基于Q-强化学习的多Agent 协商策略及算法.pdfVIP

第36 卷 第17 期 计 算 机 工 程 2010 年9 月 V .36 No.17 Computer Engineering September 2010 ol ·人工智能及识别技术· 文章编号：1000—3428(2010)17—0198—03 文献标识码：A 中图分类号：TP18 基于 强化学习的多 协商策略及算法 Q- Agent 隋 新，蔡国永，史 磊 (桂林电子科技大学计算机与控制学院，桂林 541004) 摘 要：针对传统Agent 协商策略学习能力不足，不能满足现代电子商务环境需要的问题，采用Q-强化学习理论对Agent 的双边协商策略 加以改进，提出基于Q-强化学习的Agent 双边协商策略，并设计实现该策略的算法。通过与时间协商策略比较，证明改进后的Agent 协商 策略在协商时间、算法效率上优于未经学习的时间策略，能够增强电子商务系统的在线学习能力，缩短协商时间，提高协商效率。 关键词：Q-强化学习；多Agent ；协商策略 Strategy and Algorithm of Multi-Agent Negotiation Based on Q-reinforcement Learning SUI Xin, CAI Guo-yong, SHI Lei (School of Computer and Control, Guilin University of Electronic Technology, Guiling 541004) 【Abstract 】As lack of sufficient learning ability in traditional negotiation strategy of Agents, Agents’ techniques are still unable to meet the needs of modern E-commerce. Aiming at this problem, Q-reinforcement learning theory is adapted to improve the bilateral negotiation strategy of Agents and the corresponding negotiation algorithm is designed to achieve the negotiation strategies. Comparing with the negotiation strategy of time strategy, the proposed Agent negotiation strategy is better than time strategy in terms of negotiation time and algorithm efficiency. It shows that algorithm can strengthen online learning ability of E-commerce system, shorten the negotiation time, and improve the negotiation efficiency. 【Key words 】Q-reinforcement learning; multi-Agent; negotiation strategy 1 概述 协商。 Agent 技术是一种代替人进行协商和决策的工具。传统 为提高原有协商策略的智能性和灵