Abstract: To further improve the accuracy of short-term wind power prediction, a short-term wind power prediction method is proposed considering feature reconstruction, multiple cross-attention, and the combination of a Bidirectional temporal convolutional neural network (BiTCN) and Transformer. Considering the non-stationarity, deep periodicity and trend of wind power data, we design a periodicity and trend feature enhancement module based on the parallel feature extraction of BiTCN and Transformer encoder. This module utilizes the BiTCN’s capability to extract bidirectional global dependencies, local spatial features and detailed information of temporal data, and integrates them with the Transformer encoder to capture the temporal data, long-term dependencies, and global information. The model's perception and expression capabilities are significantly enhanced through spatio-temporal feature fusion with multiple cross-attention mechanisms. Firstly, the raw power data and meteorological data are simultaneously fed into the BiTCN and Transformer encoder modules to extract bi-directional spatial features, local information, and long-term temporal features, respectively. Secondly, the complete ensemble empirical mode decomposition with adaptive noise-sample entropy (CEEMDAN-SE) algorithm is employed to reconstruct the power data into a trend sequence and a periodic sequence. These sequences are subsequently input into BiTCN and Transformer encoder modules respectively, to enhance the periodic and trend features. Finally, the short-term wind power prediction is achieved with high precision based on the results of multi-layer cross-attention fusion feature extraction. Comparing the single model and the traditional time series prediction models, the results demonstrate that the proposed method can effectively improve the prediction accuracy and reduce the prediction error. This not only validates the effectiveness and practicability of the proposed method but also has certain application value in the field of short-term wind power prediction.