Key points are not available for this paper at this time.
Abstract The performance of six subgrid-scale (SGS) models is analyzed for large-eddy simulations (LES) of wind-farm flows under stable (SBL) and conventionally-neutral (CNBL) atmospheric conditions. A precursor–concurrent technique is employed to provide fully developed turbulent inflow for simulations of a 40-turbine wind farm. Turbines are represented using the actuator-disc method, employing a baseline grid of 12 cells across the turbine diameter. The SBL precursor flow poses a challenge for LES, as it may not be able to resolve the small turbulent scales featured in this flow if the grid is coarse. For these precursor flows, the baseline grid results of all six SGS models are assessed relative to coarser and finer grids, with 6 and 45 cells across the diameter, respectively. The wall-adapting local eddy-viscosity (WALE) and Lagrangian-averaged scale-dependent dynamic (LASDD) models exhibit high grid sensitivity, while the standard Smagorinsky (Smag.), anisotropic minimum-dissipation (AMD), one-equation turbulent kinetic energy (TKE), and stability-dependent Smagorinsky (SDS) models show low sensitivity. For the wind-farm simulations conducted with the baseline grid, the AMD and SDS models predict similar wind-farm performance. In contrast, the WALE and LASDD models predict nearly 30% less power output, primarily due to their prediction of lower inflow wind speeds. CNBL simulations on the baseline grid show reduced sensitivity to the SGS model due to larger atmospheric turbulence and length scales compared to the SBL flow. Among the six models, the AMD model demonstrates ease of implementation, the least sensitivity to grid size for the SBL precursor flow, and predictions that are consistent with other models and higher-order pseudo-spectral LES solvers, making it a suitable choice for LES of wind-farm flows under both stable and conventionally-neutral conditions.
Ghobrial et al. (Tue,) studied this question.