With more and more serious global shortage of fossil fuels, the development and utilization of renewable energy has attracted more and more attention. Wind energy is one of the most widely used clean energy sources. As the main utilization form of wind energy, wind power needs to be predicted in the production work, which can be done in the short term based on the historical data recorded in daily wind field. However, the existing methods often only use the historical data in their own domain, resulting in one-sided results and large limitations. They fail to effectively use the implicit connections in the data, and are unable to suppress the model performance degradation caused by the loss of original data or outliers. To address these challenges, this paper proposes a short-term wind power prediction model based on deep migration of historical data. Firstly, the deep neural network model is built by using the automatic coding mechanism with noise reduction processing. The hidden layer is then shared by the deep migration method, and the hidden links between features are mined. Finally, the important knowledge is transferred from the wind field data with similar features and geographical locations, so as to improve the accuracy and reliability of the model. The experimental results show that the proposed method can make full use of the existing data and improve the prediction accuracy significantly.