基于能力与思维的协商公理体系多Agent 系统模型

【作者】 李晶；

【导师】 林毓材；

【作者基本信息】 云南师范大学 ， 基础数学， 2003， 硕士

【摘要】 随着现代计算复杂性和信息量的急剧增长，传统的集中计算模式已不能适应大型分布式信息处理的发展。而基于Agent的计算和以Agent为主体的高层交互可以满足现代计算和分布式信息处理系统的要求，而且多Agent系统在分析和建立人类交互模型和交互理论中可以发挥重要作用。所以对Agent和多Agent系统的研究已成为分布式人工智能（DAI）中的一个热点和重要的研究方向。本文的研究目标是以形式化的方法描述多Agent系统和其中的冲突解决机制——协商。本文提出了一种多Agent系统进行交互推理的形式化的体系——基于能力(Capability)与思维(Thought)的协商(Negotiation)公理体系的多Agent系统模型（CTN系统）。本文首先给出了该形式化系统的语义框架，包括行为、协商策略、环境、Agent、协商和多Agent系统的形式化定义，并对系统的全局状态、系统类、系统运行和可达全局状态做了描述。文中详细描述了语法规定和语义约束，该逻辑的语义是以一般的多Agent系统模型给出的，并且与模态逻辑的解释系统密切相关。之后给出基于模态逻辑的协商推理逻辑公理体系来形式化地描述多Agent系统，再对公理的合理性进行充分的解释。系统中Agent的行为是与其能力和权力密切相关的，Agent不会执行自己没能力执行的动作，在没有权力也没有获得授权时，Agent不会执行动作。这样的描述是合乎理性的。其次，本文阐述了协商推理逻辑的可靠性和完全性。在对公理的有效性进行证明之后，证明了协商推理逻辑的可靠性和无矛盾性；在得出的一系列命题和引理基础上，以构造的方法证明了协商推理逻辑的完全性，最后得出协商推理逻辑的完备性和整个CTN系统的无矛盾性。然后本文讨论了冲突解决协商和仲裁体制。在给出一些与冲突有关的定义后描述了仲裁策略，并用伪码描述了协商过程。系统中发生冲突时，先由冲突的Agent分组自主协商，各个Agent使用自己的策略与对方协商，当协商无法解决冲突时，使用系统仲裁策略进行仲裁，从外部解决冲突。本文通过具体示例说明此逻辑的用途后，讨论了CTN系统模型的特点：①CTN系统模型的语义是直觉的、合理的和自然的；②CTN系统模型是将Agent的能力和思维状态相结合，能够描述Agent访问环境时的能力差异、执行动作时的权力差异和系统发生冲突时的协商过程。最后提出今后的一些研究方向。更多还原

【Abstract】 Because of the rapid increasingly computing complexity and the explosive growth of information, the traditional centralized computing patterns are not adequate for large-scale distributed information processing, but agent-based computing and high-level interaction may satisfy the modern computing and distributed information processing, and multi-agent system (MAS) can play an important role in analyzing and developing the model and theory of human-interaction. Thus the research of agent and multi-agent system has become a hotspot and important domain in DAI. The purpose of this paper is to formally describe the MAS and the conflict resolving mechanism, negotiation in the MAS.In this paper, a formalism of reasoning about the interaction in multi-agent system is presented, which is MAS model based on negotiation axiom system of capability and thought (CTN system).Firstly, the semantic frame is presented, including the formal definition of actions, environment, Agent, negotiation and multi-agent system, and the global states, class of CTN system, RUN and reachable states for CTN system are also defined. The syntax rules and semantic constrains are described in detail; the semantic of this logic are given in terms of a general model of multi-agent system, closely related to the interpreted systems of modal logic. Then the axiomatic of negotiating logic is presented to give the system a formal description, and then appropriateness of Axioms is interpreted. The actions of agents are closely related to the ability and power of agents. An agent would not perform an action if it has not the ability, and if it has not the power and has not been entitled it would not either.Secondly, the soundness and completeness of negotiating logic is expounded. Followed the proving of validness of axioms is the proving of soundness and consistency for the negotiation logic. Based on a series of propositions and lemmas, the completeness of the negotiation logic is proved by construction. Then the logic completeness and the consistency of CTN system are presented.Thirdly, the conflict resolving mechanisms are discussed. After the definitions related to conflict, the arbitrage strategies are discussed, and the negotiation process is described in pseudo code. Negotiation is firstly used to resolve conflicts, and if negotiation can not result in agreement then arbitrage is used.Then some cases are used to illustrate the purpose of the logic. It shows that: (1) the semantic of CTN model is intuitive, rational and natural; (2) The CTN model is a formal description by integrating the ability and thought status of Agent, which can describe the difference among agents accessing the environment and performing actions, and the negotiation is also can be described when conflicts exist. Finally, and some future work are proposed.更多还原