Traffic Conflict Model Based on Bayesian Multivariate Poisson-lognormal Normal Distribution
Yan-yong GUO,Pan LIU,Yao WU,Meng-lin YANG
DOI: https://doi.org/10.19721/j.cnki.1001-7372.2018.01.012
2018-01-01
Abstract:To develop the opposing left-turn traffic conflict model of the signalized intersections,the traffic conflict data and traffic volume data were extracted from 101 hours video data at 12 signalized intersections in Vancouver by the computer vision techniques.Given the impacts of traffic flow status on the traffic conflicts,the traffic flow status was divided into 4 scenarios based on the v/c indicator.The traffic conflict model based on multivariate Poisson-lognormal distribution under multiple scenarios and the traffic conflict model based on Poisson-lognormal distribution under the single scenario were developed.The posterior distribution of the models parameters were derived by the Bayesian estimation method.Based on the Markov Chain Monte Carlo simulation,model parameters were estimated.Using the deviance information criterion (DIC) and the models' expectation variance,the goodness-of-fit and the precision of the models were compared.The results show that the goodness-of-fit of the multivariate Poisson-lognormal based traffic conflict model is superior to the single Poisson-lognormal based traffic conflict model.The precision of multivariate Poisson-lognormal based traffic conflict model under the 4 scenarios are twice,1.5 times,twice,and 1.4 times larger than that of the single Poissonlognormal based traffic conflict model.The conflicting volumes have different impacts on traffic conflicts under different traffic flow status.Results of elasticity analysis show that a 1％ increase in traffic flow of through movement may increase the opposing left-turn conflict frequency by 0.36％,0.56％,0.17％,and 0.78％ for traffic scenario 1,2,3,and 4,respectively,given that the left-turn traffic flow remains the same.Accordingly,a 1％ increase of the traffic flow in the left-turn may increase the opposing left-turn conflict frequency by 0.40％,0.67％,0.40％,and 0.51％ for traffic scenario 1,2,3,and 4,respectively,given that the through traffic flow remains the same.