Trajectory data-based severe conflict prediction for expressways under different traffic states

Conflict prediction is prevalent in the field of expressway safety management. As the precursor of crashes, severe conflict draws great concentration in previous studies. However, the conflict mechanism heterogeneity under different traffic states is always ignored. The different traffic environment affects driving behavior, resulting in various casual factors. In order to better predict and understand the influencing factors of severe conflicts, this study proposes severe conflict prediction models under three traffic states based on a 3-hour vehicle trajectory data on the Shanghai Inner Ring Expressway. Firstly, the data is labeled with three traffic states based on the Traffic State Index and road average speed. Modified Time to Collision (MTTC) of vehicle pairs is calculated to identify severe conflicts. Second, the spatial and temporal characteristics of conflicts are analyzed to explore the conflict heterogeneity under different traffic states and the impact area of different ramps. Finally, logistic regression models are established to predict the likelihood of severe conflicts. Compared with the model without division of traffic states, congestion state model and transition state model have an increase of 1.7% and 8.9% in prediction accuracy, respectively. Main casual factors of severe conflicts differ under three traffic states. The occurrence of severe conflicts is more sensitive to the change in traffic volume and ramps when the traffic state becomes more congested. Vehicles in 300-meter area upstream exit ramps/downstream entrance ramps suffers from higher risk of collisions. However, when the traffic state approaches free flow, severe conflicts are more caused by drivers' unsafe behavior such as high speed and short spacing between vehicles. The findings can help transportation managers figure out the main casual factors of expressway crashes under different traffic states, and thus develop more targeted safety management strategies. (c) 2023 Published by Elsevier B.V.