摘要
AODE是我们研制的一个面向agent的智能系统开发环境,本文以AOD为平台研究了多agent环境下的协商与学习,本文利用协商-协商过程-协商线程的概念建立了多边-多问题协商模型MMN,该协商模型支持多agent环境中的多种协商形式及agent在协商过程中的学习,系统中的学习agent采用状态概率聚类空间上的多agent强化学习算法,该算法通过使用状态聚类方法减少Q值表存储所需空间,降低了经典Q-学习算法由于使用Q值表导致的对系统计算资源的要求,且该算法仍然可以保证收敛到最优解。
Negotiation and learning in multi-agent environment in AODE, which is an agent oriented development environment for intelligent software system, is studied. A negotiation model called MMN is provided based on the concept of negotiation - negotiation process - negotiation thread. This model supports many types of negotiation and learning during the negotiation process. Reinforcement learning with soft state aggregation is adopted in AODE. As a result, the Q-learning algorithm used in AODE needs less storage space for Q-value than the standard Q-learning which uses the Q-value look-up table. And the Q-learning algorithm used in AODE is guaranteed to converge to its optimal solution under specified conditions.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2001年第3期347-351,共5页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金
高等学校博士点基金