Understanding Individual Activity Space Pattern Based On Mobile Phone Big Data

Individual human mobility is the study that describes how individual humans move within a network or system,which is a research hotspot in geographic information science,transportation science,statistical physics,urban planning and other fields.Activity space is an important indicator to measure human mobility,which means the subset of all locations within which an individual has direct contact as a result of his or her day-to-day activities,and has important applications in many fields such as urban planning,health geography and transportation geography.Traditional activity space research mostly uses activity diary,which has small sample size and short time span,making it difficult to research activity space patterns.In recent years,individual mobility data,represented by cell phone big data,has provided the possibility of analyzing and modeling the activity space pattern.Scholars have carried out a lot of fruitful research on activity space patterns using cell phone big data.Most of the existing studies measure the individual activity space as a single-center spatial structure and use the radius of gyration as a metric.However,classical behavioral geography theory shows that activity space presents a multi-center structure and hierarchical characteristics.When individuals are far apart from each other,expressing the activity space with a single-center structure will contain too many areas not visited by individuals and significantly overestimate the individual activity space.Therefore,it is necessary to develop an activity space representation method with multi-center and hierarchical structure and its corresponding metrics.In this regard,this paper proposes an activity space expression model with multi-center structure and metrics;What's more,using big cell phone data,an empirical research and model validation in Shenzhen are carried out,as follows.(1)An activity space expression model with multi-center and hierarchical structure and metrics are proposed.The proposed expression model uses spatial clustering to extract individual activity clusters adaptively,and uses each activity cluster as a unit to construct the activity space using the traditional single-center method.On this basis,the activity space metrics under the multi-center structure are explored,and the traditional radius of gyration metric is decomposed into two components: "weighted average sum of radius of gyration within clusters" and "weighted average sum of distance between clusters";the first component The ratio of the radius of gyration to the first component can quantify the extent to which the traditional monocentric model overestimates the size of the activity space.(2)An empirical study was conducted using actual big phone data in Shenzhen.The results of the study show that both the first and second components have a truncated power-law distribution,indicating that at the group level,most people have a small activity space and a small degree of separation between activity centers.The ratio of the gyration radius to the first component is calculated to be less than 60% for 95% of the users,indicating that the overestimation is greater than 40% for more than 95% of the individuals when using the singlecenter activity space method to represent the individual activity space.(3)The accuracy of the classical individual mobility model for modeling the group laws of the first and second components was analyzed.The classical individual mobility models in the literature(including EPR and d-EPR)were corrected using the actual big phone data,and the individual mobility patterns were simulated.The group statistical laws of the fitted simulation results on the first and second components were then calculated.The experimental results show that overall,the model simulation results have a better portrayal of the multi-center structure of the individual activity space,and the population statistical laws are consistent with the empirical data;in contrast,the simulation results of the EPR model tend to underestimate the first component,while the d-EPR model overestimates the second component significantly.