下车信息对公交乘客精细化分类影响研究

中山大学智能工程学院,广东广州510006

交通运输工程;公共交通;乘客分类;下车信息;组合聚类;出行特征

Impact of bus alighting information on fine classification of transit riders
LI Jun,OU Jingyi,and ZHAO Wenting

School of Intelligent Systems Engineering, Sun Yat-sen University, Guangzhou 510006, Guangdong Province, P. R. China

transport engineering; public transit; rider classification; bus alighting information; combined clustering;travel characteristics

DOI: 10.3724/SP.J.1249.2023.01109

备注

目前广泛采用的公交一票制缺乏下车信息,利用一票制数据得到的乘客分类需要进行精细化分类效果评估.本研究利用包含下车站点的大规模公交票务数据,对比分析了有无下车信息下的分类模型,探究下车信息缺失对乘客分类的影响.采用分类指标和组合聚类方法对公交乘客进行精细化分类,选取中国北京市路面公交连续28d的乘客上下车数据作为实例,分析下车信息及分类指标对分类结果的影响.结果表明,有无下车信息的2个模型均能实现乘客的有效分类,分类结果都能体现每类乘客的时空活动规律.其中,包含下车信息的分类模型能够识别具有明显出行距离特征的小样本群体,如占比0.25%的长距离出行乘客等特殊群体;而缺少下车信息的分类模型类别平均占比标准差远小于包含下车信息的分类模型,分类结果相对均衡,且各类别乘客在多维度上的规律差异显著,更能体现分类的宏观特征.
The widely used bus one ticket system lacks off boarding information, and the passenger classification based on the dataset of one-ticket bus system needs to be refined to evaluate the classification effect. This study uses the large-scale bus ticket data with alighting information to compare and analyze the transit rider classification models with and without the alighting information to explore the impact of the missing alighting information on the fine classification of transit riders. The classification indexes and combined clustering method are adopted to refine the classification of bus passengers. A case study of Beijing conventional buses which contain boarding and alighting data of 28 consecutive days is presented and used to analyze the impact of bus alighting data and classification indexes on the transit rider classification results. The result indicates that both models successfully complete the fine classification of the riders and reflect the spatio-temporal characteristics of each type of riders. The model with alighting information is able to identify the distance-concerned type with few samples such as the long-distance transit riders accounting for 0. 25%. The standard deviation of the average proportion of categories in the model without alighting information is much smaller than that in the model with alighting information. The results obtained from the model without alighting information are more balanced, and the regularity of various categories of passengers in multiple dimensions is significantly different, which can better reflect the macro characteristics of classification.
·