都市快轨交通·第30卷第2期2017年4月
doi: 10.3969/j. issn. 1672 -6073.2017.02. 010
《学术探讨
基于刷卡数据和高斯混合聚类的
地铁车站分类
岳真宏1，陈峰1,2，王子甲1,2，黄建玲3，汪波3
（1.北京交通大学土木建筑工程学院，北京100044；2.北京市轨道交通线路安全与防灾
工程技术研究中心，北京100044；3.北京市交通信息中心，北京100161）
摘要：合理的城市轨道交通车站分类对车站的规划设计及客流预测有重要作用。基于刷卡数据提取出行时间、频次、车票类型等反映车站客流特性的若干变量，运用主成分分析法（PCA）和高斯混合模型（GMM)进行车站聚类。该聚类方法不仅可以识别车站类别，同时可以根据后验概率确定混合类型的车站。以北京地铁为例.将全网233个车站分为4类，利用地理信息系统（GIS)工具可视化分类结果，并叠加地理信息描述各类车站的特征，直观地展示了部分混合性质的车站。与K-均值聚类结果比较显示，CMM方法可以更好地解释多种特性混合的车站类型。
关键调：地铁；车站分类；刷卡数据；高斯混合模型；地理信息系统
中图分类号：U231.4
文献标志码：A
文章编号：1672-6073(2017)02-0048-04
Classifications of Metro Stations byClustering
SmartCardDataUsingtheGaussianMixtureModel
YUE Zhenhong', CHEN Feng'2, WANG Zijia',2, HUANG Jianling",WANG Bo
(1. School of Civil Engineering, Beijing Jiaotong University, Beijing 100044;
2. Beijing Engineering and Technology Research Center of Rail Transit Line Safety and Disaster Prevention, Beijing 100044; 3. Beijing Transportation Information Center, Beijing 100161)
Abstract: Reasonable classification of urban rail transit stations is of great significance to station planning, designing and rider ship forecasting. This research focused on the characteristics of station ridership and proposed travel time, frequency, ticket type and other variables extracted from smart card data. Accordingly, Principal Component Analysis (PCA) and the Gaussian Mix-ture Model (GMM) were used in clustering and classifying stations. The metro stations are classified into arbitrary types and mixed types using the posterior probability generated by GMM, which revealed to what extent and by which arbitrary types a mixed station was mixed. Beijing was selected as a case and 4 clusters were determined for the 233 stations on Beijing metro network. The classification results were visualized by GIS and all types of stations were characterized by superposing geographic information, meanwhile, parts of mixed stations were presented intuitively. At last, comparative analysis was conducted be tween GMM and K means algorithm and the results showed that GMM can explain mixed stations with various characteristics preferably
Keywords: metro; station classification; smart card data; Gaussian Mixture Model; geographic information system
城市轨道交通车站的分类对车站个性化设计有重收稿日期：2016-05-03修回日期：2017-01-09
第一作者：岳真宏，男，硕士研究生，主要从事城市轨道交通规划与
设计，14121208@bjtu.edu.cn
通信作者：王子甲，男，博士，讲师，主要从事城市轨道交通规划与设
计,zjwang@ bjtu.edu.cn
基金项目：中央高校基本科研业务费专项资金资助（2016YJS102) 48URBAN RAPID RAIL TRANSIT
要意义。目前，国内对于站点分类的研究主要集中在以节点导向的分类和以场所导向的分类[1]。大多数分类仅仅基于定性分析[2-4]，而定量分析也有很大的局限性[56]。首先采用的数据量有限，不能全面反映车站的特性；其次分类结果往往是单纯的几类车站，不能反映车站可能同时属于多种类型的实际。这种间题