Abstract:In order to deeply understand the characteristics of the cross-type collision conflict between bus passengers crossing the non-motorised road to enter and exit bus stops and the longitudinal crossing of E-bikes, and to explore the factors influencing the severity of the conflict, a method for analyzing the safety impacts of E-bike crossings on the entering and exiting behaviors of bus passengers was proposed. Based on the trajectories of E-bikes and bus passengers obtained from aerial photography, the bus passenger-E-bike conflict was divided into three categories: bus passenger avoidance, E-bike avoidance, and both sides avoidance. Based on the characterization of cross-type conflicts, the relative collision time (TTR) indicator was proposed, and the conflict severity evaluation model based on improved K-means clustering was constructed by combining the post-trespassing time (PET) and safety deceleration speed (DST). Finally, five conflict severity influencing factors were extracted from the perspectives of bus passengers and E-bikes respectively, and a conflict severity level prediction model based on BP neural network was established and tested for model validity. The results show that speed difference and relative speed have the greatest influence on the conflict severity level. E-bike active avoidance has the least influence on the motion state of both parties in the conflict, and the relative collision energy is the lowest. The calculation results of the proposed conflict severity level prediction model have a high degree of conformity with the actual conflict level, and can predict the conflict severity level to a certain extent.