To further improve the performance of wind speed prediction and to explore the nonlinear characteristics and disorder of wind speed, an ultra-short-term wind speed combined prediction model based on WOA-VMD and CNN-Transformer is proposed. First, to reduce the complexity of wind speed prediction, the whale optimization algorithm is employed to optimize the parameters of the variational mode decomposition algorithm. The wind speed sequence is effectively decomposed into several modal components with different frequencies, enabling denoising of the original wind speed data. Second, CNN-Transformer deep learning combined prediction models are constructed for the decomposed wind speed sub-sequences. CNN is used to capture the local temporal features and variation trends of the input wind speed sequence, and these features are then fed into the Transformer. The Transformer model constructs long-range dependencies based on the extracted local features to capture global sequence information and complex dependencies. Finally, the predicted results of each component are superimposed to obtain the final result. Wind speed datasets from different regions are used for experiments. The experimental results show that the proposed method exhibits good prediction performance, can meet the requirements of wind speed prediction and demonstrates good practicability.