This paper proposes a combined model for ultra-short-term offshore wind power prediction based on PCA-SSA-VMD and BiLSTM. The model aims to improve the accuracy of wind power forecasting by addressing the strong randomness and time correlation issues in offshore wind power data. The approach involves several steps: first, principal component analysis (PCA) is used to reduce the dimensionality of the data. Then, the sparrow algorithm (SSA) optimizes the variational modal decomposition (VMD) to decompose the wind power time series into different frequency components, removing noise. The hyperparameters of the bidirectional long- and short-term memory neural network (BiLSTM) are optimized using the SSA algorithm, and the final power prediction is obtained. Simulation experiments show that the model effectively improves prediction accuracy and verifies the effectiveness of the prediction model. The model uses PCA to reduce the dimensionality of the data, SSA to optimize VMD for noise removal, and BiLSTM for prediction. The results show that the model outperforms other models in terms of prediction accuracy, with high R² values and low RMSE, MAE, and MAPE. The model is tested on real-world data from offshore wind farms, and the results demonstrate its effectiveness in predicting wind power output. The study highlights the importance of combining dimensionality reduction, noise removal, and advanced neural networks for accurate wind power forecasting. The proposed method is expected to contribute to the safe and stable operation of offshore wind farms by improving the accuracy of wind power predictions.This paper proposes a combined model for ultra-short-term offshore wind power prediction based on PCA-SSA-VMD and BiLSTM. The model aims to improve the accuracy of wind power forecasting by addressing the strong randomness and time correlation issues in offshore wind power data. The approach involves several steps: first, principal component analysis (PCA) is used to reduce the dimensionality of the data. Then, the sparrow algorithm (SSA) optimizes the variational modal decomposition (VMD) to decompose the wind power time series into different frequency components, removing noise. The hyperparameters of the bidirectional long- and short-term memory neural network (BiLSTM) are optimized using the SSA algorithm, and the final power prediction is obtained. Simulation experiments show that the model effectively improves prediction accuracy and verifies the effectiveness of the prediction model. The model uses PCA to reduce the dimensionality of the data, SSA to optimize VMD for noise removal, and BiLSTM for prediction. The results show that the model outperforms other models in terms of prediction accuracy, with high R² values and low RMSE, MAE, and MAPE. The model is tested on real-world data from offshore wind farms, and the results demonstrate its effectiveness in predicting wind power output. The study highlights the importance of combining dimensionality reduction, noise removal, and advanced neural networks for accurate wind power forecasting. The proposed method is expected to contribute to the safe and stable operation of offshore wind farms by improving the accuracy of wind power predictions.