【作者】 刘勇；

【导师】 程代杰； 曹泽翰；

【作者基本信息】 重庆大学 ， 计算机软件与理论， 2003， 博士

【摘要】 基于Agent的概念，人们提出了一种新的人工智能定义：“人工智能是计算机科学的一个分支，它的目标是构造能表现出一定智能行为的Agent”。所以，Agent的研究应该是人工智能的核心问题之一。斯坦福大学计算机科学系的Hayes-Roth在IJCAI’ 95的特邀报告中谈到：“智能的计算机Agent既是人工智能最初研究的目标，也是人工智能最终的目标。”现在，关于Agent和多Agent系统的研究已成为人工智能研究的热点，不仅受到了人工智能研究人员的关注，也吸引了数据通信、人机界面设计、机器人、并行工程、电子商务等各领域的研究人员的兴趣。Agent的理论和技术，特别是多Agent系统的理论和技术，为开放和分布式系统的分析、设计和实现提供了一个崭新的途径，被称为软件开发的又一重大突破。本文的目的在于深入研究Agent各意识概念内含和关系，研究出能指导工程应用的Agent意识模型和多Agent通信、协作的方法，对Agent理论应用于飞行器遥测专家系统设计作有益的探索。本文研究的方法为：运用心理学和组织行为学理论指导Agent意识概念和模型的研究，用非标准世界的模态逻辑为工具描述Agent意识概念和模型，分析各种意识概念的性质和相互关系，并且用典型问题和事例验证说明研究结果的有效性。借鉴和运用市场学方法来研究多Agent系统的组织方法，用算法形式描述相应的组织协调机制。本文的创新之处主要体现在以下几个方面： (1) 把信念概念进行了拓展，提出了知识性信念和可实现信念概念，知识性信念具有知识的进化特性和继承特性。两种信念概念的语义解释避免了逻辑全知问题，也不存在意识模型在实际中没有对应的问题。(2) 提出了逻辑理性和效用理性结合起来的语义解释模型，揭示了目标和意图产生的原因，解决了理性Agent意识模型不能给出行动决策的问题。(3) 提出了MCU意识计算模型。该模型与以往的Agent意识模型研究相比，较好地克服了意识概念关系不够明确问题, 提出的模型具有进化特性。将逻辑理性和效用理性统一在一个模型中，而效用理性是Agent价值观的体现，从而使Agent计算具有了社会性。该模型较易指导实际Agent和MAS的设计和分析。(4) 提出了基于MCU模型的通信交互语言和交互模型，给出了该模型的语义解释，提出了基于意识模型的Agent意识状态演化规则。与以往Agent通信交互语言和模型相比，本研究的通信与交互语言更为简洁，同时通信语言不存在模糊<WP=6>性和二义性，更适合于协作类型的MAS。该语言支持的MCU模型的演化策略规则使本MAS系统更能适应开放、动态的环境。(5) 在分析典型的合同网协议和合作规划系统基础上，提出了在MCU模型上基于合同网的合作规划系统，克服了合同网协议中招标者缺乏任务分解知识和合作规划系统利用先验知识判断规划可行性的缺点，及确定任务局部性的缺点，提高了整个系统的工作效率和Agent合作成功的可能性，较好地适应了开放和动态的环境。(6) 飞行器遥测专家系统研究设计。针对遥测数据类型多、处理方式复杂特点，设计了框架与规则、模型、过程相结合的面向对象的知识表示方法。为提高故障定位精度和扩大预测故障后果范围，建立了故障图诊断和预测模型。运用MCU模型建立专家结点，运用研究的通信交互语言和意识状态演化规则及协作和组织模型设计和实现了飞行器遥测专家系统。与其它类似系统相比，具有故障定位能力强，预测故障影响价值大、坚固性好特点，并具有扩展能力，适应开放动态的环境。更多还原

【Abstract】 Agent theory and technique, especially multi-agent theory and technique, provide a new approach to the analysis, design, and implementatoin of open and distributed systems. That is called an important breakthrough of the software developping. For current multi-agents theory escape from the actual problem, this thesis research relations of the agent mental state, bring up the agent mental model, cooperation and communication approach to guide engineering application. Using the theory, we have established an open distributed expert system. This thesis is focused on the research of multi-agent theory and application,which include five part of contents: In section 1, introductoin;in section 2 ,we Analysis the problem and present condition of Agent research, point out the agent theories and technique development trend;in section 3,we provide an agent mental model with utility;in section 4,we give a method of communication based on MCU model;insection 5;in section 6 we present a cooperation and oganization method of multi-agent systems;in section 7, we give some conclusions and direction for future research. The importants innovation of this thesis includes these contents:(1) Using none standard modal logic as formal tool, we research Agent and MAS model, and formally define the belief, intention, plan, worth concept of income etc., and extend the Belief into knowledge belief and achievement belief.Knowledge belief is knowledge of an Agent, it presents evolving and inheriting characteristic. Achievement belief is an event which can be established in the future, and is aware of personality or goal.Useing non standard world of the modal logic describe two kinds of Beliefs, Possible world is a state that Agent can attain or realize, it can avoid the logical omniscience problem and the logical side effect problem.(2) we present a semantic model combined with utility. On the language to present the parameter used in decise and On the semantic to present two functions of utility and cost between two worlds which be transferred by performing plan,. the problem has been solved that mental model can not make decision of activity. We Announce reason that goal and intention produce. Goal is state of that an agent can appear on in the current world. Intention is a goal which has maximum utility in the<WP=8>current world.. (3) We present MCU mental model. Compared with other model, the model overcame the problem that mental concept relation is not enough clear. There is evolving characteristic in the model. Unifing the utility reasonableness and the logic reasonableness in a model, the utility reasonableness is a worth of agent, so agent calculation have the social. The model can be easy used in MAS design and analysis. We present the expanding rule of mental state.(4) We present the communication language model based on the MCU,and explain the semantic of model,and appear the evolve rule.Compared with others, the rule and communication language make cooperation of MAS adaptting to open environment.(5) We present cooperative planning based on contract net protocol. It overcome that biders lack the task decompositing knowledge in the contract net protocol and judge that a plan can be worked using inherent knowledge in the cooperation planning system.(6) We have designed the distributed vehical expert system based on telemetring. Since there are many types in the telemeter data, and processing mathods are complex, we design the frame combined with rule,model,. For positioning accuracy and extending the estimate of fault result, we establish the fault graph diagnosis and predict model. using the MCU modelof we establish the expert nodes, use the communication language and the evolving rule and cooperation model to establish the distributed vehical expert system based on telemetring. Compared with other systems, it has strong abilityof positioning fault, robust characteristics, expandable characteristics, and adapts to opening environment.更多还原