We propose a Mamba-Transformer model integrating physical constraints and multi-scale features for ultra-short-term wind power forecasting. The proposed model employs complete ensemble empirical mode decomposition with adaptive noise to capture nonlinear and nonstationary patterns in wind power data across multiple frequency components. The decomposed modal functions are then fed into the Mamba model alongside meteorological data from the wind farm to uncover local time-varying properties. The self-attention mechanism of the Transformer model is then employed to capture long-range dependencies among multi-scale features. Finally, a one-dimensional Jensen wake model is integrated to establish physical constraints, integrating the prediction results of physical models and data-driven models through an adaptive weighting mechanism. Experimental results demonstrate that compared to the Transformer model, the proposed model reduces root mean square error （RMSE） and mean absolute error （MAE） by 41.29% and 50.26%, respectively. This model enhances wind power forecasting accuracy while exhibiting strong generalization capabilities.

wind power forecasting  /  physical constraint  /  CEEMDAN  /  Mamba  /  Transformer  /  one-dimensional Jensen wake model