摘要
传统的密度峰值聚类算法不仅具有较高的计算复杂度且未考虑路网固有的拓扑结构,无法衡量各路段之间的关联关系。针对这一问题,提出基于图密度峰值聚类算法的出行热点路段发现。该算法将交通路网用图模型结构,然后以各路段为基本单元计算局部密度及高局部密度距离并画出决策图找出聚类中心,最后结合实际区域的兴趣点分析该聚类簇成为热点路段的潜在可能。借助于图模型表达形式的优势,该算法不仅可以大幅度提升算法的计算复杂度,而且可以更加准确合理的找出热点路段。通过在滴滴‑成都轨迹数据集上的实验表明,图密度峰值聚类算法具有更高的热点路段发现精度,并且在计算效率上有大幅度提升。
The traditional density peak clustering algorithm not only has high computational complexity,but also does not consider the inherent topology of the road network.Hence,it cannot measure the intrinsic relationship between the various road segments.Aiming at this problem,this paper proposes a travel hotspot road segments discovery based on GDPC algorithm.The GDPC algorithm uses a graph model structure for the traffic road network,then uses each road segment as the basic unit to calculate the local density and the minimum high local density distance.Afterwards,the algorithm draws a decision diagram to find the cluster center,and finally combines the points of interest in the actual area to analyze the potential of the cluster to become a hot spot.With the advantage of the graph-based representation,the GDPC algorithm can not only greatly improve the computational complexity compared with traditional algorithms,but also find hot spots more accurately and reasonably.Experiments on the Chengdu Didi dataset show that the GDPC algorithm is more reasonable,and achieves a significant improvement in computational efficiency.
作者
王少帆
魏福豪
黄世雨
尹宝才
WANG Shaofan;WEI Fuhao;HUANG Shiyu;YIN Baoc(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)
出处
《中国传媒大学学报（自然科学版）》
2023年第1期31-38,共8页
Journal of Communication University of China：Science and Technology
基金
北京市自然科学基金项目(4222021)
国家自然科学基金项目(U1811463)。
关键词
智能交通
出行热点
图密度峰值聚类
热点发现
滴滴数据集
intelligent transportation
travel hotspots
graph density peak clustering
hotspots discovery
DiDi dataset
作者简介
王少帆(1982-),博士,副教授,主要从事模式识别、智能交通研究。Email:wangshaofan@bjut.edu.cn;通讯作者:尹宝才(1963-),博士,教授,主要从事数字多媒体、智能交通、多功能感知技术研究。Email:ybc@bjut.edu.cn。
