The VMD-CNN-Trans model used for ultra-short-term power prediction in wind farms.

The randomness of wind speed changes makes it difficult to achieve effective prediction of wind power. This problem has become a challenge in the consumption of new energy. To improve the accuracy of wind power prediction, a hybrid intelligent prediction model based on signal decomposition and deep learning is proposed in this paper. The model comprises Variable Mode Decomposition (VMD), 2D convolutional neural network (2D-CNN), and Transformer module. The VMD algorithm decomposes the wind speed and power change curves to construct multidimensional feature vectors. The 2D convolutional neural network is used to mine feature maps of different spatial and temporal features from multi-dimensional features. And the Transformer module is added to construct the spatiotemporal correlation between data. The example analysis is based on measured data from a wind power plant. The $E_{MAE}, E_{RMSE}$, and $R^{2}$ of the VMD-CNN-Trans model are 3.72MW, 6.86MW, and 98.90% respectively. Compared with other methods, the VMD-CNN-Trans model has certain advantages in wind power prediction. This result indicates the feasibility and advanced nature of the proposed approach.