Development of deep learning-based reduced order model for turbine wake control

Wake interactions between turbines give rise to wind farm power generation losses of 10% to 20%. Although wake interaction can be addressed by increasing the distance between the wind turbines, continuous growth in the scale of turbines may also require an increase in turbine spacing. This may result in wind farms of prohibitive dimensions. Therefore, advanced turbine control such as wake steering with yaw misalignment has been proposed to reduce the distance between turbines and maximize power production. We implemented a reduced-order model capable of predicting the wake redirection and power production of a wind farm. Large eddy simulations of Sandia National Lab scaled wind farm technology (SWiFT) facility at different wind speeds, wind directions, and yaw-misalignment of the wind turbines were used to generate the training/testing datasets. It was shown that the deep learning algorithms accurately predict the velocity field and turbulence kinetic energy in the wake of a yawed wind turbine and the secondary redirection of downwind turbines.