Hybrid yawed wake modeling: fusing steady-state wind field data and deep learning

As the wind power industry rapidly grows, wake interference within wind farms becomes increasingly significant. Yaw control is a key method to optimize power output and reduce structural loads, requiring accurate yaw wake predictions. Traditional yaw wake modeling falls into two categories: high-fidelity computational fluid dynamics (CFD) simulations and fast analytical models. CFD methods (e.g., RANS/LES/DNS) offer high accuracy but are computationally expensive and unsuitable for real-time applications. Fast analytical models are efficient but less accurate. They are adopted to handle only uniform or single-turbine yaw settings, and struggle with complex multi-turbine non-uniform yaw conditions. This paper presents a hybrid yaw wake modeling approach combining steady-state CFD simulation data with deep learning algorithm. Using a yaw-corrected actuator disk model (ADM) coupled with Reynolds-Averaged Navier–Stokes (RANS) k-ɛ turbulence model, a database of wakes under multi-turbine non-uniform yaw conditions is created. A deep neural network (DNN) is trained to predict turbine inflow velocities quickly and accurately for arbitrary yaw configurations. The proposed method achieves a balance between accuracy and computational efficiency, supporting real-time wind farm yaw control and optimization.