Multi-Agent Look-Ahead Traffic-Adaptive Control

The objective of this thesis is to create a distributed, multi-agent, approach to traffic control. This PhD thesis' focus is on the control of a network instrumented by traffic signals.A thorough literature study has been performed, reviewing the current state of the art in traffic signal control. On the basis of this literature survey, a taxonomy of existing systems was constructed. The design of a traffic-adaptive control system is as well a science as an art. Along the way compromises have to be made in order to end up with a workable system that is not only able to come up with good signal timings, but is also able to deliver them on time. The taxonomy constructed of the various traffic-adaptive control algorithms is based both on the underlying principles and on the compromises that were made to come up with a workable, albeit less optimal system. A new adaptive control algorithm is subsequently developed that incorporates the strong points of each of the algorithms reviewed. The algorithm determines a short term policy on the basis of a long-term analysis and considers the individual signal groups as the smallest controllable entity. Although state of the practice in vehicle-actuated control, look-ahead adaptive control still use stages as the smallest controllable entity, which reduced the flexibility of this approach. The developed algorithm is capable of controlling a single intersection, but can be configured for use in a network. When configured for use in a network the controller shares its intentions regarding its control plan with nearby intersection controllers and informs them of traffic that it plans to release. In order to enable cooperation controllers must be willing to adjust their locally optimal control plan for the benefit of the network. In order to achieve this controllers are informed about the cost inflicted by them to nearby controllers. Using this information, intersection controllers can iteratively adjust their plan to the benefit of the network. In order to evaluate the developed control algorithms a test bed was developed during the course of this thesis. The test bed was essential in the development and testing of the algorithm. The test bed was also used in a proof-of-concept study for the N470, whereas the performance of the algorithm was benchmarked for a corridor against freshly optimized traffic-actuated controllers.