is validated through two substantial test cases: multivariate power curve analysis for turbine condition monitoring and ultra-short-term forecasting for wind farm operations. The outcomes underscore the algorithm's proficiency in fine-tuning feature selection and providing comprehensive explanations for output behaviors. By adopting this XAI framework, wind power researchers and practitioners can navigate multivariate problems with enhanced interpretability, developing transparent models for specific applications within the wind power domain.
Wind Power Applications of eXplainable Artificial Intelligence Techniques
Astolfi D.;
2023-01-01
Abstract
The worldwide growth of wind power capacity underscores the need for more extensive data utilization in critical areas such as Operation & Maintenance, condition monitoring, and forecasting. However, the widespread use of Machine Learning in wind power carries the risk of excessive reliance on complex black-box models. To address this, our study focuses on developing an eXplainable Artificial Intelligence (XAI) framework for multivariate wind power problems. The workflow involves a Sequential Features Selection algorithm to identify suitable variables for regression. Shapley coefficients are then computed to estimate each feature's impact on the output, revealing hidden patterns and relationships. The framework effectiveness is validated through two substantial test cases: multivariate power curve analysis for turbine condition monitoring and ultra-short-term forecasting for wind farm operations. The outcomes underscore the algorithm's proficiency in fine-tuning feature selection and providing comprehensive explanations for output behaviors. By adopting this XAI framework, wind power researchers and practitioners can navigate multivariate problems with enhanced interpretability, developing transparent models for specific applications within the wind power domain.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.