Abstract
Sags are one of freeway geometric features where the gradient changes from downgrade to upgrade. As the sequential gradient change causes speed disturbances in traffic flow, traffic capacities at sags become lower than flat sections. Statistics reported that more than 60% of traffic congestion occurred on freeway networks in Japan were caused by sags. Considerable scientific attention has been paid on the bottleneck phenomena at sags during the last three decades both from microscopic aspect and macroscopic aspect. However, the detailed bottleneck mechanisms at sags are still uncovered. This paper developed a data assimilation system combining online-observations and model simulations, to grasp the traffic dynamics towards the congestion queue formation process at sags by estimating the unobservable parameters in traffic flow models. In the system, it is assumed that individual vehicle data is collected by online at a certain distance along the target section. The performances of a particle filter (PF) and an extended Kalman filter (EKF) by comparing the estimated throughputs and the observations were evaluated. It is revealed that PF presents the higher accuracy in traffic state estimation than EKF. Then, by employing the PF based data assimilation system, the variation of model parameters, critical traffic density, free flow speed and traffic capacity, before and after traffic breakdown was analyzed with or without pace-maker-light (PML). The findings are as follows: i) the traffic capacity at the block just downstream of the bottom of sag is slightly lower than the other block, which caused the queue formation at sag, ii) the capacity drop phenomena at the bottleneck block was made visible, and iii) PML may contribute to improve the travel speed, while it may decrease the critical capacity at sag.
