Prediction of wind farm wake and output power using generative adversarial network and convolutional neural network

Wake and output power prediction of wind turbine is critical for the wind farm layout optimization. Previous studies used analytical wake models and computational fluid dynamics (CFD) methods to fulfill this prediction. However, these methods either exhibit inadequate prediction accuracy or need excessive computational demands during prediction processes. Thus, a novel wind farm prediction system is established using the full-model CFD simulation and a surrogate modeling method based on convolutional neural networks and generative adversarial networks to ensure the fidelity and efficiency of the prediction. By containing an incoming speed distribution generator module (Gin), a wake distribution generator model (Gw), a rotational speed prediction module (R), and a power prediction module (P), this system can predict high-dimensional incoming data, the rotational speed, power output, and three-dimensional wake fields. The system uses the incoming wind speed (Vin) and turbulence intensity to determine the optimal placement of turbines in the wind field. The Gin, R, Gw, and P module are validated through high-resolution experimental and computational data. The system is applied to a tandem wind farm and the Horns Rev 1 wind farm. The predicted data show satisfactory agreement with high-resolution experimental and computational data, fully validating the robustness and generalization capability of the system. The proposed prediction system is of great engineering significance for optimizing wind farm layout.