尽管基于城市轨道交通自动售检票(automatic fare collection, AFC)系统采集的智能卡数据(smart card data, SCD)能够精准记录人们的出行时间和地点,但无法直接反映出行目的或活动类型. 本研究提出一种方法,将约束种子K-means算法的站点聚类与隐含狄利克雷分布(latent Dirichlet allocation, LDA)模型的客流出行目的挖掘相结合,以揭示城市轨道交通客流出行数据中的潜在活动模式. 首先,基于车站周边的人口特征、客流特征及兴趣点(points of interest, POI)分布,使用约束种子K-means算法将站点划分为8类:就业集聚型、居住集聚型、职住复合型、商业中心型、旅游景点型、综合枢纽型、对外枢纽型以及客流培育型. 其次,基于出站时间、活动时长、起点车站类型以及终点车站类型构建了LDA模型. 该模型成功识别出5类主要活动,分别为购物消费、工作、回家、休闲旅游及其他. 此外,这些模式进一步细分为若干子主题,每个子主题在时间和空间特征上具有显著差异,为深入理解节假日城市轨道交通客流出行行为提供了理论支持.
Abstract
While smart card data (SCD) collected from automatic fare collection (AFC) systems accurately records when and where people travel, they do not directly convey the trip purposes or activity types. In this study, we propose a method that integrates station clustering with an LDA model to uncover latent activities from urban rail transit passenger mobility data. First, we classified the stations into eight categories—employment,residential, mixed-use residential-employment, commercial centers, tourist attractions, composite hubs, external hubs, and ridership cultivation stations—using a constrained-seed K-means algorithm, based on demographic characteristics, ridership patterns, and the distribution of POIs around each station. Second, an LDA model is developed based on four key attributes: exit time, activity duration, origin station type, and destination station type. The model successfully identifies five primary activity types: shopping-related, work-related, home-related, tourism, and other. Furthermore, these patterns are divided into several subtopics, each distinguished by specific temporal and spatial characteristics, providing the theory support for deeply figuring out holiday travel patterns of urban rail transit passengers.
SUM L, LIUL, CHENX Y,et al .Exploration of the impact of built environment factors on morning and evening peak ridership at urban rail transit stations:a case study of Changsha,China[J].SAE International Journal of Sustainable Transportation,Energy,Environment,& Policy,2024,6(2):113-126.
HANB M, YUY R, XIZ,et al .Statistical analysis of urban rail transit operations worldwide in 2023:a review[J].Urban Rapid Rail Transit,2024,37(1):1-9.(in Chinese)
[5]
ALSGERA, ASSEMIB, MESBAHM,et al .Validating and improving public transport origin-destination estimation algorithm using smart card fare data[J].Transportation Research Part C:Emerging Technologies,2016,68:490-506.
[6]
FAROQIH, MESBAHM .Inferring trip purpose by clustering sequences of smart card records[J].Transportation Research Part C:Emerging Technologies,2021,127:103131.
[7]
KIME J, KIMY, KIMD K .Interpretable machine-learning models for estimating trip purpose in smart card data[J].Proceedings of the Institution of Civil Engineers - Municipal Engineer,2021,174(2):108-117.
[8]
HANG, SOHNK .Activity imputation for trip-chains elicited from smart-card data using a continuous hidden Markov model[J].Transportation Research Part B:Methodological,2016,83:121-135.
[9]
ZHAOZ, KOUTSOPOULOSH N, ZHAOJ H .Discovering latent activity patterns from transit smart card data:a spatiotemporal topic model[J].Transportation Research Part C:Emerging Technologies,2020,116:102627.
ZHAOZ, KOUTSOPOULOSH N, ZHAOJ H .Identifying hidden visits from sparse call detail record data[J].Transactions in Urban Data,Science,and Technology,2022,1(3/4):121-141.
CUIT, SUNC, JIAJ,et al .Enhanced K-means-type clustering algorithm with seeding constraints for the VANET[C]//2022 IEEE 95th Vehicular Technology Conference:(VTC2022-Spring). June 19-22,2022,Helsinki,Finland.IEEE,2022:1-6.
[14]
WAGSTAFFK, CARDIEC, ROGERSS,et al .Constrained K-means clustering with background knowledge[C]//Proceedings of the Eighteenth International Conference on Machine Learning.ACM,2001:577-584.
LIUL, LIUQ L .Classification and characteristics research of urban rail stations based on multi-source spatiotemporal data[J].Modern Urban Transit,2023(4):6-11.(in Chinese)
[17]
CALISTUSU C, ONYESOLUM O, DORISA C,et al .Exploring latent Dirichlet allocation (LDA) in topic modeling:theory,applications,and future directions[J].Newport International Journal of Engineering and Physical Sciences,2024,4(1):9-16.
[18]
WANGB X, MOUNTS M .Latent Dirichlet allocation mixture models for nucleotide sequence analysis[J].NAR Genomics and Bioinformatics, 2024, 6(3): lqae099.
[19]
QUANG TRUNGH,BUI X. A stochastic algorithm for solving the posterior inference problem in topic models[J]. TELKOMNIKA (Telecommunication Computing Electronics and Control),2022,20(5): 971.