Adaptive Look-ahead Optimization of Traffic Signals

ALLONS-D is a decentralized real-time traac control scheme. A description of its basic architecture, system model, and optimization algorithm is given. Through simulations, the scheme is shown to be suitable for adaptive control of an isolated intersection. An eecient tree searching algorithm is used to nd a traac signal plan that minimizes delay. The search is constrained only by maximum and minimum green times per phase. This algorithm is free to consider cyclic and non-cyclic signal plans alike. The rolling horizon concept is used for the calculation and implementation of optimal signal plans. The ALLONS-D scheme has several key diierences with other real-time traac control schemes, such as OPAC. Extensive simulation experiments on a single intersection indicate the success of the ALLONS-D scheme versus any xed-cycle plan. The tests included a full range of uniform and non-uniform demands from low levels of saturation to the over-saturated case. For arterial networks of intersections, the scheme is shown to be capable of inducing a level of coordination that results in progressive signal plans. In particular, the scheme can be used to coordinate travel times on transit routes. An extension of the scheme to provide network-wide optimization of traac signals is explained in the form of a two-layer scheme. Some preliminary results using this approach are included. Simulation experiments are performed with a newly developed traac simulator called STNS.