船舶行为异常检测研究进展  

Research Progress of Ship Behavior Anomaly Detection

作　　者：陈影玉 杨神化[1] 索永峰[1] CHEN Yingyu;YANG Shenhua;SUO Yongfeng(Navigation College,Jimei University,Xiamen 361021,Fujian,China)

机构地区：[1]集美大学航海学院,福建厦门361021

出　　处：《交通信息与安全》2020年第5期1-11,共11页Journal of Transport Information and Safety

基　　金：国家自然科学基金项目(51879119);福建省自然科学基金项目(2018J01536、2018J01484、2018J01485)资助。

摘　　要：船舶行为异常检测对于海上安全、海域的智能监管具有重要意义。异常检测算法不能满足轨迹大数据挖掘在实时性、准确性和鲁棒性等方面的需求。将异常行为进行分类,分析目前几类主要的异常检测方法:统计分析在对数据分布做出正确假设时根据概率分布获取异常情况,确定合适的异常阈值较为困难;预测法基于对历史数据的了解程度,易受多种因素影响;机器学习依赖数据特征、计算复杂度高。基于此,总结可能提高统计分析、机器学习和预测法检测效果的方法,指出将在线实时检测引入船舶检测,并展望数据处理、轨迹表示、挖掘分析和情境语义在异常检测中的可能研究方向。Abnormal detection of ship behaviour is of great significance to maritime safety and intelligent supervision.The existing anomaly detection algorithm can not meet the real-time,accuracy,and robustness requirements of trajectory big data mining.By analyzing main methods of abnormal detection,statistical analysis,when making correct assumptions about data distribution,obtain abnormal conditions according to the probability distribution,it is difficult to determine an appropriate anomaly threshold.The prediction method is based on the knowledge of historical data and is easily affected by many factors.Machine learning relies on data characteristics and has high computational complexity.Based on these,the methods that may improve the detection effect of statistical analysis,machine learning,and predictive methods are summarized.It points out that the introduction of online real-time detection into ship detection.The possible directions of data processing,trajectory representation,mining analysis,and contextual semantics in anomaly detection are expected.

关 键 词：交通控制 船舶行为 异常检测 轨迹挖掘和分析 异常行为 

分 类 号：U675.7[交通运输工程—船舶及航道工程]