Abstract: An ultra-short term wind power prediction method based on deep learning and error correction is proposed in this paper.First,a bidirectional gated recurrent unit network model is established to predict the wind power and errors of the primary model are extracted. Then,based on the primary errors,an error model is constructed using random forest algorithm to correct the primary results.Finally,using the kernel density estimation to fit the probability distribution of the corrected errors,the confidence interval is calculated. Based on the measured data of a wind farm in China,the effectiveness and applicability of the proposed method are verified by the results of multi-time scale wind power prediction.