Wind turbine wake prediction modelling based on transformer-mixed conditional generative adversarial network

Wind turbine wake prediction considering yawed condition is of great importance for wind farm applications. Tremendous efforts have been made on yawed wind turbine wake prediction by using analytical wake models and numerical simulations, while their applications are still limited due to a lack of balanced consideration on both efficiency and precision. In this work, a deep learning-based wake prediction model is developed by integrating the transformer module into the conditional generative adversarial network. The developed model takes the inflow velocity and turbulence field as the input to predict the three-dimensional wake flow under different yawed conditions. The data generation approach by both analytical and numerical ways is used to build a large wake database. Subsequently, a pretraining-finetuning strategy is adopted to improve the model training efficiency and enhance the prediction performance. The validation results show that the proposed model can achieve a good agreement with numerical simulations at different streamwise distances under various inflow conditions, with the mean absolute relative error of 5.0 % and 7.27 % for wake velocity and turbulence intensity, respectively. The model parameters are also investigated to illustrate the wake prediction improvement by transformer-mixed modelling method. The wake prediction performance of the proposed model is validated by a comparison with analytical wake model and other popular machine learning-based methods. Moreover, the power calculation of multiple wind turbines is conducted to demonstrate the easy implementation and good performance in wind farm applications.