文章编号： 1672-5328（2025）06-0124-03

单振宇
(西南交通大学交通运输与物流学院，四川成都610000)

摘要： 选取来自国际学术期刊的论文，以概述形式对城市交通理论方法、实证分析等学术研究成果 进行总结性介绍，旨在增强城市交通业界和学界对国际学术动向和研究热点的关注，促进学术交 流。针对传统基于智能体的模型(Agent-Based Model, AgBM)在行为表征、灵活性和数据依赖方面的 局限性，《迈向基于LLM智能体的交通系统建模：一个概念性框架》一文系统性地提出了一种基于 大语言模型(LLM)智能体的交通建模新框架。该框架将LLM 智能体作为人类出行者的“数字代 理”，通过构建包含身份、特征和记忆系统的精细化智能体画像，并设计感知、决策和行动模块， 使其能够模拟复杂、动态的出行决策过程。概念验证仿真结果显示，该框架在提升行为现实性、数 据利用效率和模型灵活性方面具有巨大潜力。该论文研究成果为出行需求建模开辟了一条新的研究 路径。

关键词： 交通系统建模；出行行为；出行需求建模；基于智能体的仿真；大语言模型

中图分类号： U491

文献标识码：A

Academic Dynamics on LLM-Agent-Based Modeling of Transportation Systems

SHAN Zhenyu
(School of Transportation and Logistics, Southwest Jiaotong University, Chengdu Sichuan 610000, China)

Abstract： A review of selected papers from international academic journals is presented to summarize research findings, theoretical approaches, and empirical analyses of urban transportation. The aim is to enhance the communication between industrial and academic fields in urban transportation, highlight international research focuses, and promote academic exchange. The paper Toward LLM-Agent-Based Modeling of Transportation Systems: A Conceptual Framework addresses the limitations of traditional Agent-Based Model (AgBM) concerning behavioral representation, flexibility, and data dependency. It systematically proposes a novel modeling framework based on Large Language Model (LLM) agents. This framework conceptualizes LLM as“digital agents”for real-world travelers. By constructing sophisticated agent profiles that incorporate identity, traits, and memory systems, alongside modules for perception, decision-making, and action, it enables the simulation of complex and dynamic travel decision-making processes. The proof-of-concept results indicate that this framework holds significant potential for enhancing behavioral realism, data utilization efficiency, and model flexibility. The findings of this study pioneer a new research avenue for travel demand modeling.

Keywords： transportation system modeling; travel behavior; travel demand modeling; agent-based simulation; large language model