Research On Urban Traffic Structure Based On Mobile Travel Data Mining

The Internet of Vehicles(IoV)is a derivative concept of the Internet of Things and mobile Internet in the field of transportation.By leveraging state-of-art information communication and processing technology,IoV can achieve a deep integration between vehicles and humans,vehicles,and roads,so as to increase the intelligence of vehicles and improve the operational efficiency of urban traffic.With the support of IoV,multi-vehicles and road-side infrastructures can make the collaborative decisions of route selection and road traffic guidance based on the accurate cognition of the traffic environment situation to optimize the vehicles intelligent driving and intelligent transportation control.However,the intelligent transportation devices can not cognize users' travel demand,and the telematics system has a lack of permeability,which leads the current methods can not completely and accurately depict the urban road traffic status.To this end,this paper focuses on the accurate cognition problem of urban traffic status and takes into account the correlation between urban traffic structure and traffic environment situation formed by users' travel.This paper studies the research on fine-grained urban traffic structure based on mobile travel data,and yields following results:1)In terms of fine-grained traffic structure mining,we propose a multi-source heterogeneous data fusion model to ensure the integrity and accuracy of the construction of urban traffic structure.This model considers the correlation characteristics of multi-source heterogeneous data and fuses the knowledge mined from different data sources.Firstly,aiming at urban traffic region devision,we propose a joint grid segmentation based affinity propagation clustering algorithm considering users' travel demand.The urban traffic regions can be achieved by clustering the mobile data.Secondly,due to the lack of mobile data of walking,riding and driving private cars,we propose a data-driven transportation mode recognition model based on stacked auto-encoders in terms of the continuous GPS trajectory data.Finally,by considering the spatial and temporal correlations among data,the knowledge from multiple data sources are fused to construct the complete and accurate urban traffic structure.Experimental results using mobile travel data of Beijing demonstrate that the proposed methods can effectively realize traffic region segmentation and transportation mode recognition to ensure the rationality and accuracy of urban traffic structure construction.2)In terms of the behavior of transportation mode choice,we propose a data-driven prediction model based on aggregation of travel features.The model predicts the tendency of different transportation modes which takes full account of the correlations among features of mode choice behavior.Firstly,the model defines the preference features reflecting users' transportation mode choice behavior,and discovers the correlation among features through frequent pattern mining methods.Secondly,a fused Lasso regression model considering the correlation information among features is proposed to predict the tendency of users' different transportation mode choice.Experimental results show that the proposed model can effectively utilize the correlation information among features and improve the accuracy of tendency prediction of users'transportation mode choice.3)In terms of evolution prediction of urban traffic structure,we propose an evolutionary game model for multi-mode competition mechanism.The model learns the formation mechanism of stable state of urban traffic structure to predict the evolution trend of traffic structure affected by traffic events.Firstly,a bipartite network between traffic regions and traffic network topology is designed to discover the travel crowd affected by the traffic events.Then,the influence factors of the crowd's transportation mode choice are predicted to determine the crowd's possible transportation mode.Finally,considering the heterogeneity and autonomy of travel individuals,the evolution problem of traffic structure can be transformed into a heterogeneous group evolutionary game problem.A evolutionary statable strategy learning mechanism based on deep reinforcement learning is proposed to accelerate the iterative update and search speed of statable strategy learning process.The experiments of urban road repair leading the evolution of traffic structure demonstate that the mechanism improves iteration and convergence speed of game statable strategy learning,and ensures the evolutionary game model can predict the evolution trend of traffic structure affected by traffic events accurately.