为减少超短期风电功率预测的误差，提出基于变分模态分解(variational mode decomposition，VMD)-排列熵(permutation entropy，PE)和多层双向长短时记忆(multilayer bidirectional long short-term memory, MultiBiLSTM)组合的超短期风电功率预测模型。首先，利用VMD分解算法将历史风电功率序列分解成若干个子模态分量，根据计算的PE值重构分解的子模态风电分量；然后，使用特征注意力(feature attention，FA)机制和深度残差级联网络(deep residual cascade network，DRCnet)构建MulitiBiLSTM预测模型，预测重构后的子序列；最后，重构子序列预测值，得到最终风电功率预测结果。使用贵州某风场的数据集对所提出的方法进行验证，并和其他预测模型进行对比。结果表明，使用VMD-PE-MultiBiLSTM模型能显著降低风电功率预测误差。

In order to reduce the error of ultra-short-term wind power prediction, an ultra-short-term prediction model of wind power based on variational mode decomposition (VMD), permutation entropy (PE) and multilayer bidirectional long short-term memory (MultiBiLSTM) is proposed. Firstly, the historical wind power sequence is decomposed into several sub-modal components using VMD decomposition algorithm, and the sub-modal wind power components are reconstructed according to the calculated PE value. Then, the feature attention (FA) mechanism and deep residual cascade network (DRCnet) are used to construct a MulitiBiLSTM prediction model to predict the reconstructed subsequences. Finally, the predicted value of the sub-sequence is reconstructed to obtain the final prediction result of wind power. The datum set of a wind field in Guizhou province is used to verify the proposed method and compare it with other prediction models. The results show that using VMD-PE-MultiBiLSTM model can significantly reduce the prediction error of wind power.