Dynamic stall modeling of the wind turbine blade with a data-knowledge-driven method

Dynamic stall is a typical unsteady phenomenon encountered by horizontal axis wind turbines (HAWTs), leading to abnormal blade power and load. Despite substantial growth in HAWTs, dynamic stall modeling still relies on semi-empirical models, which face challenges in large-scale and high-safety blade design. This study presents a unified data fusion modeling method to directly predict dynamic stalls for three-dimensional (3D) wind turbine blades. The components of the Leishman–Beddoes model are integrated into the data fusion neural network (DFNN) as physical knowledge. Furthermore, the DFNN is coupled with the blade element momentum method to obtain the dynamic stall behavior of the entire blade under different wind speeds , yaw angles , and radial positions . The results indicate that the DFNN can precisely capture the aerodynamic hysteresis and maximum loads associated with the NREL Phase VI blade under yaw misalignment inducing a weak coupling between dynamic stall and rotational effect . Compared to the initial BEM, the DFNN markedly improves the prediction accuracy of the loads on the 3D wind turbine blade by over four times. It is only trained by a small amount of 2D airfoil experimental data. The proposed modeling framework can form a new paradigm for wind turbine modeling and design. • A data-knowledge driven method integrating L-B model physics, trained with minimal airfoil experimental data. • The predictive accuracy of DFNN is improved by a 4-fold compared to the initial empirical model. • DFNN predicts aerodynamic effects of wind speed, yaw angle, and radial position variations.