Abstract: Accurate short-term wind power forecasting is critical to stable power system operation. To improve the short-term prediction accuracy, a combined prediction model based on variational mode decomposition（VMD）, sample entropy（SE） and improved bidirectional long short-term memory（BiLSTM） with error correction using Attention mechanism is proposed. Firstly, VMD is used to decompose the original power data into several relatively smooth subsequences and reconstruct the sample entropy similar components to reduce the complexity. Then, attention is introduced to assign corresponding weights to the state outputs of the implicit layer of BiLSTM to highlight the important influential input features, and extreme gradient boosting（XGBoost） is used to correct the error so as to further improve the prediction accuracy. Finally, the final result is obtained by adding the preliminary prediction and the revised prediction. The actual data of wind farms are used for verification, and the results show that the mean absolute error（MAE） of the proposed combined model decreases to 1.6565, and the accuracy is improved by 25.8%-56.5% compared with other models, which has a better prediction effect.