Abstract: To obtain the passenger volume or passenger flow direction from public traffic intelligent card (IC) data, the direction of each bus work shift should be obtained first. The approach chooses the clustering analysis method to obtain the station passenger volume of single bus work shift, and measures the similarity of two time series (station passenger volume of IC data and historical station passenger volume) by the similarity discovery techniques. The experiment result shows when the difference of passenger regulation in two running directions exists, the similarity discovery methods (correlation coefficient and dynamic time warping) performs better and the two methods can be both used in bus running direction identification which route has apparent different passenger regulation in two directions.

Key words: intelligent transportation systems (ITS), running direction identification, time series similar discovery, cluster analysis, public traffic intelligent card data

摘要： 在公交IC卡数据挖掘中，为了获取乘客流量及流向等信息，需要获知每个班次的运行方向.本文通过对公交IC卡数据的聚类分析，将IC卡数据解析成单班次站点客流数据，利用基于时间序列的相似性测量算法（相关性测量及动态时间扭曲法），测量单班次数据与经验数据的相似性，从而获取班次运行方向.研究结果表明,在线路客流方向性差别明显时，相似性测量方法精度较高.且经过数据聚类后，相关性测量法与动态时间扭曲法在计算精度与运算速度方面表现相近，适用于客流方向性差别较明显的公交线路.

关键词: 智能交通, 运行方向判断, 时间序列相似性测量, 聚类分析, 公交IC卡