Accurate prediction of wind power can effectively reduce grid connection fluctuation. The existing wind power prediction models have some problems, such as too many input features, difficult to select super parameters, and easy to lose important information in long time series. To this end, a short-term wind power fusion prediction model was proposed based on the sparrow search algorithm to optimize the bidirectional short-duration memory plus attention mechanism. First, SSA optimizes the super parameters such as the number of nodes, learning rate and training times of BiLSTM neural network to confirm the best parameters. Then, AM was introduced to assign different weights to input features of BiLSTM to strengthen key features. Finally, the proposed model is used to predict the wind power of 210MW wind farm in Xinjiang Province, and the results are compared with those of other models. The results show that the Root mean square error (RMSE) of the SSA-BiLSTM-AM prediction model is 5.4114, and the Mean absolute error (Mean absolute error, MAE is 3.6749. The prediction accuracy of SSA optimization algorithm and AM is significantly better than that of other models, which proves that SSA optimization algorithm and AM can effectively improve the short-term power prediction accuracy of wind turbines.