Deep learning-based wind farm power prediction using Transformer network

Accurate wind farm power prediction is of vital importance for the performance improvement of wind farms and their grid integration. In this paper, a novel method based on the state-of-the-art deep learning model (i.e. the Transformer network) is developed to tackle this issue. Specifically, the prediction task is modeled as a segmentation problem and the powerful Vision Transformer (ViT) is employed to predict each individual turbine's power generation in a wind farm with wake interaction effects. The proposed method, called Wind Transformer (WiT), is evaluated by carrying out a set of numerical experiments. The results show that the proposed method achieves accurate and efficient wind farm power prediction and it outperforms other deep learning baseline models significantly. Particularly, the maximum mean absolute percentage error by the proposed method is only 1.030%, while they are 4.350% for LSTM and 3.510% for CNN models.