多智能体Q学习在多AUV协调中的应用研究  

Research on application of multi-agent Q-learning algorithm in multiAUV coordination

作　　者：严浙平[1] 李锋[1] 黄宇峰[1] 

机构地区：[1]哈尔滨工程大学自动化学院,黑龙江哈尔滨150001

出　　处：《应用科技》2008年第1期57-60,共4页Applied Science and Technology

基　　金：黑龙江省博士后科研启动基金资助项目(LHK-04010)

摘　　要：分析了近年来提出的几种主要单、多智能体强化算法,提出一种多智能体Q学习算法,并将它用于多AUV协调中.此算法融合了Nash-Q、CE-Q及WoLF-PHC的算法思想或表达形式,程序简单、容易计算,并且具有很好的收敛性.多AUV协调控制仿真实验结果表明这个算法是有效的.Several major reinforcement learning algorithms for single agent and multi-agents, that are developed in recent years, are investigated, compared and analyzed. A multi-agent Q-learning algorithm is proposed and used in muhi-AUV coordination. This algorithm combines the thoughts or expressions of Nash-Q, CE-Q and WoLF-PHC algorithms, so that the program is simple, the computation is easy, and the convergence rate is high. Experiment results of muhi-AUV coordination and control prove the effectiveness of this algorithm.

关 键 词：多智能体 Q学习 NASH均衡 多AUV 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]