有限数据条件下智能体仿真模型构建方法  

Method for Constructing Agent-Based Simulation Models with Limited Data

作　　者：叶建红[1] 高磊 孙中文 汪道歌 周碧云 YE Jianhong;GAO Lei;SUN Zhongwen;WANG Daoge;ZHOU Biyun(Key Laboratory of Road and Traffic Engineering of Ministry of Education,Tongji University,Shanghai 201804,China;BMW(China)Services Co.,Ltd.,Shanghai 200030,China;School of Automotive and Traffic Engineering,Jiangsu University,Zhenjiang Jiangsu 212013,China)

机构地区：[1]同济大学道路与交通工程教育部重点实验室,上海201804 [2]宝马(中国)服务有限公司,上海200030 [3]江苏大学汽车与交通工程学院,江苏镇江212013

出　　处：《城市交通》2024年第4期78-86,130,共10页Urban Transport of China

基　　金：国家自然科学基金项目“基于出行场景的共享交通与公共交通竞合关系测算及优化方法研究”(52172320);中央高校基本科研业务费专项资金项目“多模式交通竞合理论与分析方法研究”(22120210542);中央高校基本科研业务费专项资金项目“城市出行服务体系规划方法研究”(2023-4-ZD-01);中国博士后科学基金“基于多智能体的城市出行高精度仿真方法研究”(2023M741429)。

摘　　要：与传统的集计评估模型不同,智能体仿真模型能够捕捉个体之间的行为交互,在交通政策精细化评估中具有优势。但智能体仿真模型对数据要求较高,在实践中面临活动计划生成、参数标定和结果验证等难点。基于MATSim建立了在有限数据条件下面向应用研究的可操作、可复现的智能体仿真模型,并且以上海市拥堵收费评估场景为例详细介绍了模型的建模流程和关键技术。重点介绍基于手机轨迹数据的活动计划生成方法以及模型的参数标定和结果验证细节。结果表明,智能体仿真模型能够满足交通政策评估需求,基准场景的上海市域出行方式划分平均相对误差为8%、快速路交通量平均相对误差为17%。进而从集计和个体两个层面对仿真结果进行分析,展示了智能体仿真模型的应用潜力。Unlike traditional aggregate evaluation models,agent-based simulation models can capture the behavioral interactions among individuals,offering advantages in the refined evaluation of transportation policies.However,agent-based simulation models require high-quality data and face challenges in practice,such as generating daily activity schedule,calibrating parameters,and validating results.This paper establishes an operational and reproducible agent-based simulation model based on MATSim,designed for application research with limited data.Using the congestion charging evaluation scenario in Shanghai as an example,the modeling process and key techniques are elaborated.The focus is on daily activity schedule generation methods based on mobile phone trajectory data and parameter calibration and results validation details of the model.The results show that the agent-based simulation model is capable of meeting the transportation policies evaluation needs,with an average relative error of 8%for travel mode classification and 17%for estimation of traffic volume on expressways in the baseline scenario of Shanghai.The simulation results are analyzed at both aggregate and individual levels,demonstrating the application potential of the agent-based simulation model.

关 键 词：智能体仿真模型 MATSim 交通政策评估 出行行为分析 手机轨迹数据 上海市 

分 类 号：U491[交通运输工程—交通运输规划与管理]