Spatio-temporal Trajectory Mining Based On Massive Floating Vehicle Data And POIs On SNS

The explosive growth of data triggers people's reconsideration towards the value of data.People raised many new thoughts,or the so-called data-driven innovations in the perspective of data,which promoted the changing of thinking patterns.In terms of transportation,many traditional TDC(Transportation Data Collections)relying on human resource was transformed to automated and intellectual TDCs,therefore a large quantity of data was accumulated rapidly.Of particular note,many OBD(On-Board Diagnostic)and Navigation Equipment had collected massive spatio-temporal trajectories.With lots of valuable traffic information,these spatio-temporal trajectories serve as major ways to deepen the understanding of travel habits and active pattern of residents,improve the level of transport management and service,optimize traffic system operation and tackle various transport problems.However,the existing studies on spatio-temporal trajectories are mainly based on sample data of taxis or that of a few private cars,so it is still difficult to explore valuable knowledge.It has became the key problem in the study of current transportation that how to explore the trajectories sufficiently,quickly and efficiently when facing those massive vehicles so as to focus on profound knowledge.Facing the transportation data service,this paper discusses the cross-exploring research based on massive spatio-temporal trajectories,information of drivers,information of vehicle attribute,and check-in information on SNS.Besides,this paper introduces a portrait method of vehicle drivers based on trajectories,by which it can further analyze the travel habits and active patterns of drivers,elaborate on the traffic needs of drivers and help to transfer those complex and invisible traffic data into easily-understanding and valuable knowledge.Through the portrait of driving characteristics,this paper also explains the driving habits of drivers based on cloud model;and by combing some factors including ambiguity,randomness,and statistical characteristics,it conducts a comprehensive evaluation of the risk level of driver's driving behavior according to cloud computing principle and reversal cloud generators.In addition to these,as to solve the problem impeding the development of EV industry,this paper put forwards a data-driven siting recommendation for EV charging piles.This recommendation method is made by the mode of “parking sites,region of interest(ROI),and charging service”,which means,massive spatio-temporal trajectories and check-in information on SNS were all included to obtain the accurate sites.In the process of data fusion,this paper presents an improved weight-based W-DBSCAN clustering algorithm to distinguish different clusters according to POIs effectively since the unsupervised DBSCAN algorithm cannot make effectively identification over clustering points with different features.Besides,this paper also solves the uneven distribution of data points in large regions in accordance with road network segmentation technique.At last,by using the siting method above,this paper conducted a case study of Wangjing area in Beijing.