ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2014, Vol. 51 ›› Issue (12): 2653-2662.doi: 10.7544/issn1000-1239.2014.20131124

• 人工智能 • 上一篇    下一篇

基于客流数据的区域出行特征聚类

冷彪,赵文远   

  1. (北京航空航天大学计算机学院 北京 100191) (lengbiao@buaa.edu.cn)
  • 出版日期: 2014-12-01
  • 基金资助: 
    基金项目:国家“八六三”高技术研究发展计划基金项目(2013AA01A601);国家自然科学基金项目(61103093)

Region Ridership Characteristic Clustering Using Passenger Flow Data

Leng Biao,Zhao Wenyuan   

  1. (School of Computer Science and Engineering, Beihang University, Beijing 100191)
  • Online: 2014-12-01

摘要: 区域功能发现对完善城市规划有着重要的指导意义.区域居民的出行特征提取与发掘可以作为建立模型分析区域功能的数据支撑.随着智能交通技术在轨道交通系统的应用,大量蕴含行人移动性和出行目的地信息的客流数据被采集得到,发现客流数据与地铁站相关区域功能有紧密联系.从地铁客流数据中提取出乘客出行模式和地铁站客流模式,并以此为基础建立概率图模型,实现了区域出行特征聚类.首先,以地铁客流数据为基础提取了乘客出行模式和地铁站客流模式,发现地铁站客流集中性和潮汐性的特性,能在一定程度上反映地铁的区域功能.然后,采用了文本分析领域经典的概率图模型,建立基于潜在狄利克雷分配(latent Dirichlet allocation, LDA)主题模型的地铁客流出行特征聚类模型,将具有出行规律相似性的地铁站聚类在一起.最后,通过分析聚类实验结果,发现在不同客流峰段内的区域功能和相互客流关系.

关键词: 出行特征聚类, 人类移动性, 概率图模型, 地铁客流, 数据挖掘

Abstract: Region function is an integral part of urban planning. The extraction and mining of ridership characteristic can be regarded as data support of region function recognition. The advance of intelligent transportation technology in metro system enables the collection of spatial-temporal passenger flow data, which conveys human mobility and indicates the similarity between metro stations, also closely related to the region function during different periods. This paper discusses the ridership characteristic clustering using passenger trip pattern and metro station flow pattern extracted from metro passenger flow data. Firstly, we identify the passenger flow centrality and station tide flow from passenger trip pattern and metro station flow pattern, which imply the region function of metro stations. Secondly, by discovering the similarity between region cluster and text analysis, we take advantage of the classical probabilistic graphical model and propose a novel LDA-based region ridership characteristic clustering model, allocating metro stations with similar ridership characteristic into the same region. Thirdly, the experimental results show the passenger flow relationship among regions and recognize the region functions during different periods. The analysis of clustering results gives us a good understanding of how passenger flow circulates during different periods and may enables many valuable services like network design and crowd evacuation.

Key words: ridership characteristic clustering, human mobility, probabilistic graphical model, metro passenger flow, data mining

中图分类号: