Analyzing And Mining Urban Human Mobility Patterns Based On Trajectory Data

Urbanization has accelerated the evolution of urban morphology and changed the urban human mobility patterns.Meanwhile,it has caused major problems such as traffic congestion,air pollution,and resource depletion.Urban human mobility analysis aims at dealing with the above-mentioned problems,as well as understanding the potential relationship between human travel behavior and urban economic development.Urban human mobility analysis is a multi-disciplinary field that combines concepts from complex network science,computational social science,statistical physics,and social economics.Prior research on human mobility mainly fo-cused on empirical analysis of the characteristics of human trajectory datasets.In today’s big data era,data has experienced an explosive gro.wth under the rapid development of computing,com-munications,and networking technologies.Intelligent mobile terminals such as mobile phones enable us to easily collect travel traj ectories.Such multisource data provides opportunities for an in-depth analysis of urban human mobility.However,analyzing and mining multisource hetero-geneous trajectory data raises some major challenges,such as semantic function representation,network node importance evaluation,and multidimensional attribute extraction.This dissertation comprehensively summarizes the latest theoretical frontiers and methods of human mobility analysis with respect to machine learning,artificial intelligence,and big data.In addition,it studies human mobility patterns in real-world urban scenarios,explores latent laws,and provides travel services to residents.1.Identification of subway station functions based on semantic vector representation.This dissertation proposes a semantic framework IS2Fun to mine the latent functions of subway sta-tions considering both static and dynamic semantic representation of subway stations.By em-bedding POI distributions and human mobility patterns,this dissertation categorizes the stations into ten functional clusters based on Station2vec.The proposed IS2Fun framework outperforms some existing benchmark methods(Doc2vec-based,LDA-based,and TF-IDF-based method)in terms of identifying the functions of subway stations.2.Ranking subway station importance based on macroscopic and microscopic indicators.Considering the evaluation of node importance in subway networks,this dissertation proposes a novel ranking method of station importance(SIRank)based on macroscopic and microscopic indicators.By analyzing the degree centrality,betweenness centrality,closeness centrality,and bi-directional passenger flow between origns and destinations in SSSN and SSPN,SIRank iter-atively generates an importance value for each station and identifies the most important station.Our experiments demonstrate that the hit ratio in SIRank is much higher than that of WMIRank and NDRank.3.Subway passenger flow prediction based on the spatiotemporal loyalty.Focusing on the multidimensional feature extraction of human mobility patterns,this dissertation proposes a s-patiotemporal loyalty-based model(STLoyal)for subway passenger flow prediction.STLoyal introduces a concept of the loyal passenger and leverages loyalty,time,location,and weather metrics to improve the accuracy of prediction.Furthermore,STLoyal shows the best perfor-mance when compared to three models such as SVR,BPNN,and GBDT.