基于交通大数据的移动模式分析综述

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...

Review of Human Mobility Pattern Analysis Based on Big Transportation Data

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.