Research On Port Ship Trajectory Prediction Method Based On AIS Data

With the continuous development of economic globalization,maritime transportation has gradually become the main mode of transportation for trade between countries,and the number,size and scale of ships are also increasing.As an important hub of maritime transportation,ports are also the gathering point of land and water transportation.They have important functions of ship berthing,cargo loading and unloading,boarding and disembarking passengers and cargo provision,and play an important role in the field of land and water transportation and maritime transportation.However,at the same time of economic development,the incidence of port traffic accidents has also increased.Compared with ordinary waters,the channel in the port waters is narrow,the ships are crowded,and the navigation environment is more complex.Improving the safety of ships in the port waters has become an urgent problem to be solved.Ship trajectory prediction can effectively reduce the accident rate of ships in port waters,and ensure the safe operation of ships.It plays a vital role in ship collision early warning and path planning tasks.In order to ensure the safety of ships in port waters,this paper proposes a port ship prediction method based on AIS trajectory data,which is mainly composed of two parts.First,the AIS-R mode ground-based backup navigation system uses radar positioning to obtain the true trajectory information of port ships,so as to ensure the authenticity of AIS data.In view of the non-line-of-sight error problem generated in the process of radar positioning,A nonline-of-sight error suppression method based on Dense Net is proposed.Secondly,by comparing the characteristics of port waters and ordinary waters,a port ship trajectory prediction model based on Bi GRU-Seq2 Seq is proposed.The main research contents are as follows:(1)This paper proposes a non-line-of-sight error suppression method based on Dense Net.This method mainly includes two parts: feature enhancement method based on packet location and Dense Net network model.First,the least square method is used to eliminate the linear error value caused by the hardware equipment of the base station,and reduce the impact of the hardware equipment of the positioning system on the positioning accuracy.Then,each base station of the radar system is divided into positioning base station groups by arrangement and combination,and the trilateral positioning algorithm based on PSO(Particle Swarm Optimization)is used to initially locate each group of base stations.Secondly,the coordinate matrix containing the base station position coordinates,the target actual position coordinates and the target preliminary positioning coordinates is converted into a two-dimensional pixel matrix through the plane rectangular coordinate system,and the geometric position information of the positioning system is expressed through the feature image,and the feature image data set is constructed.Finally,the Dense Net network model is trained using the feature image data set,and the trained model can effectively suppress the nonline-of-sight error,so as to calculate more accurate positioning information.The simulation results show that this method is superior to other range-based positioning algorithms in the LOS/NLOS mixed environment.In the coastal scene with four radars,the positioning error of the algorithm is less than 180 m.(2)In this paper,a method of port ship trajectory prediction based on Bi GRUSeq2 Seq is proposed.This method improves and optimizes the data preprocessing part and prediction model part of ship trajectory prediction.First of all,this paper uses the motion state interpolation method based on vector function to transform irregular time series data into time series data with equal time interval to solve the problem of uneven distribution of original data acquisition time.Secondly,this paper uses the track division algorithm based on speed and course to divide the ship track into direct and curved tracks to solve the problem of complex ship tracks in port waters.Finally,Bi GRU-Seq2 Seq model is used to realize the future multi-step and multi-variable port ship trajectory task.The model is tested on the ship trajectory data set with a radius of10 km in New York Port,which proves the advantages of the model in terms of convergence and prediction accuracy.