划时区分段的动态时间规整算法  

Dynamic time warping algorithm for time zone segmentation

作　　者：康军[1] 许卫强 段宗涛[1] 黄山 Kang Jun;Xu Weiqiang;Duan Zongtao;Huang shan(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《计算机应用研究》2020年第11期3330-3333,共4页Application Research of Computers

基　　金：陕西省重点科技创新团队项目(2017KCT-29);陕西省重点研发计划资助项目(2019ZDLGY17-08,2019ZDLGY03-09-01)。

摘　　要：轨迹聚类是城市交通数据挖掘的重点之一,交通轨迹聚类算法是按照一定的相似度指标将轨迹划分成若干个类簇。在复杂的路网环境下,针对目前如DTW、SDTW等相似度计算方法准确性不高的问题进行了研究,提出了一种划时区分段的动态时间规整算法(SDTW+)进行相似度计算。该算法充分考虑了轨迹形状因素,能有效提高准确性。实验部分利用不同相似度算法,并结合层次聚类算法对实际车辆轨迹进行聚类,最终以平均轮廓系数和聚类成功率为评价指标,判断不同相似度算法的聚类效果。实验结果表明,采用所提算法相对于采用DTW、SDTW的平均轮廓系数分别提高33.86%、12.94%,同时聚类成功率也得到一定提高。Trajectory clustering is one of the key points of urban traffic data mining,whose algorithm divides the trajectory into several clusters according to certain similarity indicators.In the complex road network environment,aiming at the inaccuracy of current similarity calculation methods such as DTW and SDTW,this paper proposed a dynamic time warding algorithm(SDTW+)for time zone segmentation to calculate the similarity.This method took the shape of the trajectory into conside-ration and could effectively improve the accuracy.The experimental part used different similarity algorithms and combined the hierarchical clustering algorithm to cluster the actual vehicle trajectories.Finally,this paper selected the average contour coefficient and the cluster success rate as the evaluation index to judge the effect of clustering with different similarity algorithms.The experimental results show that the average contour coefficient of the proposed algorithm is 33.86%and 12.94%higher than that of DTW and SDTW,respectively,meanwhile with the clustering success rate improved.

关 键 词：城市交通 轨迹聚类 数据挖掘 相似度 DTW 

分 类 号：U121[交通运输工程]