基于强化学习的异步动态定价算法  

Asynchronous dynamic pricing algorithms based on reinforcement learning

作　　者：王金田[1,2] 唐昊[1,3] 程文娟[1] 毕翔[1,3] 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009 [2]安徽省审计厅,安徽合肥230001 [3]教育部安全关键工业测控技术教育部工程研究中心,安徽合肥230009

出　　处：《系统工程学报》2011年第5期664-670,共7页Journal of Systems Engineering

基　　金：教育部留学回国人员科研启动基金资助项目(教外司2008890);安徽省自然科学基金资助项目(070416242;090-412046);安徽高校省级自然科学研究重点资助项目(KJ2008A058;KJ2010A256)

摘　　要：研究电子零售市场上两个销售商在彼此没有信息交互情况下的异步动态定价问题.基于性能势理论,建立了同时适用于平均和折扣两种优化准则下的异步定价策略的Q学习和WoLF-PHC算法,通过一个数值例子比较了相关算法的学习优化效果.仿真结果表明,Q学习和WoLF-PHC算法都能较好地解决异步动态定价问题,但由于后者采用混合策略和可变学习率,故能更好地适应环境变化,并具有更好的学习优化效果.This paper studies the asynchronous dynamic pricing problems of two sellers in an electronic retail market without information exchange between them. Based on the concept of performance potential, a Q- learning algorithm and a WoLF-PHC algorithm are proposed to yield the asynchronous pricing policies that are suitable for either average- or discounted-reward criteria. A numerical example is used to compare the learning performance of different algorithms. The simulation results show that both the proposed algorithms are effective for solving asynchronous dynamic pricing problems. Compared with Q -learning, the WoLF-PHC algorithm is more adaptable to environmental change due to the utilization of mixed strategies and variable learning rates.

关 键 词：异步动态定价 多AGENT 性能势 WoLF—PHC算法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]