Skalierbare agentenbasierte Simulation : Werkzeuge und Techniken zur verteilten Ausführung agentenbasierter Modelle
Year: 2009
Type: dissertation
Abstract: \n \nPawlaszczyk, Dirk: \nSkalierbare agentenbasierte Simulation : Werkzeuge und Techniken zur verteilten \nAusführung agentenbasierter Modelle / Dirk Pawlaszczyk \nIlmenau : Univ.-Verl. Ilmenau, 2009. - XVI, 327 S. \nISBN 978-3-939473-59-6 \n \nResearch on systems of autonomous agents, called multiagent systems (MAS), has \nreceived much interest in the domain of (distributed) artificial intelligence \nin recent years. MAS are most suitable for the development of distributed \napplications with uncertain and a dynamically changing environment. Simulation \nis the imitation of a systems behaviour and structure in an experimental model \nto reach findings which are transferable to reality. In multiagent-based \nsimulation (MABS) real world systems are modelled using multiple agents. \nThe system emerges by interaction of the individual agents as well as their \ncollective behaviour. Agents send messages with respect to some communication \nprotocol and are disposed at some discrete point in time. In this context a \nsoftware agent is defined as a program that acts autonomously, communicates \nwith other agents, is goal-oriented (pro-active) and uses explicit knowledge. \nAgent-based modelling and simulation seems to be the right tool for domains \ncharacterized by discrete decisions and distributed local decision makers. MABS \nis an appropriate method if we wish to understand the evolution of a \ndistributed system, minted with non-linear dynamics. As we wish to resemble \nthis complex real world system, there is an increasing demand to simulate large \nmodels, composed with hundreds or even more agents. With growing complexity, \nthe scalability of a simulation environment becomes a crucial measure of its \nability to cope with the complexity of the underlying system. To simulate an \nincreasing number of entities, the underlying simulation system needs to be \nscaleable. Since agent-based simulation moves out of the academic world towards \ncommercial applications it becomes certainly demanding to investigate in \nscalability issues. The objective of this thesis is to discuss an approach for \nscalable simulation, which considers the deviations of agent-based simulation \napproach effectively. Therefore, we outline some commonly used notions of \nscalability and begin to examine what scalability means within the context of \nagent-based simulation. Furthermore we introduce the concept of a scalable time \nmanagement service. With respect to the experimental results it is argued that \nparallel and distributed simulation can help to ensure scaleable simulation of \nagent based models.
Author: Dirk Pawlaszczyk
Language: German
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W1661752833