Dirk Helbing, Stefan Lammer

Paper #: 10-09-019

A coordination of vehicle flows is usually reached by a cyclical operation of traffic lights, and by synchronizing these cycles. The typical conditions, however, for which traffic lights are normally optimized for, never occur exactly. Large fluctuations in the number of vehicles arriving during one cycle time may lead to an inefficient usage of green times, which are often either too short or too long. The method we propose here allows for variable adjustments not only of the duration, but also of the order of green phases, while it reaches at least the same intersection throughput capacity as an optimized fixed-time controller. This is particularly important, when intersections are highly saturated, road networks are heterogeneous, or if public transport is to be prioritized.
We reach the stabilization of queues and red-time durations by a decentralized supervision of (potentially unstable) locally optimizing traffic light controllers. The proposed supervisory concept makes sure that all network flows get a green light regularly and long enough. In addition, it enables flexible responses to local fluctuations in the demand, and it favors a self-organized coordination of traffic flows. In a simulation of the city center of Dresden, Germany, we could compare the proposed concept with an adaptive state-of-the-art controller (which has been optimized within the same simulation suite and includes green waves). Results indicate that not only the mean value of travel times is reduced, but also in their variance, which is positive for the reliability of public transport and individual trip scheduling.