In this paper, a meta-model called IRM4MLS, that aims to be a generic ground to specify and execute multi-level agent-based models is presented. It relies on the influence/reaction principle and more specifically on IRM4S (Michel, 2007a,b). Simulation models for IRM4MLS are defined. The capabilities and possible extensions of the meta-model are discussed.

International audienceThis paper describes four design patterns that aim at systematizing and simplifying the modelling and the implementation of multi-level agent-based simulations. Such simulations are meant to handle entities belonging to different, yet coupled, abstractions or organization levels. The patterns we propose are based on minimal typical situations drawn from the literature. For each pattern, we present use cases, associated data structures and algorithms. For genericity purposes, these patterns rely on a unified description of the capabilities for action and change of the agents. Thus, we propose a precise conceptual and operational framework for the designers of multi-level simulations

1. Dans le cadre de cet article, le mot complexe n'est pasà prendre au sens de « compliqué » mais se réfèreà l'idée de nombreuses entités en interaction. La notion de système complexe est définie plus précisément dans la section 2.1.