【目的】短期海上风电功率预测为国家电力系统的安全与稳定发挥着重要的作用，然而由于海上监测站点稀疏，导致风速、波浪等关键气象数据不足，且由于海上风浪流等多物理场耦合与风电机组电力电子设备开关控制相互作用的影响，让海上风电本身存在非线性和多频等特性，使得海上风电功率预测困难，为提高海上风电功率预测的准确度，更好地支持电力系统的稳定性。【方法】本文提出一种基于VMD-BiGRU与自注意力机制融合的短期海上风电功率预测方法。首先，利用变分模态分解（variational mode decomposition，VMD）算法，将原始数据分解为不同频率的本征模函数（intrinsic mode function,IMF），然后将IMF 和原始数据输入，用双向门控循环单元（bidirectional gated recurrent unit，BiGRU）网络和自注意力机制（self-attention mechanism，SAM）对短期风电功率进行预测，其中BiGRU与SAM采用残差连接和维度拼接的方法进行特征融合。最后，为了 验证所提出方法的可行性和优越性对比利时2024年海上风场短期风电功率进行预测，并与其他三种预测方法进行比较。【结果】结果表明，基于VMD-BiGRU与自注意力机制融合的短期海上风电功率预测方法将均方根误差减小至0.016 4，与其他方法相比具有更加优越的预测性能。【结论】基于VMD-BiGRU与自注意力机制融合的方法有效提升了短期海上风电功率预测精度，其多模块协同设计增强了模型的时序特征捕捉能力和鲁棒性。未来可通过参数优化和数据集扩展进一步提高模型的泛化能力。

[Objective] Short-term offshore wind power prediction plays an important role for the safety and stability of the national power system, however, due to the sparse offshore monitoring stations, resulting in insufficient key meteorological data such as wind speed, waves, etc., and due to the influence of multi-physical fields such as offshore wind and wave currents coupled with the interaction between the switching and control of the wind turbine's power electronic equipment, the offshore wind power itself has nonlinear and multi-frequency characteristics, making offshore wind power prediction difficult.[Methods] Aiming at this problem, a short-term offshore wind power prediction method based on the fusion of VMD-BiGRU and self-attention mechanism is proposed. Firstly, the original data are decomposed into intrinsic mode functions (IMFs) of different frequencies using the variational mode decomposition (VMD) algorithm,and then the IMFs and the original data are used as inputs for the prediction of short-term offshore wind power using the bidirectional gated recurrent unit (BGRU). bidirectional gated recurrent unit (BiGRU) network and self-attention mechanism (SAM) for short-term wind power prediction, in which BiGRU and SAM are fused using residual linking and dimensional splicing. Finally, in order to verify the feasibility and superiority of the proposed method to predict the short-term wind power of offshore wind farms in Belgium in 2024, and to compare it with the other three prediction methods. [Results] The results show that the short-term offshore wind power prediction method based on the fusion of VMD-BiGRU and self-attention mechanism reduces the root-mean-square error to 0.016 4, which provides a more superior prediction performance compared with other methods. [Conclusion] The method based on the fusion of VMD-BiGRU and self-attention mechanism effectively improves the accuracy of short-term offshore wind power prediction, and its multi-module synergistic design strengthens the ability of time-series feature capture and robustness of the model. The generalisation ability of the model can be further improved by parameter optimisation and dataset extension in the future.