Wind Turbine Siting by Using Mesoscale Model Data Assimilation and Computational Fluid Dynamics

Abstract

Accurately predicting energy production for wind turbines in realistic environmental conditions can be challenging due to terrain and uncertainty in weather conditions. To enhance predictive capability we have developed an approach in which a high-fidelity Computational Fluid Dynamics (CFD) computer code is coupled to a Mesoscale numerical weather prediction model. The CFD model captures local terrain and environmental conditions while the Mesoscale model defines the inflow boundary conditions as well as interior wind profiles for the CFD model. The novel approach of data assimilation is incorporated that uses either meteorological data or Mesoscale model predictions to obtain a flow solution that is closer to observations. A case study in a valley near Rock Springs, PA is analyzed. I. Introduction iting wind turbines accurately can be a complex undertaking. Traditional techniques such as using wind roses can lead to improper placement of a turbine because local effects such as terrain and surface heating are not normally assessed. Additionally, atmospheric motion is described by nonlinear dissipative dynamical systems and it is sensitive to initial and boundary conditions. While traditional Unsteady Reynolds Averaged Navier Stokes (URANS) computer codes are used in this class of problems because they can model complex terrain and include surface roughness models, they still have an underlying problem in that they employ ensemble averaging with turbulence closures which can be inconsistent with the realization that is occurring and that of the ensemble average that the equations predict. This mismatch could lead to poor prediction of the energy that a turbine might extract from a particular location. If additional information about a flow field is available from mast observations, or from mesoscale models, better approximations to an actual realization should be possible. Haupt et al. (2009, Beyer-Lout 2007), used this type of data assimilation to more accurately predict realizations. They were able to do this by first identifying the characteristics of the realization and then use this data to back-calculate better flow modeling variables to match that realization (Haupt et al. 2009, Beyer-Lout 2007). This current work seeks to predict details of fine-scale motion that include the impact of local terrain, heating information, land-use processes, and input from a mesoscale numerical weather prediction model. The challenge is to assimilate such information into a standard computational fluid dynamics (CFD) RANS model. Such an effort requires new assimilation techniques that merge profiles at several locations as computed by the mesoscale model into the CFD simulation without double counting the subgrid scale motions and that is smooth enough to prevent prohibitive gravity wave action. Recent work toward siting (Laporte 2009) incorporates mesoscale modeling as boundary conditions to CFD models. We try to extend this idea with data assimilation. The new technique is tested in complex terrain near Rock Springs, PA with computed profiles input from the Weather Research Forecast (WRF) model run at Penn State. Section II describes the site. The mesoscale model as well as the CFD model are described in section III. That section also discusses the assimilation procedureand gives some preliminary results. Section IV summarizes this work and discusses prospects for future work.