High-fidelity prediction of wind turbine wakes: Enhancing wake models using LES-trained machine learning algorithms

An innovative machine learning (ML) model is introduced to efficiently predict high-fidelity three-dimensional velocity fields in the wakes of utility-scale wind turbines. The model takes low-fidelity velocity fields from an analytical engineering wake model as input and produces high-fidelity velocity fields. Large-eddy simulations (LES) of the Sandia National Lab Scaled Wind Farm Technology (SWiFT) facility at different wind speeds, wind directions, and yaw misalignments of the turbines were performed to generate high-fidelity velocity fields for training and validation. The input to the ML model consists of the three-dimensional velocity field of the SWiFT facility obtained from the Gauss Curl Hybrid (GCH) model. When compared against LES results, the ML model reduced prediction errors of the GCH model from 20% to less than 5%. Additionally, the ML model accurately captured the non-symmetric wake deflection observed for opposing yaw angles in wake steering cases, thereby increasing the accuracy over the GCH model. The computational cost of the trained ML model is comparable to that of the GCH model while providing results nearly as accurate as the high-fidelity LES.