基于交通大数据的移动模式分析综述

Review of Human Mobility Pattern Analysis Based on Big Transportation Data

交通大数据可为揭示交通主体显性出行行为背后的深层规律(即移动模式)提供重要基础。精确掌握大数据驱动下交通主体的移动模式,可为需求预测、客流组织、土地利用、事件管理等应用提供理论依据。交通大数据主体繁多、时空多态、关联性复杂的特性迫使小数据时代下的移动模式分析方法转型和升级,但仍可能遇到移动模式一致性表达难、异常类型检测难、复杂关联性表达难、时空多态性建模难、一体可视化分析难等普遍问题。针对这些问题,利用科学知识图谱,对2010~2020年期间3 747篇文献的热点关键词分布、发表趋势分布、出版刊物分布等特点进行归纳总结,结合常见的移动模式分析数据集,系统综述现有研究在移动模式常态分析、非常态分析、关联分析、预测分析和可视化分析方向上的阶段性进展。其中,移动模式常态分析综述了个体活动特性分析、出行类别划分、特定群体分析等应用的研究进展。移动模式非常态分析综述了基于模板匹配和数据驱动的非常态事件检测方法的发展脉络。移动模式关联分析综述了面向不同数据源的关联特性检测方法的发展近况。移动模式预测分析综述了基于数理统计和数据驱动的交通属性预测方法的发展状况。移动模式可视化分析综述了用户交互设计、移动模式宏观可视化、微观可视化和整体可视化的发展近况。最后,系统总结各个分支方向面临的主要问题与挑战,并从数据整合、模型创新、机制变革等角度提炼移动模式分析研究的未来发展趋势,期望为后续研究采用新理论技术开展移动模式分析提供参考。

Big data in transportation provide a basis for revealing hidden mobility patterns in observable travel behaviors of moving objects. Understanding these patterns may provide theoretical foundations for practical applications, such as travel demand prediction, passenger flow organization, land use planning, or incident management. Despite these merits, several issues are encountered in these applications, as big data have high-dimensional entities, various spatiotemporal dynamics, and complex mobility connections. These issues include expressing the patterns in a consistent way, detecting multiple outliers among them, portraying their inner complex correlations, modeling their spatiotemporal polymorphism, or visualizing them in an integrated manner. To deal with these complex characteristics, previous methods adopted in the era of small data have to be upgraded. Thus, we reviewed 3 747 academic studies published between 2010 and 2020 and explored the distributions of co-occurring hot keywords, topic variations, and publication preferences using a knowledge-mapping tool. A systematic summary was given to describe the progress made in five mobility-pattern-based research directions, namely normality analysis, abnormality analysis, correlation analysis, prediction analysis, and visual analytics. In particular, in the field of normality analysis, the progress of activity pattern analysis, travel category segmentation, and special group analysis was first reviewed. Based on the normality analysis, two types of approaches in the field of abnormality analysis were further summarized, namely module-based and data-driven approaches. With the knowledge of the ways identifying normal or abnormal behaviors, correlation analysis was further reviewed to determine the developing trends of detecting mobility correlations based on different data sources. All of the above research results serve as solid bases for prediction analysis of mobility patterns, which was also reviewed in this paper. The review results show that two main types of approaches are used to fulfill the task: statistics-based and data-driven estimations. Finally, previous studies on visual analytics were reviewed to determine how transportation data can be visualized based on user interaction, macro exploration, micro exploration, and overall exploration. Overall, we identified several potential challenges in the aforementioned research directions, and proposed three promising development trends in data integration, model innovation, and scheme revolution. Our ultimate aim is to provide potential guidance for future studies involving mobility pattern analysis based on new theories and technologies.