Information Technology Reference
In-Depth Information
well-designed composition of regions is therefore necessary to improve
autonomy of the agents and enhance scalability.
2.3.4 DEVS and variable structure models
The DEVS (discrete event system specification) formalism allows the
description of event-based systems in a formal way [151]. DEVS sup-
ports hierarchical modeling, thus coupling existing models to create
more complex models. As Uhrmacher points out, 'original DEVS
assumes, as do other formal discrete event simulation approaches, a
static structure of the system' [131], [17]. This static model struc-
ture is in sharp contrast to the very idea of agent-based models,
which assumes lots of entities interacting in manifold, not completely
predetermined ways and changing their behavior pattern if necessary.
Although DEVS provides a strict formalism and is well-understood
for modeling discrete event systems, the possibilities to describe agent-
based models using DEVS seem limited. This is due to several reasons:
First and foremost, it seems hardly possible, if at all, to describe all
possible changes in the structure of a model in advance. As almost
all aspects of an agent-based model (like the environment or the
agents and their embedding) are subject to manifold changes, the
chances of defining all these structural changes in advance are low.
Secondly, agent-based models are inherently modeling the micro-level
and depend heavily on the communication and interaction between the
agents. Explicitly modeling all possible communication and interaction
relations between the agents of a specific model is about as hard as
defining all possible structural changes.
Therefore, the current assessment is that DEVS is not ideally suited
for the description of agent-based models as a whole. On the other
hand, DEVS may be well suited for describing the behavior of single
agents. Furthermore, the strict distinction made by DEVS between
model, abstract simulator and actually implemented simulator is
worth mentioning.
With the limitations of DEVS in mind, several extensions of the
original DEVS formalism have been proposed to support models with
