Dynamic Traffic Assignment For Large-scale Urban Networks Based On Multi-source Sensor Data

Dynamic traffic assignment(DTA)is a vital component of road traffic network analysis.To establish a DTA model for large-scale urban road networks that meets the characteristics of urban road traffic flow in China based on multi-source traffic data,has great practical significance for the development of intelligent traffic management systems.Existing studies still have some limitations:(1)the OD estimation methods are highly dependent route choice assumptions,(2)insufficient description capabilities for traffic flow loading and propagation,and(3)low efficiency of DTA solutions.These limitations result in that the DTA-based intelligent traffic management technologies are difficult to be applied in large-scale urban road networks.In view of this,considering the development and application requirements of urban road intelligent traffic management systems in China,a DTA model for large-scale urban road networks based on multi-source data is proposed in this research.The research tasks of the study involving three aspects: dynamic OD pattern estimation,dynamic loading modeling for urban road networks,DTA modeling and its solutions.The details and conclusions of the research work are summarized as below.Current OD estimation methods for large-scale urban road networks heavily dependent on the assumption of route choice behaviors.Considering there are a large amount of trajectory data in the data records collected by electronic bayonet systems,a data driven-based dynamic OD pattern estimation method is proposed multi-source traffic flow data.Then,the effect of trajectory data heterogeneity on OD pattern estimation is quantitatively analyzed based on a spatial statistics method,and further proposes an OD adjusting method for fine-tuning the specific OD flows.Finally,the dynamic OD patterns of urban area of Kunshan city at morning peaks,evening peaks and noon non-peak were estimated.Experimental results show that the estimated trip generation/attraction intuitively reflects the changes and spatial distribution of traffic demand at different time periods.The average MAPE value at each time period is within 20%.The estimation accuracy varies with the sampling rate of trajectory data,and it reduces with the decreases of the sampling rate,and the minimum acceptable sampling rate is 60%.Spatial statistical analysis results show that the local spatial autocorrelation of trip generation/attraction variations hardly changes with the trajectory sampling rate,and the trajectory spatial distribution.The impact of heterogeneity on the number and location distribution of hotspots(traffic zones)has a strong correlation with the location of the test area.Moreover,the OD fine-tuning method can effectively improve the estimation accuracy by adjusting a limited number of OD flows.Existing dynamic network loading model has insufficient ability to describe intermittent traffic flow under the influence of signal control,this paper proposes a novel dynamic loading model for urban road network flow based on an improved path-based cell transmission model.First,the signalized intersection nodes are divided into multiple turning cells,and the flow transmission and state update equations of turning cells are constructed.Then,an improved path-based cell transmission model suitable for urban roads is proposed.Finally,a dynamic network loading model of urban road networks based on a mesoscopic simulation with path flows as loading unit is constructed.Considering that traditional methods which manually dividing cells is time-consuming and laborious,an automatic cell division approach based on road network topology is proposed,which realizes rapid construction of cell models for large-scale road networks.The network loading model was tested on a partial road network composed of the Changjiang Road and its surrounding roads in Kunshan city.Testing results show that the proposed model can reasonably load the given dynamic OD to road network under the premise of deterministic path assignment,and accurately simulate the propagation of traffic flow and the "formation-diffusion-dissipation" process of traffic congestion.The improved p-CTM model approximates the impact of signal control on the traffic flow at signalized intersections through turning cell capacity constraint.The model uses delay of the approach(travel time of cell on the entrance approach)to replace signal control delay,hererin can provide accurate and reliable dynamic path impedance for DTA modeling,and the MAPEs of path travel time are less than 15% in most periods.Most of previous DTA models are very time-consuming in real large-scale road network application scenarios,a novel DTA model for large-scale road network based on spatial domain decomposition is constructed,and a parallel computing method is employed to improve the solution effectiveness.First,to simulate the traffic states of road network,the proposed meso-dynamic network loading model are employed as a flow propagation constraint,a DTA model based on dynamic user equilibrium assignment criterion is constructed.Then,a simulation-based continuous time-period DTA solution algorithm based on the method of successive average(MSA)is proposed.Second,considering that the calculation complexity of the proposed DTA model is related to the number of paths passing through cells,a new variable-the number of cell paths(cell-path)is defined,and a spatial domain decomposition method for road networks based on breadth-first searching with regional load balancing as its objective is proposed.Finally,a simulation-based DTA solution method based on parallel computing is constructed.The proposed model was tested in urban road network of Kunshan city using 15-minute interval dynamic ODs as inputs.Results show that the proposed method achieves good convergence performance.The model in almost each period can converged after 13 to 19 iterations.The optimal threshold of relative gap(RG)is 0.4%.The local RG index of most OD pairs is less than 1%,and the travel time of the selected paths varies in a small interval.The assigned flow of the shortest path is much higher than those of the non-shortest paths,indicating that the assigment result satisfies the dynamic user equilibrium conditions.The spatial distribution of the estimated traffic speeds and volumes are consistent with real traffic conditions.The simulated path travel times and their observed average values maintain a basically consistent change trend,and the errors between them are within 8 minutes.The simulated travel times in most periods are lower than the average,but closer to the lower bound of the observed values,indicating that the path assignment of actual road network is not an ideal dynamic user equilibrium state.The parallel computing method can significantly improve the efficiency of the model,an optimal speedup(6.78)is reached with 14 processers,and the calculation time is reduced to less than 10 minutes.Therefore,the proposed DTA model realizes the balance of assignment accuracy and calculation efficiency,and thus can be applied in real-time intelligent traffic management systems.