地理科学 ›› 2021, Vol. 41 ›› Issue (9): 1505-1514.doi: 10.13249/j.cnki.sgs.2021.09.002

• • 上一篇    下一篇

复杂网络视角下时空行为轨迹模式挖掘研究

张文佳(), 季纯涵, 谢森锴   

  1. 北京大学深圳研究生院城市规划与设计学院,广东 深圳 518055
  • 收稿日期:2020-08-21 修回日期:2020-12-27 出版日期:2021-09-25 发布日期:2021-11-17
  • 作者简介:张文佳(1985-),男,广东云浮人,助理教授,博导,主要研究面向城市研究的大数据与人工智能技术开发、时空行为与结构优化、基于行为的建成环境规划等。E-mail: zhangwj@pkusz.edu.cn
  • 基金资助:
    国家自然科学基金项目(41801158);深圳市自然科学基金重点项目(GXWD20201231165807007-20200810223326001);深圳市基础研究自由探索项目(JCYJ20180302153551891);深圳市自然科学基金面上项目(JCYJ20190808173611341)

Pattern Mining of Spatio-temporal Behavior Trajectories by Complex Network Analysis

Zhang Wenjia(), Ji Chunhan, Xie Senkai   

  1. School of Urban Planning and Design, Peking University Shenzhen Graduate School, Shenzhen 518055, Guangdong, China
  • Received:2020-08-21 Revised:2020-12-27 Online:2021-09-25 Published:2021-11-17
  • Supported by:
    National Natural Science Foundation of China(41801158);Shenzhen Municipal Natural Science Foundation (Key Project)(GXWD20201231165807007-20200810223326001);Shenzhen Municipal Basic Research Project (Free Exploration)(JCYJ20180302153551891);Shenzhen Municipal Natural Science Foundation(JCYJ20190808173611341)

摘要:

针对时空行为轨迹大数据的序列性、时空交互性、多维度性等复杂特性,构建结合时间地理学与复杂网络的分析框架,建立时空行为路径与时空行为网络之间的转换关系,利用复杂网络社群发现算法对时空行为轨迹进行社群聚类、模式挖掘与可视化。基于北京郊区居民一周内活动出行GPS轨迹数据的案例分析发现:① 复杂网络分析方法可以有效挖掘具有相似行为的群体特征和识别出典型的行为模式。② 可以灵活处理多元异构与多维度的行为轨迹大数据以及满足不同叙事、不同空间相互作用、不同时序的应用需求。③ 北京郊区被调查居民的行为模式存在日间差异与空间分异。

关键词: 时间地理学, 复杂网络, 行为轨迹大数据, 模式挖掘, 社群发现算法

Abstract:

The big data of spatio-temporal behavior trajectory have multiple complex characteristics, such as sequential properties, spatio-temporal interaction, and multidimensionality. Facing these complexities, this paper explores a new pattern mining method of spatio-temporal behavior trajectories to provide a more flexible approach to pattern mining, particularly for the trajectory big datasets. This paper explores a new spatio-temporal behavior pattern mining method by three steps. First, based on Social Affiliation theory, we develop a conversion method from space-time paths to spatio-temporal networks by individuals, by incorporating the analytical framework of time geography and complex network. The temporal and spatial attributes of activities are saved in the spatio-temporal networks. Second, this study uses the Louvain Method to detect communities, that is, the clustering trajectories or behavior patterns in a behavioral network. This community-detection method is widely used in the field of network science, particularly for handling a large set network data. Third, relying on the visualization techniques from time geography, this study integrates the advantages of 2-dimentional and 3-dimentional charts to analyze and display the characteristics of spatio-temporal behavior patterns based on multiple perspectives. By mining similar or cohesive communities, this study further explores the characteristics of spatial heterogeneities in behavioral patterns and their day-to-day variabilities. By adopting a weekly-long trajectory data from a GPS-based individual activity-travel survey in 2012 Beijing, this study reveals three major findings. First, complex network analysis can effectively extract grouping patterns with similar behaviors, along with identifying representative behavior patterns from messy trajectories. Besides, the new approach provides a new perspective for further exploring the spatio-temporal interaction of human activities in time geography and social geography. It has the capacity to flexibly handle heterogeneous and multidimensional behavior trajectories and detect patterns from trajectory big data by varying narratives of activity-travel events, spatial interaction levels, and lengths of time series or sequences. Second, in the case study, considering the time allocation characteristics and sequence of activities, activity and sequence narrative are selected for analysis, and 6 major behavior patterns based on similarity mining are analyzed and compared. Results show that the residents have a typical behavior pattern in which they worked from 9 AM to 7 PM after sleeping on weekdays, and then conducted leisure, entertainment, housework and private affairs before 9 PM. Third, this study extends the complex network method to behavior pattern mining with multi-day spatio-temporal data. The effect of spatial interaction was considered when measuring the individual connections, and the spatial distribution characteristics of community were compared by adjusting the distance attenuation factor. This finding suggests significant spatial heterogeneities in behavioral patterns of surveyed residents in the suburb of Beijing. The residents also have significant day-to-day variabilities in spatio-temporal behavior patterns, mainly between weekday and weekend as well as between Saturday and Sunday.

Key words: time geography, complex network, big data of behavioral trajectories, pattern mining, community detection algorithm

中图分类号: 

  • K901