Vehicle group risk modeling based on risk propagation mechanism.

Surrogate safety measures (SSMs) have been widely used for traffic risk assessment. However, most existing approaches remain focused on the risk state of individual vehicles, providing limited representation of the relational and propagative nature of risk in highly interactive traffic environments. In such environments, the risk observed at one vehicle may reflect the cascading propagation of disturbances through a vehicle group. To address this limitation, this study proposes a novel framework for vehicle group risk modeling based on the perspective of risk propagation. First, Modified Time to Collision is adopted to quantify single vehicle risk as the foundation for group-level safety characterization. Second, vehicle group categorization is conducted using a spectral clustering based method, through which vehicles exhibiting continuous temporal and spatial interactions are separated from the traffic flow as stable analytical units. Third, directed risk propagation relationships within vehicle groups are identified through granger causality analysis, and the corresponding propagation intensity is quantified using a generalized additive model. On the basis of these components, a vehicle group risk indicator is constructed by integrating the basic risk of individual vehicles with the propagation induced effects arising from the interaction structure within the group. Experiments are conducted on the highD dataset, and the results validate the effectiveness of the vehicle group risk assessment framework proposed in this study. The proposed approach offers an interpretable and propagation aware basis for vehicle group risk assessment, providing methodological support for extending traffic safety analysis from individual vehicles to vehicle groups.