地理学报

地理学报 ›› 2009, Vol. 64 ›› Issue (4): 445-455.doi: 10.11821/xb200904007

• 产业发展与消费行为 • 上一篇    下一篇

多代理人系统在商业街消费者行为模拟中的应用 ———以上海南京东路为例

朱玮1, 王德2, Harry TIMMERMANS1   

  1. 1. 埃因霍芬科技大学城市规划组, 埃因霍芬5600MB, 荷兰;
    2. 同济大学城市规划系, 上海200092
  • 收稿日期:2008-10-15 修回日期:2009-01-29 出版日期:2009-04-25 发布日期:2009-04-25
  • 通讯作者: 王德(1963-), 教授, 博导。中国地理学会会员(S110005378M), 主要研究方向： 城市研究方法论、大城市空 间结构、城市与区域规划领域的教学与研究。E-mail: dewang@tongji.edu.cn
  • 作者简介:朱玮(1978-), 博士, 荷兰埃因霍芬科技大学城市规划组。博士后， 德国马普人类发展研究所适应性行为与 认知中心。主要研究方向： 决策和行为建模、消费者空间行为、模型方法在城市规划、交通、零售业及其 他相关领域的应用。E-mail: zhu@mpib-berlin.mpg.de
  • 基金资助:

    国家自然科学基金项目(40871080)

Applying Multi-Agent Systems in the Simulation of Consumer Behavior in Shopping Streets: The Shanghai East Nanjing Road Case

ZHU Wei1, WANG De2, Harry TIMMERMANS1   

  1. 1. Eindhoven University of Technology, Urban Planning Group, Eindhoven 5600MB, The Netherlands;
    2. Department of Urban Planning, Tongji University, Shanghai 200092, China
  • Received:2008-10-15 Revised:2009-01-29 Online:2009-04-25 Published:2009-04-25
  • Supported by:

    National Natural Science Foundation of China, No.40871080

PDF (PC)

赞

摘要：

用模型方法研究消费者行为经历了从集合模型到个体模型的过程。随着计算机技术的 进步, 近十余年来多代理人技术得到了长足的发展。当前, 将消费者个体行为模型和多代理 人系统相结合来模拟商业街消费者活动, 已经成为相关研究和实践领域的前沿方法。其主要 优点在于能够使相关人员以单个消费者为基础, 整合复杂的相关因素, 预测整体或局部商业 空间的动态运行效果, 作为规划开发的依据。本文介绍由笔者基于NetLogo 开发的一个模拟 商业街消费者行为的多代理人模拟平台, 包括系统的建构和模拟的流程。该模拟系统的核心 由四个基于有限理性原理的消费者决策模型构成： 回家决策、方向选择决策、休息决策、商 店光顾决策。然后应用这个平台于研究上海南京东路消费者行为的实例。该数据收集于2007 年, 现场记录了消费者在南京东路的活动过程, 并被用于拟合行为模型。将模拟的代理人个 体行为集合后, 于三方面同观察到的消费者实际行为作比较, 包括从事不同活动的消费者数 量随时间的变化、在各街道段中活动的消费者数量随时间的变化, 以及单个商店中的消费者 人次和延时，同时分析了模拟中的问题。结果显示该模拟平台总体上能够较好地预测消费者 集合行为。

关键词: 多代理人系统, NetLogo, 消费者行为, 决策模型, 模拟, 商业街, 南京东

Abstract:

Modeling consumer behavior has experienced from using aggregate models to individual-based models. During the latest decade, multi-agent technique has been extensively developed with the help of the ever-advancing computer technology. Currently, the method of combining individual-based consumer behavior models and multi-agent simulation systems has become the state-of-the-art method in related research and practical fields. The major advantage of such type of method is that it enables the related professionals to integrate the complexity of various aspects, predict the dynamic processes of the whole or part of the shopping space, and support planning and development based on the behavior of individual consumer. The paper introduces a multi-agent simulation platform developed by the authors based on NetLogo, including the construction of the platform and the procedures of the simulation. The core of the simulation system consists of four major consumer decision models: go-home decision, direction choice decision, rest decision, and store patronage decision, which are constructed differently from conventional choice models based on principles of bounded rationality. The simulation platform is applied to the study of consumer behavior in East Nanjing Road as a validation to the model system. A survey was carried out in the street in 2007. The data recorded the activity diary of consumers and were used for estimating the models. The simulation is carried out using the given distributions of consumers' entry locations and start time, to replicate the observed behavior. Aggregate consumer behavior is more concerned in practice. Therefore, after twenty times of simulation for reducing random fluctuations, individual agent behavior is aggregate. Three aspects of the aggregate simulated behavior are compared with the observed aggregate behavior. The first aspect is the number of consumers conducting different activities overtime, including the total number of consumers in the street, the number of consumers in the stores, the number of consumers taking rests, the number of consumers who are walking, and the number of consumers who have gone home. The second aspect is the number of consumers in street segments over time. The third aspect is the number of consumer visits and duration in individual stores. The results show that in general the simulation platform can predict the aggregate consumer behavior well. Particularly, the activity-time distributions are simulated very well except that the walking behavior is simulated poorly mainly due to the small number of observations. The number of consumers in street segments are simulated not well enough although the general trend in segments with a large number of observations are captured. This means the proposed model system can capture the aspatial behavior better than spatial behavior. This might be due to the fact that the complexity of consumer spatial behavior was affected by many environmental factors and personal variations, and however, it was only modeled using limited explanatory variables. Further studies are needed to be done on consumer behavior in relation with the environment.

Key words: multi-agent system, NetLogo, consumer behavior, decision model, simulation, shopping street, East Nanjing Roa