AI for wind farm layout optimization
In the wind energy industry, simulations of wind farms are required at the design stage but also for short (days) and long (months to years) term power predictions. At the design stage, the objective is to assess the potential of the farm and to optimize the placement of the wind turbines on the terrain. Short and long term forecasts are necessary for the operator to assess the wind resources available in the coming days or months. In all cases, an accurate description of the airflow through topography, generally involving complex and heterogeneous terrains, and the wind turbine effects are necessary.
For such so-called microscale simulations, the wind industry currently resorts to solving the Reynolds-averaged Navier Stokes (RANS) equations and even more simplified models to understand the wind at a site prior to wind farm construction. Thanks to ever-growing computational power and improved algorithms, LES is gaining popularity thanks to the superior accuracy it can provide compared to RANS, for example for modelling separated flows, as shown in Figure 1.
Figure 1: Turbulence generated by a cliff on Bolund Island, Denmark.
The various wind turbine models come with different degrees of approximation and computational costs; The simplest approximation is the actuator disc (AD) model, providing an acceptable accuracy for the wakes at a distance greater than 2.5 diameters downwind. The actuator line (AL) model enables to account for the rotational effect, depending on airfoil properties and the angular velocity. While AD simulations are typically steady, AL simulations are transient and thus significantly more expensive. Despite their advantages over AD models, AL models have limitations, i.e., vortices close to the blades and 3D effects cannot be simulated. These different levels of approximation are illustrated in Figure 2.
Figure 2: Different levels of wind turbine modeling in wind farm simulations.
With the advent of Exascale computers, high-accuracy simulations resolving rotating blades via full rotor models for whole wind farms and including FSI become possible. They will foster improvements in the blade, turbine, and wind plant design/optimization, by providing new knowledge at fidelity that is unattainable by any other numerical or experimental means. Such simulations are the target of Exascale projects in the U.S. and Europe represented by EoCoE-II despite the extremely innovative haracter of the proposed objectives of EoCoE-II, these simulations cannot be envisaged on a daily basis, and compromises need to be made. To overcome this, the wind turbines are modeled through DL techniques in CoE RAISE, i.e., through surrogates. The surrogate is generated for specific inflow conditions, using different wind velocities, and levels of turbulence intensity, through high-fidelity LES of full rotors. These models are incorporated in the simulation of complete wind farms containing hundreds of wind turbines.
Barcelona Supercomputing Center
Torre Girona c/Jordi Girona, 31, 08034 Barcelona (Spain)
e-mail: info [@] bsc.es