Adaptive agent abstractions to speed up spatial agent-based simulations

Authored by Abbas Sarraf Shirazi, Timothy Davison, Sebastian von Mammen, Joerg Denzinger, Christian Jacob

Date Published: 2014-01

DOI: 10.1016/j.simpat.2013.09.001

Sponsors: National Science and Engineering Research Council of Canada (NSERC)

Platforms: No platforms listed

Model Documentation: Pseudocode Other Narrative Flow charts Mathematical description

Model Code URLs: Model code not found

Abstract

Simulating fine-grained agent-based models requires extensive computational resources. In this article, we present an approach that reduces the number of agents by adaptively abstracting groups of spatial agents into meta-agents that subsume individual behaviours and physical forms. Particularly, groups of agents that have been clustering together for a sufficiently long period of time are detected by observer agents and then abstracted into a single meta-agent. Observers periodically test meta-agents to ensure their validity, as the dynamics of the simulation may change to a point where the individual agents do not form a cluster any more. An invalid meta-agent is removed from the simulation and subsequently, its subsumed individual agents will be put back in the simulation. The same mechanism can be applied on meta-agents thus creating adaptive abstraction hierarchies during the course of a simulation. Experimental results on the simulation of the blood coagulation process show that the proposed abstraction mechanism results in the same system behaviour while speeding up the simulation. (C) 2013 Elsevier B.V. All rights reserved.

Tags

agent-based simulation Optimization Abstraction Online learning