海上交通的船舶异常行为挖掘识别分析  

Vessel Abnormal Behaviors Mining Recognition Analysis on Maritime Traffic

作　　者：姜佰辰[1] 关键[2] 周伟[3] 何友[2] JIANG Bai -chen ZHOU Wei SUN Lu GUAN Jian(Graduate Students Brigade, Navy Aeronautical and Astronautical University, Yantai Shandong 264001, China Department of Electronic Information, Navy Aeronautical and Astronautical University, Yantai Shandong 264001, China Department of Information Fusion, Navy Aeronautical and Astronautical University, Yantai Shandong 264001, China)

机构地区：[1]海军航空工程学院研究生管理大队,山东烟台264001 [2]海军航空工程学院电子信息系,山东烟台264001 [3]海军航空工程学院信息融合研究所,山东烟台264001

出　　处：《计算机仿真》2017年第6期329-334,共6页Computer Simulation

基　　金：国家自然科学基金(61501487;61531020;61471382;61401495);山东省自然科学基金(2015ZRA06052)

摘　　要：针对大数据背景下海上交通的船舶数据繁杂,利用航迹数据挖掘船舶航道进而发现船舶异常行为和目标意图识别问题,通过烟台港至大连湾实测数据采用了一种构建混合高斯模型及主成分分析模型分析目标轨迹行为的分析方法。分析了当下海上交通船舶对于异常行为的判别原则,并对疑似异常航迹信息进行预处理的基础上利用混合高斯模型拟合航迹曲线,将航迹转化为可用于分析的概率模型,借助主成分分析方法得到特征值判别,得到异常行为识别准确率。实验最后通过判读未知航迹信息准确发现潜在船舶异常行为,并在数据层面上定量地说明了方法的有效性,以此可以为有关海上监管部门提供相关预警。In order to recognize the abnormal behavior as well as the intention for maritime transport vessels under the background of the complex track data,by mining vessel channel, a behavior analysis method is presented. Based on the data measured from Yantai to Dalian port, an analysis method involves the Gaussian mixture models and principal component analysis. Discrimination principles of maritime transport vessels are analyzed, while pre - processed track information are fitted by using Gaussian mixture model, the track is changed into probabilistic model and as a basis for preliminary determination of abnormal trajectory. By means of principal component analysis, eigenvalues are determined to improve the recognition accuracy. Finally, potential abnormal behaviors are measured by means of interpreting unknown track information. Measurement effectiveness is surely illustrated on the level of quantitative data, thus early warnings can be provided to the relevant regulatory authorities to improve the maritime traffic.

关 键 词：异常行为识别 轨迹分析 混合高斯模型 主成分分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]