城市交叉口Agent间的多遇交互历史学习协调方法  

A Multi-Interaction History Learning Approach for Coordination of Urban Intersection Agents

作　　者：夏新海[1] 许伦辉[1] 

机构地区：[1]华南理工大学土木与交通学院,广东广州510640

出　　处：《华南理工大学学报（自然科学版）》2011年第3期114-119,共6页Journal of South China University of Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60664001)

摘　　要：为信号控制的城市道路交叉口定义一个Agent结构模型,利用双人对策Nash平衡理论构建了城市交叉口Agent间的多遇交互模型,每一交叉口Agent与相邻交叉口Agent进行多次交互学习,根据选择策略获得的效用值来更新它的混合策略.利用记忆因子δ、学习概率α、交叉口交通流变化概率βi等参数分析了交叉口Agent间的循环学习协调过程.设计了交叉口Agent多遇交互历史学习协调算法,在此算法里交叉口Agent可以通过对其他相邻交叉口Agent以往历史交互行为特别是最近的历史行为的记忆学习达到协调.以数个交叉口相连接的干道为例分析了δ、α、βi等参数对算法性能的影响.通过干道上交叉口交通信号协调的实例分析,证明了该协调学习方法的有效性.Proposed in this paper is a multi-interaction history learning approach for the coordination of urban intersection agents.In the investigation,first,each signalized intersection is defined with an Agent controller.Next,a multi-interaction model for urban intersection Agents is built based on the two-person Nash equilibrium game theory to make each intersection Agent to perform multi-interaction learning with its neighbours and to update its mixed strategy according to the utility value of the selected strategy.Then,the iterative interaction learning process of intersection Agents is analyzed by using the parameters such as memory factor δ,learning probability α and local traffic change probability βi at each intersection.A multi-interactive history learning algorithm was constructed.In the proposed algorithm,intersection Agents coordinate by taking into consi-deration all history interactive information（especially the recent one） coming from neighbouring intersection Agents.Finally the effects of parameters δ,α and βi on the algorithm performance is also analyzed by the experiment of traffic signal control at some connected intersections.The results show that the proposed coordinative learning approach is effective.

关 键 词：AGENT 交通信号控制 学习 协调 

分 类 号：U491.54[交通运输工程—交通运输规划与管理]