Abstract: Passenger quantity of subway normally varies significantly by different time period like peak and non-peak hours. Reasonable classification of operation intervals is essential for adaptable adjustment of traffic plan for peak and non-peak hours. Actually, the classification method which is manually set based on experiences, is subjective and lack of accuracy. Taking 10 minute as a unit time interval, the daily operating period (6∶00~23∶00) can be divided into 102(10-min)time intervals. This article proposes affinity propagation algorithm merge time samples into different categories, together with arriving passenger volume alongside stations which are used as describing variables. Clustering validity indexes such as CH Hart and IGP are introduced to examine clustering result, so optimal operation intervals classification and switch time are finally confirmed. The study of Tianjin subway 2th line indicates that operation intervals classification based on clustering algorithm could respond the fluctuation of real passenger quantity more accurately. On the base of that, the optimized traffic plan causes obvious decrease of passenger average waiting time.

Key words: subway operation intervals classification, arriving passenger volume alongside stations, affinity propagation cluster, clustering validity assess, passenger average waiting time

摘要： 地铁在每天不同时段客流量差异较大,运营时段的科学划分,是低峰与高峰列车运行计划合理交替的前提。目前地铁运营时段划分主要依据人工经验,主观性强且难以保证精度。以10min为时间间隔,把全天运营时间6∶00～23∶00分为102个时间点样本,将地铁沿线各车站每个时间点的进站客流量作为样本描述变量。采用近邻传播聚类算法将各时间点归并为不同类别,并引入CH、Hart以及IGP等聚类有效性评估指标对聚类结果加以检验以确定最优类别数,从而得到运营时段的最优划分方案和最佳时段分割点。天津地铁二号线实例研究表明,基于近邻传播聚类算法得到的运营时段划分结果更能体现实际客流需求波动特性,在此基础上优化行车计划后,旅客等待时间明显下降。

关键词: 地铁运营时段划分, 进站客流量, 近邻传播聚类, 聚类有效性评估, 旅客平均候车时间