Giovanni Petri (London):

Local and nonlocal information in a traffic network: how important is the horizon?

Abstract:

Recent advances in distributed sensor network technology have changed the landscape of traffic optimization in which small, mobile devices are able to sense local information and communicate in real time with one another. Naive optimization algorithms that operate solely on the local or global level are inherently flawed, as global opti- mization requires every local sensor to communicate with a central- ized base-station, creating prohibitive bandwidth, robustness, and security concerns, while local optimization methods are limited by a near information horizon as they are unable to propagate or react to information beyond their immediate vicinity. This paper inves- tigates an intermediate approach where individual sensors are able to propagate congestion information over a variable distance that is determined in real-time. This strategy consistently out-performs a naive strategy where every car simply takes the shortest path to its destination, but does worse than a simpler optimization algorithm that only incorporates local information. This is most likely because the intermediate solution directs cars along the same alternate path when attempting to free a congested area, thus creating new con- gestion along the detour. The results suggest that local information might set an upper bound on performance in models of cascading in- formation. Further work is required to confirm this observation and develop an algorithm able to join both local and global information to effectively diffuse traffic around congestion.

download document (pdf, 2,49 MB)