Ship Trajectory Prediction Model And Application Based On Improved Bi-LSTM

Against the backdrop of economic globalization,the maritime transportation industry is booming,with cargo transportation accounting for nearly 90% of global transportation volume.Every day,over 100,000 vessels sail on the sea,and vessels are gradually becoming larger and more specialized.Ensuring the safety of vessel navigation is of great significance to maritime safety.Improving the accuracy of ship trajectory prediction can ensure the safety of ship navigation.By using deep learning prediction models to predict ship trajectories,the location information of vessels in the future can be predicted in advance,which enables proactive operation and warning,thus ensuring the safety of ship navigation.With the widespread deployment of AIS devices,researchers have gained access to massive amounts of ship trajectory data,which provides favorable conditions for fully exploring the features of vessel data.Therefore,based on AIS data,this study uses deep learning models to predict,research,and analyze ship trajectories and conduct long-term trajectory prediction using the model to plan vessel navigation and reserve sufficient coordination time for drivers and VTS staff,thereby ensuring the safety of vessel navigation.The main work includes the following aspects:(1)The full-year 2020 data and January 2021 data from Tianjin Port were selected to provide accurate data for the experiment.Firstly,the raw AIS data was decoded to form a preliminary dataset.Then,abnormal data values were removed,and the dataset was processed using cubic spline interpolation,resampling,and mean-variance normalization.These procedures enabled the AIS data to become a smooth and continuous sequence with the same dimensions and fixed intervals,thereby improving the quality of the AIS data.Ultimately,the required dataset for the study was obtained.(2)To ensure the safety of ship navigation and effectively improve the accuracy of ship trajectory prediction,this study proposes an improved deep learning prediction model that combines the Radam optimization algorithm with the Bi-LSTM prediction model.The traditional prediction model based on the Adam optimization algorithm often suffers from high variance in the learning rate and slow convergence during the early stages of training,which can easily lead to local optimal solutions.By applying the proposed model to ship trajectory prediction,the effective prediction of ship positions is achieved.(3)In order to further improve the accuracy of ship trajectory prediction,this study combines the Lookahead optimization algorithm with the Bi-LSTM model to construct an improved prediction model.The proposed model takes advantage of the Lookahead optimization algorithm to specify an internal loop optimizer and update the fast and slow weights,which effectively enhances the prediction accuracy.(4)Based on the prediction of ship trajectories,the improved model is used to achieve long-term prediction of multiple ships.Through the deduction and arrangement system,the behavior of ships deviating from the route can be monitored and warned in a timely manner,eliminating the danger of ship navigation and ensuring the safety of ship navigation.At the same time,it also provides reference and theoretical support for the decision-making of maritime regulatory authorities.In this study,deep learning prediction models were used to predict the "longitude" and "latitude" features of ships.The results show that the Radam-enhanced Bi-LSTM model and the Lookahead-enhanced Bi-LSTM model are suitable for predicting ship trajectories,and both improved models have smaller root mean square error(RMSE)and mean absolute percentage error(MAPE)than RNN,LSTM,and Bi-LSTM neural network prediction models based on the Adam optimization algorithm.Furthermore,the prediction accuracy of the Lookahead-enhanced Bi-LSTM model is further improved compared to the Radam-enhanced Bi-LSTM model.The prediction results demonstrate the feasibility of using deep learning prediction models for ship trajectory prediction,providing a reference for the future spatial and temporal positions of ships sailing in the channel,and preparing collision avoidance warning data,ensuring the safety of ship navigation,and providing theoretical support for maritime regulatory decision-making.