Schedule-driven intersection control

Model-based intersection optimization strategies have been widely investigated for distributed traffic signal control in road networks. Due to the form of ''black-box'' optimi- zation that is typically assumed, a basic challenge faced by these strategies is the combina- torial nature of the problem that must be solved. The underlying state space is exponential in the number of time steps in the look-ahead optimization horizon at a given time reso- lution. In this paper, we present a schedule-driven intersection control strategy, called SchIC, which addresses this challenge by exploiting the structural information in non-uni- formly distributed traffic flow. Central to our method is an alternative formulation of inter- section control optimization as a scheduling problem, which effectively reduces the state space through use of an aggregate representation on traffic flow data in the prediction hori- zon. A forward recursive algorithm is proposed for solving the scheduling problem, which makes use of a dominance condition to efficiently eliminate most states at early stages. SchIC thus achieves near optimal solutions with a polynomial complexity in the prediction horizon, and is insensitive to the granularity of time resolution that is assumed. The perfor- mance of SchIC with respect to both intersection control and implicit coordination between intersections is evaluated empirically on two ideal scenarios and a real-world urban traffic network. Some characteristics and possible real-world extensions of SchIC are also discussed.