Towards a Formal Semantics of Event-Based Multi-agent Simulations

Abstract

The aim of this paper is to define a non-ambiguous operational semantics for event-based multi-agent modeling and simulation, applied to complex systems. A number of features common to most multi-agent systems have been retained: 1) agent proactive as well as reactive behavior, 2) concurrency: events can arrive simultaneously to an agent, an environment or any simulated entity and the actual change only depends on the target according to the influence/reaction paradigm [1], 3) instantaneity: if reaction takes time, perception as well as information diffusion is instantaneous and should be processed separately, 4) structure dynamics: the interaction structure (who is talking to whom) changes over time, and the agents as well as any simulated entity may be created or destroyed in the course of the simulation.

For each of these features, a solution inspired by the work on $\mathit{DEVS}$ (Discrete EVent Systems, [2]) is proposed. Proactive/reactive behavior is naturally taken into account by $\mathit{DEVS}$. Concurrency is dealt with using $/\!/\!\!-\!\!\mathit{DEVS}$ (in [2]), a variant of the pure $\mathit{DEVS}$. Instantaneity is managed by distinguishing the physical events producing state transitions and the logical events realizing only perception and information diffusion. The structure dynamics is achieved by using a variant of ρ-$\mathit{DEVS}$ (cf. [3]) where the expressiveness allows to manage hierarchical structures. The operational semantics is given as abstract algorithms and the expressive power of this formalism is illustrated on a simple example.