Data-driven quantification of model-form uncertainty in Reynolds-averaged simulations of wind farms

Computational fluid dynamics using the Reynolds-averaged Navier–Stokes (RANS) remains most cost-effective approach to study wake flows and power losses in wind farms. The underlying assumptions associated with turbulence closures are biggest sources of errors uncertainties model predictions. This work aims quantify model-form RANS simulations farms at high Reynolds numbers under neutrally stratified conditions by perturbing stress tensor through a data-driven machine-learning technique. To this end, two-step feature-selection method is applied determine key features model. Then, extreme gradient boosting algorithm validated employed predict perturbation amount direction modeled toward limiting states on barycentric map. procedure leads more accurate representation anisotropy. trained high-fidelity data obtained from large-eddy simulation specific farm, it tested two other (unseen) distinct layouts analyze its performance cases different turbine spacing partial wake. results indicate that, unlike data-free which uniform constant entire computational domain, proposed framework yields an optimal estimation uncertainty bounds for RANS-predicted quantities interest, including velocity, intensity,