考虑多维特征的船舶轨迹分层聚类算法  

Hierarchical clustering algorithm for ship trajectories considering multi-dimensional features

作　　者：苏俊杰 兰培真[1,2] SU Junjie;LAN Peizhen(Maritime Traffic Safety Institute,Jimei University,Xiamen 361021,Fujian,China;National Engineering Laboratory for Emergency Information Technology of Traffic Safety,Xiamen 361021,Fujian,China)

机构地区：[1]集美大学海上交通安全研究所,福建厦门361021 [2]交通安全应急信息技术国家工程实验室,福建厦门361021

出　　处：《上海海事大学学报》2022年第4期30-36,共7页Journal of Shanghai Maritime University

摘　　要：为准确聚类复杂的船舶轨迹和辨识隐蔽轨迹簇,提出一种考虑多维特征的船舶轨迹分层聚类算法。用核心萤火虫算法(core firefly algorithm,CFA)解决具有噪声的基于密度的空间聚类(density-based spatial clustering of applications with noise,DBSCAN)算法的邻域查询冗余和参数敏感问题,并在传统船舶轨迹聚类特征的基础上引入水域环境、轨迹线型和时隙特征来分层建立轨迹相似性度量指标,最终实现轨迹的逐层递进聚类。以厦门港及其附近水域的AIS数据验证算法的有效性,检验结果表明:船舶轨迹由算法聚类为9簇;簇内动态时间规整(dynamic time warping,DTW)距离均值为5.199,簇间DTW距离均值为18.032;聚类结果符合实际的船舶交通流情况,聚类准确率为91.50%。可见,提出的算法相比其他常用的轨迹聚类算法能更有效地辨识轨迹地理分布和船舶运动特征的异同,更容易发现隐蔽的轨迹簇。由提出的算法聚类的同簇轨迹,其船舶运动特性更相似,聚类结果可为船舶交通流特性分析及船舶行为模式识别等提供典型的轨迹样本。In order to accurately cluster the complicated ship trajectories and identify the hidden trajectory clusters,this paper proposes a hierarchical clustering algorithm for ship trajectories considering multi-dimensional features.The core firefly algorithm(CFA)is used to solve the neighborhood query redundancy and parameter sensitivity problems of DBSCAN(the density-based spatial clustering of applications with noise)algorithm;based on the traditional ship trajectory clustering features,the water environment,the trajectory line shape and the time slot feature are introduced to establish in layers the similarity indices of trajectories,and finally the progressive clustering of trajectories is realized layer by layer.The effectiveness of the algorithm is verified by AIS data of Xiamen Port and the nearby waters,and the test results are as follows:the ship trajectories are clustered into 9 clusters by the algorithm;the mean value of dynamic time warping(DTW)distance in a cluster is 5.199,and the mean value of DTW distance between the clusters is 18.032;the clustering results are consistent with the actual ship traffic flow situation,and the accuracy of the clustering is 91.50%.It is clear that,compared with other common trajectory clustering algorithms,the proposed algorithm can more effectively distinguish the geographical distribution of ship trajectories and the differences of ship motion characteristics,and it is easier to find hidden trajectory clusters.The trajectories in a cluster clustered by the proposed algorithm are of more similar ship motion characteristics,so it can provide typical trajectory samples for applications such as ship traffic flow characteristic analysis and ship behavior pattern recognition.

关 键 词：船舶轨迹聚类 相似性度量 层次聚类 核心萤火虫算法(CFA) 具有噪声的基于密度的空间聚类(DBSCAN) 

分 类 号：U675.79[交通运输工程—船舶及航道工程] TP18[交通运输工程—船舶与海洋工程]