Towards data-driven turbulence modeling for wind turbine wakes

The Dutch energy strategy expects renewable energy sources like wind and solar to provide around 70% of the yearly electricity by 2030. In order to achieve these targets, models that efficiently and accurately capture the flow around wind turbines would be immensely helpful for both planning and operation of wind farms.

For wind turbine wake interaction, computationally cheap and simple engineering models fail to capture the more complex flow physics, whereas LES based models do very well but are computationally too expensive. An alternative is to use Reynolds-Averaged Navier-Stokes (RANS) solvers which lie somewhere between LES and engineering models. However, these models have structural shortcomings for many applications and development of better models has stalled in the past decades.

More recently, data-driven techniques have been used to try and derive better, application-specific models. In this work, a combined methodology between a baseline RANS model and a data-driven correction is presented. The resulting models give significantly better predictions than the baseline model for both velocity and turbulent kinetic energy. Similar to traditional Nonlinear Eddy Viscosity Models, the models initially showed numerical instability, but a pragmatic solution was found for this.

The novelty of the results presented in this thesis lie in the application of the methodology to higher Reynolds numbers and 3D test cases. However, there still remains much to be done before data-driven models can be useful in industrial practice. This would require larger datasets and more efficient algorithms for both training and testing of the data-driven corrections to the baseline turbulence model.