Design And Implementation Of Ship Trajectory Prediction Model Based On Machine Learning And AIS Data

In today’s increasingly developed economy and technology,the connection between countries has become more and more close.Marine transportation is one of the main ways of international communication.With the development of shipping industry,maritime traffic accidents happen from time to time.Therefore,the research on the trajectory of ships has become more and more important.Through trajectory research,the ship’s route can be analyzed,and the sailing status and destination of the ship can be monitored,the abnormal trajectory of the ship can be found in time,and the safety of maritime traffic can be guaranteed.Based on the data obtained by Automatic Identification System(AIS)and machine learning method,this thesis analyzes the data structure and characteristics of the AIS data,and extracts the ship’s trajectory,and designs and implements a ship trajectory prediction model by combining clustering and regression algorithm.The main work and innovation of this thesis are as follows:(1)Analyze and study the AIS data,extract the trajectory of the ship from it,and process the missing values and abnormal data in the trajectory.For the interpolation of latitude and longitude,this thesis proposes a two-way weighted average interpolation method.Through bidirectional interpolation of the two endpoints of the trajectory segment containing missing values,the interpolation sequence of missing values to the other end point is recursively obtained.After that,the two interpolation sequences of missing values are weighted and averaged to get the desired interpolation value of missing data.Experiments show that the method has high precision for longitude and latitude interpolation when there are many missing data.(2)A method of ship trajectory clustering based on regional similarity is proposed.In this method,the navigation area is gridded,and the original ship trajectory sequence is changed into grid sequence.By judging the similarity of grid sequence,the ship trajectory with similar grid sequence is grouped into one class,so as to achieve the purpose of ship trajectory clustering.(3)Through the improvement of the Seq2 Seq model,multiple trajectory points in the future were predicted.This thesis improves the Seq2 Seq model,uses GRU as the model unit,and combines the output of the encoder with the last element of the input sequence as the input of the decoder,and applies it to the prediction of trajectory points at multiple consecutive moments.By comparing with other model experiments,this model can improve the prediction accuracy of continuous trajectory points.(4)Clustering and regress algorithm is proposed to improve the accuracy of trajectory prediction.By combining trajectory clustering and trajectory prediction,different trajectory sets obtained by trajectory clustering were trained for modeling.The difference between samples of the same type of trajectory set is small,and regression prediction on it can improve the accuracy of trajectory prediction.