Research On Trajectory Data-based Monitoring In Urban Transportation Networks

Recently,the urban transportation networks are faced with serious traffic load,restricting the transportation efficiency.Therefore,it is of great significance for fully monitoring the transportation networks,such as optimizing the routing platforms and improving the transportation plans.As the infrastructure,transportation networks interact with the urban space and the citizens' activities,mining the interaction pattern helps monitor the transportation network from a macro view.The conventional fixed monitoring system,such as the traffic cameras,is of limited cover range,whereas,the trajectories in the transportation systems are of much larger coverage,and the trajectory data involves activity information which helps fully monitor the transportation networks.Thus the trajectory data-based research on monitoring in urban transportation networks attracts an amount of concentration.Under this background,we aim to study the trajectory data-based monitoring methods in this paper.Specifically,driven by the trajectory data,we study the monitoring methods from the aspects of dynamic fine-grained monitoring and static structure understanding.We proposed a road network speed prediction method with sparse data,a dynamic road network impedance estimation model with context-aware information,a framework for community structure detection driven by POI(points of interest),and a model for mining the significant POIs.The main contribution of this paper can be concluded as following aspects:(1)Based on the map matching,we proposed a road network speed prediction methods with sparse data to realize the network-level monitoring.Specific to the limited coverage of trajectories and complex traffic context,we proposed a sparse data-based context-aware speed prediction framework.The ratio between speed and free speed is applied to estimate the road congestion state,and the temporal congestion state is utilized to cluster roads.Then we propose a probabilistic PCA-based prediction algorithm.The clustering step sets the covariance matrix as inputs and the EM algorithm is used to optimize parameters of the prediction algorithms,which both make the framework capable of tackling sparse data and the clustering step makes the framework fit for parallel and distributed systems.Experiments demonstrate the effectiveness and robustness of our proposed method.(2)By extracting the road travel time from the matched trajectories,we proposed a context-aware road network traffic impedance estimation method to realize the dynamic network-level monitoring,which can help optimize the routing system.The traffic impedance denotes the travel time of a road segment.Conventional model-driven methods can not realize the dynamic monitoring,whereas,based on the trajectory data,we proposed a network-level traffic impedance estimation framework.Faced with the challenges of complex traffic context and sparse data,the framework fuse the congestion state and the POI feature in a road network.We first estimate the road congestion level based on the speed ratio using FCM and construct two 3-order tensors of traffic impedance and congestion probability.Then we proposed a POI matrix-based tensor decomposition method.We further propose a 2-order road network traffic impedance estimation algorithm with the congestion probability as a weighted factor.Experiments on a real trajectory dataset demonstrate the effectiveness and robustness of the proposed method.(3)By analyzing the origin and destination of trajectories,and distribution of POIs,we proposed a POI-driven transportation network community detection method to help understand the spatial structure.Regards to the explanation of the spatial community formed by the trajectory ODs,we propose an analytical framework of the consistency between POIs and communities.We first partition the urban area into grids and construct the network with grids as nodes and origin-destination travel as edges weighted by the amount of ODs.Then we utilize the community detection algorithms to detect the community structure.Accordingly,the OD is connected to POIs in nodes,and we propose a logistic regression-based consistency analysis model between the POIs and community structures.Finally,using the real dataset of Shanghai and Beijing,we verify the effectiveness of our proposed method and discuss the interpretability between POIs and communities.(4)Based on the transportation network community detection,and analyzing the consistency between communities and POIs,we proposed a recognition method of significant POI categories based on the logistic regression,which aims to monitor the driven mechanisms of the transportation network communities.As the reference category of the logistic regression model influences the regression fitness,we propose a model which considers the significance frequency and community entropy of POI categories.In detail,each community is iteratively set as the reference label,and we calculate the frequency of POI category that meets the two restrictions: the estimated parameter is nonzero and it meets the significance level.On this basis,we calculate the significance frequency entropy of every POI category in each community,multiplied by the norm of the POI category.Case studies with the real dataset of Shanghai and Beijing find that the government,hotel,and traffic infrastructure are of the most influence on the transportation network community pattern.