Real-Time Safety Evaluation at Signalized Intersections: Hierarchical Bayesian Extreme Value Theory Models Based on Different Conflict Types

Real-time safety evaluation of urban signalized intersections is a prerequisite for proactive traffic safety management. Due to its independence from historical data, the traffic conflict technique has gained increasing popularity as a tool for real-time safety evaluation in transportation systems. However, different types of conflicts (e.g., rear-end and side-impact conflicts) may lead to differences in safety evaluation, which has led previous studies to typically analyze crash risk based on various conflict types separately. Therefore, this study develops the hierarchical Bayesian extreme value theory (block maxima (BM)) model based on different conflict types to form a novel real-time safety evaluation approach. The proposed model is applied to the real-time safety evaluation of five signalized intersections in Harbin, China. The results show that (1) the proposed BM model exhibits good prediction performance when there is sufficient observation duration and a sufficient number of samples; (2) the proposed model is superior to other baseline models developed based on only one conflict type in terms of prediction accuracy. The empirical findings of this study establish innovative frameworks and theoretical foundations for advancing proactive safety protocols and autonomous mobility systems.