Buoy Drift Prediction Modeling Based on Long Short-Term Memory Recurrent Neural

Accurately capturing the drift motion patterns of navigation buoys and reasonably predicting their short-term drift positions and distances are crucial for ensuring vessel navigation safety and collision avoidance. To this end, this study develops a buoy drift prediction model based on multi-source sensor data using a Long Short-Term Memory (LSTM) algorithm. Experimental results show that the Root Mean Square Error (RMSE) is 3.9959, the Mean Absolute Error (MAE) is 3.2126, the Mean Absolute Percentage Error (MAPE) is 1.9788%, and the drift distance error is 8.5 m. The proposed model achieves high prediction accuracy and outperforms the ARIMA benchmark. This LSTM-based trajectory prediction method provides important references for short-term buoy drift prediction and buoy management.