Home // SIMUL 2024, The Sixteenth International Conference on Advances in System Modeling and Simulation // View article

Multi-agent Dynamic Interaction in Simulation of Complex Adaptive Systems

Authors:
Hantao Hua
Feng Zhu
Yiping Yao
Wenjie Tang

Keywords: complex adaptive systems; graphical modeling and simulation; multi-agent interaction; automatic generation of dynamic data filter

Abstract:
Agent-based modeling and simulation is an effective approach to study complex adaptive systems. With the increasing scale of the simulated system, the interactions between autonomous agents become more and more complex. It is difficult to describe the dynamic interaction between agents by model code intuitively. In addition, the static interaction structure in a multiagent model leads to long running time and more memory resource consumption. Therefore, this paper proposes a method to graphically describe the dynamic interactions between different agents, named Multi-Agent interaction Graph (MAG). MAG takes agent model class as the basic element of graphical composition. The data communication between agent models is established by using subscribe/publish mechanism, and the interaction between agent model instances is accurately determined based on the dynamic attribute filtering algorithm, which is generated by the large models include Large Language Model (LLM) and Large Vision Model (LVM) automatically from the MAG. The transmission of irrelevant communication data between agent model instances is reduced, and the simulation execution time and memory consumption are reduced. Taking two scenarios as case studies, this paper proves that MAG can model the dynamic interaction between agents, and the execution time of different numbers of agents situation is reduced by 20% - 60%, which can effectively support the scalability of the number of agent model instances. Additional scenario experiments were also conducted to demonstrate the stability and generality of the dynamic attribute filtering algorithm for large model generation.

Pages: 34 to 43

Copyright: Copyright (c) IARIA, 2024

Publication date: September 29, 2024

Published in: conference

ISSN: 2308-4537

ISBN: 978-1-68558-197-8

Location: Venice, Italy

Dates: from September 29, 2024 to October 3, 2024