Abstract
The coexistence of autonomous vehicles (AVs) and human-driven vehicles (HDVs) in mixed traffic will persist for an extended period, inevitably introducing new safety challenges due to variability in human drivers' behavior, particularly in complex traffic scenarios. To better understand the safety dynamics of emerging mixed traffic flow, this study investigated the safety implications of mixed traffic at freeway merging areas, considering driving behavior. First, human drivers' driving behaviors were classified into aggressive driving, normal driving, and conservative driving using a k-means clustering algorithm based on field dataset analysis. Next, an improved lane-changing model of HDVs, accounting for driving behavior, was developed by incorporating lane-changing duration and a lane-changing motivation function within a multi-objective optimization framework. This model was validated by comparison its performance with the SL2015 model in the SUMO simulation platform. Finally, simulations of mixed traffic flow were conducted under various flow rates, AV penetration rates and ramp flow ratios. The results indicated that a low AV penetration rate (< 20%) did not enhance mixed traffic safety, However, when the AV penetration rate exceeded 20%, safety improved, as evidenced by reductions in the conflict rate, time to collision (TTC) frequency, lane change time to collision (LCTTC) frequency and conflict severity. Additionally, it was found that higher ramp flow ratios and merging flow rates increased safety risks, particularly in scenarios with low AV penetration rates. These findings provide valuable insights for the development of mixed traffic safety improvement strategies and the effective deployment of AVs.