Statistics of air flow past wind turbines from a Fokker Planck operator

Wind turbines can replace the generation of electricity from the burning fossil fuels and thereby ameliorate climate change. Simulations of the turbulent flow of air past turbines are needed to optimize the design of wind power farms. The simulations, however, are computationally expensive and require highly trained scientists to carry out. Machine learning offers possible relief from these two pressures. In this work, we investigate the performance of a Fokker-Planck Operator (FPO) constructed by machine learning from Large Eddy Simulations (LES) to predict the statistics of the turbulent flows. We identify and classify fundamental flow structures in phase space that constitute the basic building blocks of wind farm flows. Linear interpolation between different training data sets enables the accurate prediction of flow statistics for new configurations that have not been explicitly simulated. We conclude by discussing potential optimization strategies for improving FPO performance in this application.