Urban Road Network Dynamic Vehicle OD Estimation Based On Multi-source Heterogeneous Data

As an important prerequisite for research on road network traffic flow and various traffic simulation models and an important basis for fine-grained urban traffic control,how to obtain OD(Origin-Destination)demand is an important research topic in the field of transportation.For this reason,a large number of studies have considered the actual propagation process of traffic on the road network based on various static OD estimation models and methods to establish dynamic mapping relationship between OD demand and segment traffic counts(including link traffic counts and intersection turning traffic counts),that is dynamic assignment matrix,so that the original static OD estimation models have been transformed into dynamic estimation models.With the application of advanced communications and computer technologies in road traffic detection applications(such as vehicle-mounted GPS,video license plate recognition,etc.),the detection of road traffic flow has broken through previously only to obtain cross-section traffic,occupancy and speed,the driving path information of single vehicle can be further obtained.Based on this,this paper comprehensively uses a variety of existing data,including GPS data,video license plate identification data and cross-sectional flow data.On the basis of fully mining the traffic flow information contained in various data,this paper constructs a dynamic traffic OD estimation model for urban road network based on multi-source heterogeneous data.In order to make the proposed model could reflect the real OD distribution,this paper firstly established road-network-grade traffic zones that are suitable for dynamic traffic OD estimation based on taxi GPS data.Secondly,on the premise of the determined distribution of traffic zones,the demand observations between OD pairs were extracted using GPS data and video license plate identification data;at the same time,the reliability of the observations of the extracted OD requirements was performed based on the probability statistical theory.Conditionally,considering that the video license plate identification data and GPS data could not cover all vehicles in the road network,the observed OD demand were only the part of the complete demand in the road network.Therefore,this paper further divided OD demand into two parts: the observable and the unobservable.For unobservable parts of the OD demands,this paper used a Kalman filter model to be subject to linear state constraints to estimate these uncertain OD demands,where those OD demand were used as a state variable to construct the state transfer equation;The dynamic mapping relationship between OD demands and traffic counts that was established based on dynamic traffic assignment theory to build the observation equation.At the same time,this paper established the linear state constraint conditions based on the observed actual traffic demand of the traffic zones.Evidently,the observations of the traffic flow in the road network include both the observable part of the OD demand and the unobserved part of the OD demand.In order to obtain the cross-section traffic counts for estimating the unobservable OD,this paper reconstructed the missing trajectories of those observed OD demands using the shortest path algorithm to determine the part of the trajectory of observable OD demand in the cross-sectional flow and remove these counts from the corresponding traffic counts,and the remaining part was used to estimate the unobserved part of the OD demand.Finally,the paper comprehensively evaluates the performance of the constructed OD estimation model based on the actual road network.The results showed that the dynamic traffic OD demand estimated based on the proposed method can better reflect the real traffic demand of the road network.;and compared with ordinary Kalman Filtering Model,the proposed method had advantage in terms of accuracy and reliability.