基于聚类距离计算的船舶轨迹异常检测方法

A Trajectory Data Anomaly Detection Method Based on Trajectory Cluster Distance

针对船舶自动识别系统(Automatic Identification System)数据,提出了一种基于聚类距离计算的异常检测方法,该方法在船舶AIS历史轨迹点聚类中提取移动点聚类重心向量和停止点聚类的采样集,定义了聚类相对距离和聚类角度距离两种聚类距离用来衡量轨迹点在距离上、航向和航速上与聚类的相似性。通过计算目标轨迹点与重心向量和采样点之间的聚类距离,可检出AIS数据中的距离异常点、航向异常点和航速异常点。基于真实AIS数据集组织了实验,验证了方法的有效性。

In order to analyze trajectory data from automatic identification system,a new method based on trajectory cluster distance calculation is proposed for the ship trajectory anomaly detection.The method proposes gravity vectors and sample stop points from the trajectory clusters generated from history data,introduces cluster relative distance and cluster angular distance to measure the similarity between target trajectory points and cluster characteristics.By calculating the distance between target trajecto⁃ry and clusters,the proposed method can detect distance anomaly,heading anomaly and speed anomaly.Experiments are made based on real AIS data,which shows the effectiveness of this method.