Research On Travel Time Characteristics Of Urban Roads Based On License Plate Recognition Data

Urban road travel time information is essential for traffic management and policy making by governors and for travel planning,route selection and congestion avoidance by travelers.A main source of such information is the travel time distribution(TTD),which provides information such as traffic state condition and travel time reliability.Therefore,it is of practical importance to conduct an in-depth study of the route travel time characteristics and to model the estimation of route travel time.However,the estimation of route travel time distribution faces the problem of sparse sample size,and the direct method of simply summing the link travel time lacks the consideration of the spatio-temporal correlation between the adjacent links.This study takes the link travel time and the route travel time as the research objects,analyses the influencing factors of travel time characteristics quantitatively,and constructs a route travel time distribution estimation model considering the correlation between links based on GMM clustering,Markov chain and hierarchical Bayesian theory.The specific research content includes the following two aspects:1)Route travel time characteristic analysis.Firstly,the research based on the license plate recognition data,obtains the high-quality travel time dataset by data filtering method.Then,according to the distribution histogram,the research analyses the travel time distribution characteristics under different link lengths and different periods,proposes to use Burr distribution and Gaussian mixture distribution model to fit the travel time data,determines the appropriate distribution model under each condition through goodness-of-fit test,and explains the relationship between travel time distribution,reliability and route length quantitatively,And finally,based on the grey correlation theory,this study analyzes the key influencing factors of travel time characteristics from three aspects: route length,number of intersections and route flow.2)Modeling and effect verification of route travel time distribution estimation.Firstly,the research mines the traffic state transition relationship between the upstream and downstream links based on the GMM clustering algorithm.Then,a multi-order Markov chain model is constructed,proposing the use of autocorrelation coefficients to dynamically represent the influence weights of the traffic state of each upstream links,so as to obtain the conditional Markov path probability.In addition,according to the hierarchical Bayesian theory,this paper constructs a hierarchical model,takes the historical travel time information as a prior information,and obtains the posterior inference of travel time distribution parameters by real-time data.Finally,based on the license plate recognition data in Xiaoshan,Hangzhou,this study compares the estimation effects of the four types of models,and verifies that the improved model proposed can make the most accurate estimation of the route travel time distribution and reliability.At the same time,this paper compares and analyzes the effectiveness and applicability of the model from three aspects:sampling rate,estimated time period and link group division.This study presents a quantitative analysis of the travel time characteristics of urban road and constructs a three-stage estimation model of route travel time distribution.All the methods in the estimation model can be described as explanatory models,which is of great significance for estimating travel time distribution in the case of limited route travel time data samples.