부산대학교 도서관

Traffic congestions have been a major problem in metropolitan areas worldwide, causing enormous economical as well as ecological damage. At the same time, in densely populated areas with high vehicle traffic, central information gathering and distribution to vehicles takes too long for providing accurate, let alone optimal routing directions (which would have to be available in due time before vehicles arrive at road intersections). Accurate information provided too late may even add to congestion problems. In this paper we present a bottom-up, multi-agent online approach termed BeeJamA (Bee-Inspired Traffic Jam Avoidance) for individual vehicle routing which, on its network communication layer, is taking advantage of our novel self — organizing network routing algorithm BeeHive/BeeAdhoc. This Swarm Intelligence based method has been largely inspired by the behavior of honey bees. As a distributed algorithm BeeJamA does not rely on global information, and scalability is not a critical issue. BeeJamA features dynamic deadlines. The quality of the algorithm has a strong impact on the acceptance rate through the drivers, for installing and operating communication features (navigators and routing-related software) as well as on driver adherence to routing directions. This, in turn, requires a high amount of flexibility for routing algorithms considering (unpredictable) resetting of destinations by drivers, making dynamic real-time reactions a critical issue. For a comprehensive and comparative realistic evaluation reflecting the aforementioned aspects/parameters we have developed a generic routing framework (GRF) which allows to run BeeJamA and other routing algorithms on different scientific or commercial traffic simulators. (Each of them serves different purposes and is therefore considerably abstracting from reality.) While we (briefly) report on extensive simulation experiments on the MAT-Sim simulator which verify BeeJamA's superior performance compared to existing models we will also outline — as part of our current research — how to create an incremental procedure for performing realistic field studies where in ever larger areas the abstract simulation is replaced, and observed, through real traffic. This imposes very strict requirements for the real-time, or online, performance of the simulator. Comprehensive results results of these altogether novel experimental investigations will be subject of upcoming publications.