On-line Estimation Of Dynamic OD Flows For Urban Road Networks Based On Traffic Propagation Characteristic Analysis

Accurately and reliably capturing urban road traffic system capacity and demand in real time,and then to achieve the dynamic balance between the two sides is the core component of urban road traffic management,meanwhile is an important way to effectively relieve traffic congestions.Focusing on the side of traffic demand,the majority of existing studies in estimating the dynamic traffic demand,i.e.dynamic OD flows,for urban road networks are still having many problems.The problems include how to make full use of multi-sensor traffic data,how to consider the distribution and propagation characteristics of traffic demands in network level during the OD estimation process,and etc.Accordingly,it is still difficult for many traffic control and route guidance-oriented management techniques to be effectively applied in a large scale urban road network.Therefore,a dynamic OD flows estimation method was proposed by considering traffic propagation characteristic,using the link counts and intersection turning counts that can be readily detected and obtained.The research work of the study involving three key parts,including the methodology of on-line dynamic OD flow estimation,the modelling of dynamic traffic assignment,and the empirical verification and evaluation based on a real urban road network.The details of the research work are summarized as below.In the research of on-line dynamic OD flow estimation,the traditional idea of OD structure decomposition was adopted to consider the structural and stochastic characteristics of OD flows,and a new framework based on a time-series decomposition technique was proposed to decompose the on-line dynamic OD estimation process into two parts,i.e.the off-line estimation of dynamic structure OD and the on-line estimation of dynamic structural deviation OD.First,the off-line dynamic structural OD flows were estimated through the Q-PARAMICS simulator,in which the pre-extracted observable structural link and intersection turning counts were used as the simulation inputs.Second,a widely used state space model was constructed to estimate the on-line dynamic structural deviation OD flows,in which the deviations between the on-line detected counts and the pre-extracted structural counts of the observable links and intersection turnings were treated as the measurement variables.Focusing on the three problems that use state space models for dynamic OD estimation in relevant studies in describing the dynamic OD flow structure,considering the state space errors statistic characteristics,and analyzing the state space system observability in the OD estimation state space,corresponding improvements were made in the research.For the first problem,considering the fact that real OD flows are hardly obtained,the autocorrelation of the historical structural deviation counts of the observable links and intersection turnings instead of the OD flows themselves was tested,and the results were used to describe the dynamic structure of the OD flows and according were treated as the rationale for the state transition model construction.For the second problem,the basis of using the structural deviation counts of the observable links and intersection turnings in the process of random walk as the statistic characteristics of the state errors was proposed,and an adaptive Kalman filtering algorithm was then designed to the structural deviation OD flows state space model.For the third problem,the observability rank condition of the proposed structural deviation OD flows state space model was derived in details and was used in the empirical analysis to guarantee the convergence of the Kalman filtering process.The main purpose of dynamic traffic assignment modeling in this research is to output the dynamic traffic assignment matrix that is used as the measurement matrix of the proposed state space system for the dynamic structural deviation OD flows.For such purpose,the whole dynamic traffic assignment process was divided into two procedures,i.e.the off-line estimation of the route choice probability and the on-line estimation of the path-flow contribution rate.In the aspect of estimating the route choice probability,two methods respectively of extending an intersection and extracting a dynamic road network topology were first proposed based on fact that turning traffic delays for different turning movements should be fully considered in the process of estimating the urban road network impendence.Then,an off-line route travel time estimation model that considers traffic propagation was proposed,along with a time-dependent K shortest path searching algorithm based on the estimated off-line route travel time.Finally,by combing the constraint conditions for route choice set generation and meanwhile considering problem of path overlapping,a PS-Logit model was selected to achieve the off-line estimation of the route choice probability.In the aspect of estimating the path-flow contribution rates,the method that uses the traditional vehicle package hypothesis was adopted in this research,while the ideal assumption of constant headway was improved by estimating the on-line headway that was calculated based on the on-line estimated travel time of the first and last vehicles in the vehicle package.In order to capture the propagation process of vehicles on a path,especially at a signal controlled intersections,a LWR theory-based virtual vehicle trajectory reconstruction technique was proposed to provide an effective way to estimate of the on-line travel time of the first and last vehicles in the vehicle package.A real urban road network that includes 23 zones,31 intersections,and 133 directed links was used to validate and evaluate the proposed dynamic OD estimation method.The validation and evaluation process include three parts,i.e.the detailed calculation procedure of the dynamic traffic assignment matrix,the determination of the state space system based on the analysis of the system observability for the OD estimation and the analysis of the convergence of the Kalman filtering process,and the evaluation of the performance of the OD estimation and the dynamic traffic assignment.For the OD estimation performance,the results show that the overall filtering accuracy is about 87%,and the one-step forecasting accuracy is about 80%.The evaluation results under different demand levels show that the estimation for OD pairs with high OD flows perform better than those OD pairs with medium and extremely low OD flows.The filtering and one-step forecasting accuracy of the OD pairs with high OD flows is up to 95%and 86%,respectively.Under different congestion levels,the results show that the performance during off-peak hours is better than that during the peak hours.The accuracy of filtering and one-step forecasting during the off-peaks hours reaches as high as 91%and 85%respectively.During the peak hours,however,the accuracy of filtering and one-step forecasting is around 80%and 75%respectively.Compared with the augmented state space model,the proposed model produced more accurate OD estimation results with much lower computing time.For the evaluation of the dynamic traffic assignment,the GEH index was used to evaluate the assignment accuracy.It is found that 84.75%of the observable links and intersections turnings get the GEH value smaller than 5,and all the observable links and intersections turnings get the GEH value smaller than 10.Generally,the assignment accuracy in off-peak hours is much better than those in peak hours.Overall,the empirical results show that the proposed method performs well in both OD estimation and dynamic assignment.