Innovative sparse data reconstruction approaches for yawed wind turbine wake flow via data-driven and physics-informed machine learning

This paper explores innovative approaches for reconstructing the wake flow field of yawed wind turbines from sparse data using data-driven and physics-informed machine learning techniques. The physics-informed machine learning wake flow estimation (WFE) integrates neural networks with fundamental fluid dynamics equations, providing robust and interpretable predictions. This method ensures adherence to essential fluid dynamics principles, making it suitable for reliable wake flow estimation in wind energy applications. In contrast, the data-driven machine learning wake flow estimation (DDML-WFE) leverages techniques such as proper orthogonal decomposition to extract significant flow features, offering computational efficiency and reduced reconstruction costs. Both methods demonstrate satisfactory performance in reconstructing the instantaneous wake flow field under yawed conditions. DDML-WFE maintains comparable performance even with reduced measurement resolution and increased noise, highlighting its potential for real-time wind turbine control. The study employs a limited number of measurement points to balance data collection challenges while capturing essential flow field characteristics. Future research will focus on optimizing turbine control strategies in wind farms by incorporating multi-scale modules and advanced data-driven techniques for temporal prediction of wake flow fields.