基于DBSCAN和iForest算法的船舶异常行为分析

Analysis of Vessel Abnormal Behavior Based on DBSCAN and iForest Algorithms

为更好地从船舶自动识别系统(AIS)数据中挖掘信息,科学地感知水上交通态势,针对异常船舶进行识别,论文提出一种基于改进的密度空间聚类算法(DBSCAN),结合孤立森林算法(Isolation Forest,iForest)在包含大量不相关船舶分布数据中选取数据集内在分布规律和聚类效果的变化来发现异常;同时针对船舶位置(LAT,LON)及对地航速(SOG)进行聚类,将速度与船舶密度结合,对船速分簇后结合地理领域知识和iForest构建的异常边界,更好地发现船舶异常行为。以北部湾船舶AIS数据为例,应用该文的算法,发现异常并进行分析,可以简化异常识别监控方式,为海事检测部门提供良好的船舶监控环境,降低监控人员的工作强度,为航道建设部门,应急处置部门及海事监管部门等提供合理建议。

In order to discover maritime traffic situation scientifically from automatic identification system(AIS)information in the data mining for the abnormal vessels identification better,in this paper,an improved density-based spatial clustering of applications with noise(DBSCAN),combined with the isolation forest algorithm(iForest)is proposed. The inner distribution law and the change of clustering effect of the dataset containing a large amount of uncorrelated vessel distribution data are selected to find the anomalies by this method. At the same time,the vessel location(LAT,LON)and speed over ground(SOG)are clustered to combine the speed with the vessel density. After clustering the vessel speed,the abnormal boundary constructed by geographical knowledge and iForest is adept at finding the distinct abnormal behavior of the vessel. Taking the AIS data of Beibu Gulf as an example,the algorithm in this paper can be used to find and analyze anomalies,which can simplify the monitoring method of anomaly identification,provide a good vessel monitoring environment for maritime detection departments,relieve the work burden of monitoring personnel,and provide reasonable suggestions for channel construction departments,emergency response departments and maritime supervision departments.