Research And Application Of The Mining Framework Of Trajectory Data For Vessels

With the development of world economy,the water transportation is increasingly heavy,the water traffic is more complicated.In order to ensure the safety of property and to prevent illegal activities for state sovereignty,strengthening the regulation of water traffic is very important.Traffic flow is mainly obtained by visual observation and radar observation,which is inefficient and lack the details.Besides,the method of manual supervision have long been very inefficient.In addition,misinformation is easily caused by regulators' decreased attention and fatigue.The paper is intended to acquire information such as vessels' motor patterns and maritime anomaly detection,based on massive historical trajectories saved by AIS,consequently,aid the decision-making for waterway construction and path planning,raise the efficiency of maritime management,lower the risks,improve efficiency of traffic and save transportation costs.The paper take vessels' trajectory as research objects,and studies the mining method by statistical analysis and cluster analysis.The works are as follows.(1)Summary of preprocessing including removing noises is made,besides,the identifier packaging problem is found after analyzing the abnormal phenomenon of checksum error during the collection of AIS data from VTS in Tianjin.A method to screen by circularly checking identifier is given.Practice proves that the new method can avoid the problem and raise the quality of AIS message collection.(2)A method of determining threshold of the Douglas-Peucker algorithm is presented by mathematical statistics and a concept called Length Sum of sub-trajectories'.Practice proves that the new method can compress data and still reserve the main feature of the raw data,therefore it can be used for partition of vessels'trajectory(3)In order to analyze vessels' trajectories which is unevenly distributed in the area,a method to adaptively determine parameter of DBSCAN algorithm based on statistical analysis for 'Core-distance' and amount of noise is presented,thus improve the algorithm in a way hierarchically.The analysis of an example indicates that the new algorithm can distinguish vessels' trajectory and gathering them which are similar,therefore it can be used for modelling of vessels' behavior.(4)A framework based on "Partition by Douglas-Peucker—Similarity by structured distance—adaptively hierarchical DBSCAN/Spectral Clustering/Kernel Density Estimation" and its applications for detecting anomaly vessels based on sample data is given.Applying VisualStudio2013,MatlabR2014a,ECDIS,paper conduct the analysis experiment by using AIS data in Qiongzhou strait,besides,a vessel detecting system is built to make the simulation of vessel monitoring and its results prove the validity in the field of maritime supervision and management,besides,the framework shows its applicability to find regularities of distribution for the massive disordered vessels'trajectory,and to timely recognition abnormal vessels.