Accurate wind power prediction plays an important role in maintaining the safety and stability of the power system and improving wind energy utilization efficiency. This paper proposes a multi-step prediction method based on a hybrid deep learning model of Transformer and BiGRU to address the issues of significant error accumulation and insufficient adaptability to complex meteorological scenarios in wind power multi-step prediction. By combining Transformer and BiGRU networks, the global attention mechanism of Transformer and the temporal modeling capability of BiGRU are integrated to effectively suppress the accumulation of errors in multi-step prediction, and the optimal selection of hyperparameters is achieved through multiple experiments. The calculation results of different meteorological and power data sets of a wind farm in 2022 show that the improved Transformer-BiGRU hybrid network model is superior to the single model (GRU, BiGRU, Transformer) in the two to four step prediction. The results show that the prediction accuracy of the proposed method is better than that of the single model. Taking two-step prediction as an example, the proposed model has a coefficient of determination R2 of 0.952, indicating the highest accuracy. Compared with the three single models, the mean absolute error MAE, root mean square error RMSE and mean absolute percentage error MAPE decreased by 2.55%, 1.28% and 11.33% respectively, and the prediction accuracy was improved.