A short-term wind power prediction model based on SSA (sparrow search algorithm), optimized BiLSTM (bidirectional long and short-term memory neural network), and singular spectrum analysis was proposed in order to address the characteristics of wind power sequences with volatility and increasing complexity. First, features were extracted from historical power data using singular value analysis, and noise information interference was minimized using the denoising method. Second, the sparrow method was used to optimize the hyper-parameters of the BiLSTM model, and the wind power prediction model was constructed utilizing the BiLSTM, improving the effectiveness of model training. Finally, operational data from a wind farm was used to assess the model's accuracy and logic by comparing it to other models. Based on the experimental results, the improved model can effectively improve the accuracy of short-term wind power prediction, reducing the absolute error by 14.2% and the root-mean-square error by 4.24% relative to the baseline model. The improved BiLSTM model proposed in this paper has better prediction performance.