Abstract: To discuss the characteristics of taxi spatiotemporal behavior, a cluster analysis of taxi behavior was carried out by adopting the non-negative matrix factorization from two dimensions of time and space. Firstly, the spatiotemporal unit was divided, the spatiotemporal behavior of taxis was represented by the spatiotemporal distribution characteristics of pickup points and the spatiotemporal behavior matrix was established. Then, the non-negative matrix factorization was adopted for taxi clustering. Finally, a case study of Guangzhou was carried out. The cluster results show that taxis can be divided into four categories under the influence of the spatial structure of Guangzhou, and taxis were less affected by the time dimension, but there are obvious differences in behavior in different periods. In addition, the taxis form a specific spatiotemporal behavior pattern that they operate around the commercial center in space and adjust the scope of operation to adapt to the travel activity demand in different periods of time. The study reveals the relationship among taxis spatiotemporal behavior, urban spatial structure and urban travel activities, which can provide reference for the development of taxis and urban management.

Key words: traffic engineering, taxi, spatiotemporal behavior, non-negative matrix factorization, cluster analysis

摘要： 为探讨出租车的时空行为特征，采用非负矩阵分解法从时间和空间两个维度对出租车行为进行聚类分析。首先划分时空单元，以上客点的时空分布特征表征出租车的时空行为，并构建时空行为矩阵，然后采用非负矩阵分解法进行出租车聚类。最后以广州市为例进行研究，聚类结果表明：出租车受广州市空间结构特征的影响聚成了4类，而受时间维度的影响较小，但不同时段的行为也有明显差异。4类出租车都形成了明显的围绕区域商业活动中心集中运营的空间行为模式以及调整运营范围以适应不同时段出行活动需求的时间行为模式。研究揭示了出租车的时空行为与城市空间结构、城市出行活动的联系，能为出租车和城市管理发展提供参考。

关键词: 交通工程, 出租车, 时空行为, 非负矩阵分解, 聚类分析