CASIA OpenIR  > 毕业生  > 硕士学位论文
Thesis Advisor王珏
Degree Grantor中国科学院自动化研究所
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Abstract本文包括三部分内容:(1)基于意图不完全知识的Pareto最优协 商算法;(2)基于意图不完全知识的结盟机制;(3)基于Stacke1herg 平衡的agent类型划分。 我们在[王96]的基础上,针对其协商方法的不足,重新定义意图, 基于新的意图给出一种生成Pare to合理协商集的不完全知识协商算法(协 商A):在协商集上对联合意图进行协商(协商B)时,为了增加协商的 灵活性,我们假设agent的效用值可以转移,即一个agent为了达到自己 偏好程度高的状态,可以把自己效用值的一部分支付给另一个agent作为 补偿,从而避免agent双方因为不同状态间的效用值相差太大而协商失败: 并且还证明在协商A、B过程中可以防止I~IV类欺骗。 在基于意图不完全知识协商的基础上,我们建立了一个基于意图不完 全知识的结盟机制。每一轮结盟丌始时,所有agent两两协商。agent根 据协商结果的偏好程度对其它agcnt排队,尽量和偏好程度最高的agent 结成一个联盟。在联盟中,两个agent不是平等的伙伴关系,联盟中为达 成协议而把自己的效用值的一部分支付出去的agent成为联盟的管理者 (managing agent或leading agent),另一个agent则称为随从 (following agent),其中管理者是联盟的全权代表。联盟的能力是联 盟中所有agent能力的总和,其对世界状态的评价值及其它agent的偏 好程度则和管理者一样。因此i联盟可以看成是作为管理者的agent的 个扩展,管理者代表联盟参加下一轮的结盟。因此,每一轮结盟都可以看 成是agent的两两协商,从而火人地简化了问题的复杂性。我们的结盟机 制所生成的联盟是稳定的。 由于在基于意图不完全知识的结盟机制中,agent不是平等的关系,其 中leading agent是联盟的全权代表,它处于比following agent 更强的 地位,following agent必须在接受leading agent的策略的前提下作出 理性的反应,我们利用对策论中的Stackelberg平衡对这种情况下的agent 类型进行划分。我们讨论了leading agent和following agent的两种类 型的策略(合作类型和自私类型)。一般而言,自私类型策略是稳定的但 不是高效的,而合作类型策略则是高效的但不能保证是稳定的。但当 leading agent和following agent的反应策略集合均是只有一个元素时, 则自私类型策略等同于合作类型策略。
Other AbstractThis thesis can be devided into three major parts: (1) intention-Based Imcomplete Knowledge Pareto-Optima] Negoation; (2) Intention-Based Imcomplete Knowledge Coalition;(3)Agent Interaction Strategy Types Based on Stackelberg Equibrium. in order to overcome the drawback of the negotiation method in [Wang96],we give a negotiation method based on our new definition of intention. Having agreed on a Pareto Optimal negotiation set in the first step of negotiation, one agent can tranfer on part of its utility to the other agent in order to select its prefering state in the negotiation set as the their mutual intention. Four kinds of possible cheat in the negotiating process are discussed and can be advoided in our mechanism. On the base of Intention-Based Incomplete Knowledge Negotiation, we present an Intention-Based Incomplete Knowledge Coalition Mechanism. At the beginning of' each round of coalition, all the agents negotate with each other and try to form a coalition with the agent with the highest preference degree. The two agents in the coalition are not symmetric and the agent who transfers one part of its utility to its partner acts as the manager of the coalition. The managing agent has the authority to negotiafe on behalf of the coalition to form larger coalitions and the coalition can be viewed as an meta-agent, i.e.,the managing agent with extended ability. On salient feature of our mechanism is that the forming coalition is stable. We present two types of strategy(one selfish and the other cooperative) for the interaction between the managing agent and the following agent where the roles of the two agents are not symmetric, i.e.,the managing agent is in the powerful position to enforce its strategy on the following agent and the following agent must take into account the managing agent's claimed stategy before it reacts rationally. Generally, the selfish strategy is stableand the cooperative is more effient. The selfish strategy is equal to thecoperativeone if the rational strategy set for the managing agent and the coorresponding rational reaction strategy set of following agent are both singletons.
Other Identifier431
Document Type学位论文
Recommended Citation
GB/T 7714
阮永韶. 多agent系统的协商和结盟[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,1997.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[阮永韶]'s Articles
Baidu academic
Similar articles in Baidu academic
[阮永韶]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[阮永韶]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.