Resumen
The inter-driver variability of follow-the-leader behavior is well recognized as a main cause of bottleneck phenomenon on expressway basic segment. The bottleneck phenomenon observed at sag sections on Japanese expressways is an example where the accumulation of driving behavior variability results in obvious capacity drop and probabilistically occurring breakdown. Conventional theoretic analysis on drivers? car-following behavior generally treats the variability as a random error around a standard behavior pattern, which is not sufficient to study the exact pattern and extent of the variability to provide any acceptable explanation to the probabilistic occurrence of breakdown as well as applicable solution to unify the behavior. Trajectory data from video processing provides a source for this research to analyze every single driver's follow-the-leader behavior pattern and extent of variability from a microscopic view as well as their impact on overall traffic flow condition. 875 time series profile of single drivers? behavior such as acceleration, speed, distance and road geometry is studied. The follow-the-leader behavior of every driver is fitted into a car-following model, where drivers decide their acceleration based on the speed difference, spacing with its leading vehicle, its own reaction delay time, target spacing as well as the road geometry. Reaction delay time and desired spacing are obtained for every driver through correlation analysis and regression. Other parameters are obtained with heuristic optimization algorithm. The car-following model is calibrated with objective function and performance indicators as well as simulation. The objective function and performance indicators are studied to ensure that the model and its parameters can reproduce car-following behavior well. The simulation is carried out to reproduce the probabilistic nature of traffic congestion occurrence at a bottleneck of sag section including the effect of vertical slope increase. It successfully reproduces the tendency that the congestion occurrence probability grows with traffic demand. Finally, the variability of car-following behavior in drivers at sag section is studied with car-following parameter distribution. The desired spacing parameters are jointly distributed. The vertical effect parameter has two peaks concentrating on the value range, which indicates that there are drivers who are not affected by the vertical grade while others are.