Modeling Car-Following Behaviors on a Merging Section in an Expressway On-Ramp Bottleneck

The car-following behaviors at on-ramp bottlenecks are complex, especially those at the acceleration lane, where the vehicles are not only affected by the preceding vehicles but also the vehicles at the mainline. In this paper, the empirical trajectory data set of the Hongxu on-ramp bottleneck at the Yan’an Expressway in Shanghai, China is collected. Firstly, both time gap and space gap standards are proposed to identify the car-following state. Then the random forest method is used to select the key variables from 17 possible factors that affect the car-following behaviors at the acceleration lane. Finally, a parameterized model, Gazis-Herman-Rothery (GHR) model, and a non-parameterized one, a Bayesian network (BN) model, with the consideration of lateral stimuli are developed. As a result, 4 key variables are selected, and a half of them are related to vehicles at the shoulder lane. Comparing with the conventional GHR and BN model without lateral stimuli, both R-square and Ossen index are reduced, which implies the significant influence from the mainline.