Wind farm wake modeling based on deep convolutional conditional generative adversarial network

Abstract

Modeling of wind farm wakes is of great importance for the optimal design and operation of wind farms. In this work a surrogate modeling method for parametrized fluid flows is proposed for wind farm wake modeling, based on the state-of-the-art deep learning framework i.e. deep convolutional conditional generative adversarial network. Based on the proposed method and the data generated by high-fidelity large eddy simulations, a novel wind farm wake model is developed. The developed model is first validated against high-fidelity data and the results show that it achieves accurate, efficient, and robust prediction of wind turbine wake flow, at all the streamwise locations including both near wake and far wake, for both streamwise and spanwise velocity components, and at the cases with different inflow wind profiles. Then an extensive parametric study is carried out and the results show that the model generalizes well to unknown flow scenarios. Furthermore, a case study for a wind farm is investigated by the developed model. The prediction results are then compared with high-fidelity simulations, showing that the model can predict the wind farm wake flow (including both the streamwise and spanwise velocity fields) very well.