Information Technology Reference
In-Depth Information
Learning Dynamic Adaptation Strategies
in Agent-Based Trac Simulation Experiments
Andreas D. Lattner
1
,Jorg Dallmeyer
1
,andIngoJ.Timm
2
1
Information Systems and Simulation, Institute of Computer Science
Goethe University Frankfurt, P.O. Box 11 19 32, 60054 Frankfurt, Germany
2
Business Informatics I, University of Trier, D-54286 Trier, Germany
Abstract.
The increase of road users and trac load has lead to the
situation that in some regions road capacities appear to be exceeded
regularly. Although there is natural capacity limit of roads, there exist
potentials for a dynamic adaptation of road usage. Finding out about
useful rules for dynamic adaptations of trac rules is a costly and time
consuming effort if performed in the real world. In this paper, we intro-
duce an agent-based trac simulation model and present an approach to
learning dynamic adaptation rules in trac scenarios based on supervised
learning from simulation data. For evaluation, we apply our approach
to synthetic trac scenarios. Initial results show the feasibility of the
approach and indicate that learned dynamic adaptation strategies can
lead to an improvement w.r.t. the average velocity in our scenarios.
Keywords:
Agent-based trac simulation, supervised learning, adap-
tation strategies.
1
Introduction
In the past decades, the number of road users and tra
c load has constantly
increased. This has lead to the situation that road capacities in some areas
seem to be exceeded and congestions occur regularly. Although there exists a
natural limit of road users, a road infrastructure can deal with, there are still
potentials for optimization of road usage and activities to prevent or reduce
trac jams, e.g., by re-routing of road users or by introducing trac regulations
aiming to avoid risky situations. Current solutions provide already means in
order to dynamically reroute road users with GPS-based navigation systems
utilizing trac information as well as to adapt trac regulations by setting a
speed limit if high trac load is present or if congestions have already occurred
in the subsequent course of the road. Different measurement techniques are
used to get online information about the current trac status at certain control
points [11].
Finding out about useful rules for dynamic adaptations of trac rules or
trac rerouting is a costly and time consuming effort if performed in the real
world. Although certain general experiences about trac can be utilized in
order to set up such rules, emergent effects of changes are not obvious in all
