情感多主体系统的设计与实现  

Design and Implementation of an Emotional Multi-agent System

作　　者：黄国锐[1] 曹先彬[1] 徐敏[1] 王煦法[1] 

机构地区：[1]中国科学技术大学计算机科学技术系,安徽合肥230026

出　　处：《小型微型计算机系统》2003年第7期1268-1271,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金 ( 699710 2 2 )资助 ;国家博士点 ( 19990 3 5 82 6)基金资助

摘　　要：多 Agent在复杂多变的环境中行为的自组织是当前人工智能研究的一个热点 .本文借鉴生物系统控制的理论 ,设计了一种多 Agent的体系结构 ,在此结构中包含有遗传、神经和内分泌控制子系统 .本文重点论述了内分泌控制子系统的作用 ,提出了用情感进行行为学习的新方法 .在此方法中 ,神经系统的作用是记忆和行为决策 ,其行为决策的效果通过内分泌系统进行反馈 ,从而避免了神经网络的自学习 ,使得算法具有较好的求解性能 .为了验证该算法的有效性 ,本文做了倒立摆控制的仿真实验 .Self organization of behaviors in dynamic environment is a hot topic in AI. Motivated by the biological system control mechanism, this paper designs an agent architecture that includes genetic, neural and endocrine control sub system. We discuss detailedly the endocrine control sub system that has been somehow neglected. The paper further puts forward a new method of using emotion to learn to behave. In this algorithm, the function of neural system is memorizing and decision making, and the result of decision is fed back to neural system through endocrine system. So the algorithm avoids the self study of neural system and gets better performance. In order to prove the validity of this method, we use it to control an inverted pendulum. Experimental evidence of successful controlling is also presented.

关 键 词：神经网络 遗传算法 内分泌控制 情感学习 倒立摆 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]