A hybrid prediction model based on learning Osprey Optimization Algorithm and Transformer‐BiLSTM is proposed to address the nonlinear, high volatility, and time sequence dependence issues in short‐term wind power forecasting. Firstly, by introducing adaptive inertia weights and dynamic reverse learning strategies to improve the traditional Osprey algorithm, the algorithm's global optimization capability is enhanced; Secondly, a Transformer‐BiLSTM dual channel feature extraction network is constructed, which utilizes the multi head attention mechanism of Transformer to capture long sequence global features, and combines with a bidirectional long short‐term memory network to extract bidirectional local features of time‐series data; Finally, LOOA is used to adaptively optimize the hyperparameters of the model. Experiments based on actual wind farm data show that compared to the standard Transformer‐BiLSTM model, the Root Mean Square Error (RMSE) has decreased by 41.33%, and the Mean Absolute Error (MAE) has decreased by 44.33%, verifying the effectiveness of the model in temporal feature mining and improving prediction accuracy. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.
Xue et al. (Sun,) studied this question.