Abstract: Accurate wind power prediction helps grid operators grasp the power generation trend of wind farms in advance, thereby optimizing the power dispatching plan, achieving a dynamic balance between power supply and demand, and effectively alleviating the impact of wind power fluctuations on the stability of the power grid. To this end, this study proposes a wind power prediction model that integrates the attention mechanism with the deep learning model. The attention mechanism performs non-uniform weighting on the contribution of the input feature vector, thereby optimizing the learning objective. Convolutional neural network (CNN) is helpful for extracting short-term attributes to obtain high-dimensional attributes, while gated recurrent unit (GRU) can solve the problem of inaccurate predictions caused by the merging of original data. The experimental results show that the performance of the single LSTM and GRU models is relatively poor; the CNN-LSTM and CNN-GRU models have achieved a good balance in terms of accuracy and computational efficiency, and are suitable for scenarios that have certain requirements for prediction accuracy and hope to obtain results quickly. Although the CNN-GRU-Attention model takes a long time to calculate, it can make predictions most accurately. If the requirement for computing time is not high and high-precision prediction is pursued, CNN-GRU-Attention is a better choice.