Estimation And Prediction Of Dynamic Travel Time On Urban Signalized Arterials

Urban traffic congestion management is a package solution that combines strategy and tactics,short term and long term,planning and implementation,and also involves cooperation among government,enterprises,scholars and the public.In comparison with other congestion countermeasures,intelligent transportation system is considered as an important measure because of its small investment and quick effection.Travel time,as a major parameter of intelligent transportation subsystems,is widely used in travel information service,dynamic route guidance,network performance evaluation,traffic demand management and trip decision.Obviously,the estimation and prediction of dynamic travel time on urban signalized arterials include multiple links can achieve current perceptions and future predictions.This will contribute to the smart decision,precise management and efficient service of urban traffic.This paper has two objectives,one is to build up a physical model based on the mechanism of traffic phenomenon to process obtained traffic data except travel time,which can estimate dynamic travel time accurately and indirectly under both steady and oversaturated state.And the other is to explore the non-linear relationships between path travel time and its historical values or link travel time with the use of data-driven method,then the direct prediction of short-term arterial dynamic travel time will be implemented in real time.Thus the main works include the following 5 aspects:(1)Basic theory study of dynamic travel time estimation and predictionAfter distinguishing the concepts of “dynamic travel time estimation”and “dynamic travel time prediction”,the two are classified by data types and core methods respectively.Then the related literatures are systematically reviewed,including its research objects,basic data,theoretical methods,variables,models,accuracy,efficiency and scope of application,besides,the research status is summarized.(2)Study of shockwave tracing model(STM)based on traffic flow wave theoryEach whole link can be treated as one cell,then a link-based shockwave tracing model on the basis of traffic flow wave theory is proposed.Taking advantage of that traffic states within a link can be simplified as arrival,saturated,jammed conditions,traffic dynamics are simulated by tracing four major shockwaves: queuing,discharge,departure and compression waves.An arterial corridor consisting of consecutive links can be divided into multiple cells.Then a shockwave tracing model for signalized arterial networks is presented by considering the boundary conditions,upstream and downstream association.According to the known external traffic demand,traffic flow characteristic parameters,signal timing scheme,road geometric parameters,STM describes shockwave propagation and queuing evolution dynamically,besides,the spatial-temporal distribution of traffic state within each link are estimated in real time.(3)Study of virtual vehicle trajectory reconstruction based on traffic flow wave theoryThrough analyzing 6 typical trajectories on each link,a summary of its traversal traffic states,encountered shockwaves and traffic signals is summarized.Accordingly,the trajectory reconstruction can be generalized into 3 cases.Then the formulas for 6 key time points of vehicles entering the link are given,so that 3 time periods corresponding to 3 cases are divided.For a virtual vehicle,the reconstruction case is judged by its entry time,the maneuver decision at each time step is determined by its own status and surrounding traffic conditions,and the step-by-step decision continues until it arrives at the destination.Considering the spatial-temporal evolution,the departure time of upstream section can be taken as the arrival time of downstream section,thus the reconstruction of vehicle trajectories on arterials is implemented with the repeated use of link-based method.(4)Study of arterial dynamic travel time estimation based on shockwave tracing and trajectory reconstruction of virtual vehicleFirstly,the shockwave tracing model is used to estimate traffic state by analytically deriving the velocity and trajectory of four shockwaves,and the queue length at every moment;besides,the maximum queue(both length and time),minimum queue(both length and time),the initial queuing time and clearing time during each cycle are extracted.Secondly,all the above results are served as input data to reconstruction method to calculate the cumulative distance of the virtual vehicle at each time step.Hence,the difference between the starting time and the ending time is the arterial dynamic travel time,and the average travel time can then be calculated as the aggregated value at certain intervals.The numerial example validates that the proposed estimation model can generate accurate arterial dynamic travel time without relying on streaming data.Additionally,it can be applied to steady state and oversaturated state with residual queue.(5)Study of arterial dynamic travel time prediction based on particle filterConsidering the intrinsical nonlinearity of urban transportation system,and the assumption of Gaussian noise may not be consistent with collected travel time data.Therefore,a real-time dynamic travel time prediction method by particle filter is presented because its advantages in addressing nonlinear non-Gaussian systems.Specifically,the arterial travel time and link travel time are defined as state variable and measurement variable,respectively.The state variable can be approximated by a collection of particles,the state value of each particle corresponds to the arterial travel time at a certain period on history day,and its weight can be calculated by comparing the dissimilarity between current and historical traffic pattern.The improved algorithm updates particles using data sequences rather than explicit state transition,represents the traffic pattern as travel time matrix containing several time intervals along all links,and adopts partial resampling strategy,which is different to the traditional methods.After that,the detailed realization of 5 core steps of the algorithm is elaborated,and its complete logical framework design is given.The numerical example shows that the single-step prediction has the smallest error,the multi-step prediction within 15 minutes can provide reasonable results,which indicates the accuracy and effectiveness of the prediction model.