Analysis of Sideswipe Collision Precursors Considering the Spatial-Temporal Characteristics of Freeway Traffic

摘要

To avoid bias caused by the high percentage of rear-end collisions in the generic models, only the observed sideswipe collisions on Highway I-5 in the central Puget Sound, Washington State, area are used to analyze the matched case-control logistic regression model. Considering the spatial-temporal characteristics of traffic flow and short-term variation of the sideswipe collision occurrence, a comprehensive analysis of sideswipe collision occurrence and its relationship with the freeway flow across lanes and detector locations are studied. The results imply that sideswipe collisions are more likely to occur at straight and level segments of multilane freeways in off-peak hours. High average occupancy and low average flow and speed variance upstream of collision location tend to increase the probability of sideswipe collision in congested scenarios. In contrast, high average speed and coefficient of variation in speed, low speed variance and coefficient of variation in occupancy, and high average absolute difference in speed between adjacent lanes downstream of collision location may decrease the likelihood of sideswipe collisions in uncongested conditions. Furthermore, the average absolute difference in occupancy and speed between upstream and downstream of the collisions location leads to an increase in the occurrence of sideswipe collisions, while standard deviation of absolute difference in occupancies and speeds exerts opposite effects. The sideswipe collision model proposed is validated using sideswipe collision data of SR 520 freeway, and both the transferability and stability of the proposed sideswipe collision model are statistically tested.