Abstract:
Arterial roads are the framework of urban road network,where the crash occurs frequently due to com-plex traffic environment.It is necessary to conduct corresponding safety analysis,in order to propose constructive coun-termeasures.Geometric features,land use,traffic volume and average speed are gathered at a total of 1 76 road segments from 18 arterial roads in Shanghai.Average segment speed is calculated from floating car data (FCD),which solved the problems related to speed data collection with sensors installed on fixed locations.Considering correlations among seg-ments along an arterial roads,a set of Bayesian hierarchical Poisson log-normal models are developed.The Full Bayesian Method is used for parameter estimation and different prior distributions are tested.Crash features vary depending on time,so the models for peak and off-peak hours are developed separately.Results indicate that hierarchical models im-prove the goodness-of-fit of the data because deviance information criterion (DIC)values of hierarchical models are signifi-cantly less than maximum likelihood estimation (MLE)prior models.The reliability of parameter estimation can be im-proved by MLE prior.The standard deviations of parameters of the MLE prior models are less than those of non-informa-tive models.Along arterial roads,the longer the segment length,the more crashes.At the segment level,geometric fea-tures and land use are substantially associated with crash frequencies.Higher traffic volume is associated with increased crash frequencies especially during peak hours.Average segment speed contributes to increasing crash occurrence during peak hours.
