Research On Vessel Trajectory Prediction Method Based On Recurrent Neural Network

Nowadays,with the deeply development of world economic globalization,economic trade through maritime traffic gradually becomes a popular trend,the maritime transportation is also getting busier,traffic is growing rapidly,water bearing burden,the number of ocean vessels and the density of vessels in important waters and waterways are increasing,causing frequent maritime traffic accidents.Therefore,it is imperative to reduce the maritime traffic accidents and improve the maritime navigation safety,in which the key task is to implement the trajectory prediction of vessels.In the application of maritime search and rescue as well as customs anti-smuggling,vessel trajectory prediction is a key problem.Accurate and efficient trajectory prediction is of great significance for reducing maritime traffic accidents and improving the decision-making level of relevant maritime departments and vessel traffic service system.In this paper,the problem of marine vessel trajectory prediction is studied.A large number of historical vessel voyage data are collected through Automatic Identification System(AIS),and a data-driven vessel trajectory prediction method is proposed.This method divides the trajectory prediction task into two stages: data preprocessing and trajectory prediction.In the data preprocessing stage,the existing problems of AIS data are analyzed.Firstly,the interpolation preprocessing is carried out based on the linear interpolation method to solve the problem of uneven data distribution caused by sparse AIS data and partial data loss.Then,a similarity measurement method based on Symmetrized Segment-Path Distance(SSPD)is designed to eliminate the influence of redundant data and noise in the AIS data and obtain the reliable correlation data set required by the prediction model.In the trajectory prediction stage,two kinds of prediction model are constructed based on Recurrent Neural Network(RNN),namely the Long Short-Term Memory(LSTM)and the Gated Recurrent Unit(GRU),the reliable data set is trained as the input of the model,and then finally realizes the accurate and efficient latitude and longitude coordinates of future prediction of the vessel.A large number of AIS data are compared and the experimental results verify the effectiveness of the proposed vessel trajectory prediction method in the application of the practical problem.