In order to improve the ultra-short-term prediction accuracy of wind power, this paper proposes an ultrashort-term prediction model for wind power based on integrated learning and deep neural network (Bagging-DNN). The method is based on Bagging and DNN, using Bootstrap method to generate sample sets, and training for different training subsets to obtain multiple integrated deep neural network models. By calculating the output of each model, the ultra-short-term prediction results of wind power are obtained. The analysis based on a case study of an offshore wind farm in Shanghai shows that the MAPE of this paper’s model is 5.0078% and 10.9658% when predicting 1h and 4h in advance. The MAPEs are on average 2.4% and 6% smaller than those of other methods, indicating that their prediction accuracy is higher than that of other prediction models. Compared with the BP and LSTM models, the square model in this paper has a narrower width of prediction interval, shorter training time, and more superior performance. In the process of iterative training, the model in this paper has a superior fitting effect.
