Estimation Of OD Flow Uncertainty For Urban Road Network Based On Vehicle License Plate Recognition Data

OD flow estimation of road network is the core content of studying the relationship between system capacities and traffic demands to alleviate urban traffic congestion,traffic safety and traffic pollution,and is also the critical support for urban road traffic management and service intelligence.The existing OD flow estimation methods of road network rarely take into account the uncertainty of traffic variables and the traffic distribution process,are prone to overfitting,require high data quality,and lack research on real-time OD flow uncertainty estimation.The intersection electronic police devices have a high coverage rate in urban central areas,and its high equipment sampling rate and recognition rate can effectively capture vehicle trajectories with high penetration rate,which can provide important data support for the estimation of OD flow uncertainty.Emerging methods such as artificial intelligence and big data provide new ideas and new ways for solving the OD flow uncertainty estimation of road network.Based on the vehicle license plate recognition data,this study uses Bayesian neural network and computational graph method to estimate the OD flow uncertainty of urban road network,including constructing the framework for OD flow uncertainty estimation,designing the framework solution process(including historical estimation and real-time estimation),and analysing the accuracy and reliability of estimated results.The specific research contents are as follows.In view of the problems existing in the OD flow estimation methods,this paper builds a road network OD flow uncertainty estimation framework based on Bayesian neural network and computational graph(hereinafter referred to as "Bayesian computational graph").Firstly,study the logical relationships and representation methods of the OD flow uncertainty estimation process in the Bayesian computational graph framework.Secondly,according to the generation and distribution rules of traffic flow on the real road network and the process of vehicle flow distribution,path selection,flow aggregation,three physical layers are designed,which named as the OD flow layer,the route flow layer,and the road section flow layer.Finally,according to the actual road network information,the mapping relationships between the key layers is established to realize the accurate mapping between the OD flow layer and the route flow layer,the route flow layer and the road section flow layer,and make the framework easier to capture the actual distribution of traffic flow on the road network.Based on the constructed urban road network OD flow uncertainty estimation framework,the historical OD uncertainty estimation solution process and real-time OD uncertainty estimation solution process are designed.The estimation process of historical OD flow uncertainty includes two steps.Firstly,the sample OD,sample route flows and sample road section flows obtained from the historical vehicle license plate recognition data are input into the framework to obtain a set of training parameters of sample parameters,and then combine the sample parameters with the historical road section flows extracted from the vehicle license plate recognition data,and input them into the framework to obtain the mean value and distribution interval of the historical OD flows.The real-time OD flow uncertainty estimation takes historical OD and route flows at the same time period,OD and route flows in previous periods,and real-time road section flows as input,and the mean value and distribution interval of real-time OD flows are obtained after the framework is trained.Based on the Bayesian back-propagation algorithm and variational inference method,a loss function suitable for the OD flow uncertainty estimation framework is constructed.The accuracy and reliability of the OD flow uncertainty estimation framework are evaluated for an example road network in the central area of Kunshan.Firstly,determine the number of neurons in each layer of the Bayesian computational graph framework,and select the mean absolute error index,use the grid search method to calibrate the adjustable parameters in the process of solving framework.Secondly,the average absolute percentage error,the root mean square error and the GEH indicators are selected to analyze the accuracy of the historical OD flow uncertainty estimation results,and the road section flow distribution of the real-time OD flow uncertainty estimation process.Finally,the invalid coverage rate and confidence interval width indicators are selected to evaluate the reliability of the historical OD flow uncertainty estimation results and the road section flow distribution during the real-time OD flow uncertainty estimation.The results show that the proposed method can effectively quantify the uncertainty in the historical and real-time OD flow estimation process on the basis of ensuring high accuracy,and also meet the real-time requirements of OD estimation.