Microscale Wind Simulations over Arbitrarily Complex Terrain using Cartesian Methods and GPUs

The variability of the wind is a major concern for efficient and reliable grid integration of wind energy. Utility companies and balancing authorities need accurate wind power forecasting methods to balance the variable wind energy with other energy resources. Depending on the forecast range, persistence and regional numerical weather prediction models are blended, sometimes along with other techniques, to forecast power from wind parks. Such blended models appear to meet current industry expectations when applied to relatively flat terrain. However, large forecast errors have been reported when these models are applied to wind farms over complex terrain. Persistence models do well at very short time horizons, but tend to lose accuracy for forecast horizons longer than an hour. Application of persistence models for complex terrain areas is also questionable. The approach we present here is a microscale computational fluid dynamics (CFD) simulation that resolves arbitrarily complex terrain with an immersed boundary approach and with spatial resolutions on the order of ten meters.

Our approach resolves turbulence at the microscale over short-term time horizons (0-6 hours) using a large-eddy simulation technique with dynamic subgrid-scale models. The issue of near-surface modeling is addressed by resorting to a Reynolds-averaged description of turbulence in the length-scale definition. The boundary conditions at an arbitrarily complex terrain are imposed using a Cartesian grid immersed boundary method with a logarithmic reconstruction scheme to ensure consistency with the length scale definition in the turbulence model. Forecasting mode in the computations is enabled by designing the entire simulation for clusters of graphics processing unit (GPU) with an MPI-CUDA implementation.

We use the OpenFOAM CFD model to assess the performance of our new microscale model in terms of execution speed and accuracy of adopted approaches. We investigate the differences between our Cartesian approach and the terrain-fitted approach of OpenFOAM. We adopt well-known complex terrain cases such as Askervein Hill and Bolund Hill along with other areas of arbitrarily complex terrain to observe differences in mean velocity profiles and other turbulent statistics. We explore different approaches to lateral boundary conditions, investigating different inlet profiles and the use of inflow conditions from precursor simulations. Investigating the lateral boundary conditions provides insight into coupling a microscale complex terrain approach with a mesoscale weather model to create a multi-scale forecasting engine that we will be part of our future work.