A Large-Eddy Simulation of Wind-Plant Aerodynamics

Abstract

In this work, we present results of a large-eddy simulation of the 48 multi-megawatt turbines composing the Lillgrund wind plant. Turbulent inflow wind is created by performing an atmospheric boundary layer precursor simulation, and turbines are modeled using a rotating, variable-speed actuator line representation. The motivation for this work is that few others have done large-eddy simulations of wind plants with a substantial number of turbines, and the methods for carrying out the simulations are varied. We wish to draw upon the strengths of the existing simulations and our growing atmospheric large-eddy simulation capability to create a sound methodology for performing this type of simulation. We used the OpenFOAM CFD toolbox to create our solver.The simulated time-averaged power production of the turbines in the plant agrees well with field observations, except with the sixth turbine and beyond in each wind-aligned. The power produced by each of those turbines is overpredicted by 25-40%. A direct comparison between simulated and field data is difficult because we simulate one wind direction with a speed and turbulence intensity characteristic of Lillgrund, but the field observations were taken over a year of varying conditions. The simulation shows the significant 60-70% decrease in the performance of the turbines behind the front row in this plant that has a spacing of 4.3 rotor diameters in this direction. The overall plant efficiency is well predicted. This work shows the importance of using local grid refinement to simultaneously capture the meter-scale details of the turbine wake and the kilometer-scale turbulent atmospheric structures. Although this work illustrates the power of large-eddy simulation in producing a time-accurate solution, it required about one million processor-hours, showing the significant cost of large-eddy simulation.

Observing and Simulating Wind-Turbine Wakes During the Evening Transition

“…Moreover, after the evening transition, the hub-height wind-speed deficit persists for more than 15 km downwind, while the turbulence enhancement vanishes beyond 10 km downwind. In contrast to LES studies with constant atmospheric stratification (Churchfield et al 2012;Mirocha et al 2014Mirocha et al , 2015Abkar and Porté-Agel 2015;Vanderwende et al 2016a), the wake structure illustrated herein varies both temporally and spatially downwind of the wind farm. The turbine locations in these simulations, based on an actual wind farm, contribute to this variability.…”

“…However, modelled turbine wakes grow horizontally twice as rapidly, and the wake meanders from the wake centreline to a greater extent in convective than in stable conditions (Abkar and Porté-Agel 2015;Vanderwende et al 2016a). Unfortunately, examining the wake behaviour of a sizable wind farm using LES is computationally expensive (Churchfield et al 2012), and the simulation of wind-farm impacts using mesoscale meteorological models is more practical.…”

Thek-ε-fPmodel applied to double wind turbine wakes using different actuator disk force methods

“…Prior to the first iteration in CFD, UH,∞ is guessed based on the free-stream wind speed, which provides the scaling coefficients CT , CP and Ω, that determine the AD forces. After the first iteration, UAD is probed at the AD and a new UH,∞ is calculated with equation (4). Subsequently, the scaling coefficients and AD forces are updated.…”

“…However, a similar assumption is made when tabulated airfoil data is used for the AD forces, to represent a wind turbine rotor from which the actual blade geometry and corresponding aerodynamics are unknown. This assumption is used by Porté-Agel et al [19] and Churchfield et al [4]. In their work the AD forces of the Siemens SWT-2.3-93 wind turbine are modeled by using a newly designed wind turbine, that mimics the known power production.…”

“…Wind turbine wake interaction has been studied numerically by many authors, using models that range from simple engineering wake models such as the N.O. Jensen model [3], to high fidelity Computational Fluid Dynamics (CFD) methods, including Large-Eddy Simulations (LES) [4,5]. Reynolds-Averaged Navier Stokes (RANS) solvers use a CFD method that is roughly three orders cheaper than the LES [6].…”