The Lattice Boltzmann Method for Wind Farm Simulations: Validation and Application

Many new challenges in wind energy require the use of large eddy simulation for accurate modeling of wind farm flows. However, the immense computational cost hinders its use in research and industry. The lattice Boltzmann method is the most promising candidate to date to achieve the highest level of accuracy while decreasing computational cost by orders of magnitude compared to traditional methods. In this thesis, I present further development of the lattice Boltzmann method for wind energy and compile various applications, such as industrial use, generation of training data for machine learning, and analysis of wind farm control paradigms. In order to evaluate the requirements of different industrial stakeholders, we conduct a survey among industry experts on the use of large eddy simulation and show that the run time requirements indicated by many respondents can be met with the current state of the lattice Boltzmann method. In a validation study, the lattice Boltzmann method is as accurate as traditional Navier-Stokes solvers, while reducing computational cost by one to more orders of magnitude. A convolutional neural network is trained to predict average flow velocities in the wake of a single turbine. The predictions exhibit very high accuracy at execution times similar to engineering models. The lattice Boltzmann method enables the generation of larger training sets at a feasible computational cost. A proof of concept is provided for the use of reinforcement learning to discover new, cooperative wind farm control mechanisms. In an extensive analysis of the helix approach, its physical mechanisms are elucidated and a thorough parameter study of the wake of a single turbine is provided. We also study the interaction of multiple helical wakes, providing a way to extend the approach from a pair of turbines to wind farms