导　　师： 朱梧槚

学科专业： 081101

授予学位： 博士

作　　者： ;

机构地区： 南京航空航天大学

摘　　要： Agent理论与技术是当今人工智能与软件工程中的热门研究课题。所谓agent就 是具有一定自主性、主动能力、应激能力、社会能力的软件或硬件实体。目前，如何 实现agent的自主性、主动性、应激性及多agent间的交互等尚存在着诸多理论问题 与实现技术上的困难。 本文旨在建立适用于不同类型、不同程度自主性agent的设计、构造及实现的智 能agent结构和面向智能agent程序设计语言。具体获得的研究成果包括如下： （1）从面向agent的程序设计语言实现的角度，提出了一种框架问题的解决方法， 给出了效应公理的规则表示形式，为智能agent结构的建立提供了必要的理论基础。 （2）结合BDI结构和情境演算的优点，提出了一个能刻画agent的多种特征， 尤其是自主性的智能agent结构IASC。此结构既能表示agent的信念、目标、意向等 心智成份，又能进行行动推理和规划，为解释agent的自主性提供了统一的平台。 （3）提出了用推理复杂性作为度量agent自主性的标准，给出了“简单自主 agent（simple agent）”、“慎思自主agent（deliberate agent）”以及“规划agent（planning agent）”等不同程度自主性的agent的结构，从而为实际应用中根据特定的应用背景 选择建构具有何种推理能力的agent提供了必要的理论基础。该工作的另一方面的重 要意义在于，目前，从事智能agent理论研究的人员往往关注于理论模型的建构，但 是这种理论模型的复杂性往往给实际的实现带来困难，甚至难于实现。我们认为，实 际构造agent时并不需要实现完全的理论模型，从而根据具体的问题如何对理论模型 进行简化归约是重要的，只有这样才能使agent的理论研究对工程实现具有指导意义， 但是这方面的研究未见报道，本文这部分的工作正是对此做了一些探讨。 （4）讨论了如何用IASC构造具有主动性、应激性、社会性、合作性的agent； 基于智能agent结构IASC，实现了一个基于联合意向的合作模型；说明了如何在结 构中刻画agent的实时处理及适应性等特征。 （5）提出了一种面向智能agent的程序设计语言—IAPL，即智能agent程序设计 语言。给出了IAPL语言的形式规范；设计了一组心智状态操作原语（信念、意向等 心智成份的增加、删除、修改、查询等原语）及agent之间的通信原语；给出了IAPL 语言的语义；提出了在线执行与离线规划相结合的IAPL程序执行方式。 （6）开发了IAPL语言原型支撑系统IASS，为进一步研制多agent系统集成开发 环境奠定了基础。 Agent theory and technology is one of active researching areas in artificial intelligent V and software engineering. The term agent is used to denote a hardware or software-based computer system that enjoys the following properties: autonomy, pro-activeness, reactivity and social ability. However, there exist a number of theoretical problems and technical difficulties in implementing of agent.? This dissertation aims to present an intelligent agent architecture and agent-oriented programming language /(AOPL/), respectively. They may be adopted to design and various types agents, and agents with different levels of autonomy. The main results may be summarized as follows: /(1/) For the sake of convenience in AOPL programming, a solution to the frame problem is proposed, and rule-based representation for effect axioms is introduced, which establish theoretical basis for constructing agent architecture and implementing AOPL. /(2/) Based on BDI structure and situation calculus, we present an intelligent agent architecture called IASC, which can depict various features, especially autonomy of agent. Since mental states such as belief, goal, intention and so on can be represented, and reasoning about action and planning can be executed in IASC, IASC may serve as a uniform platform for interpreting the autonomy of agent. /(3/) Using reasoning complexity, a criterion for measuring the level of autonomy is put forward. Various architectures for different types agents, such as simple agent, deliberate agent and planning agent, are given. These works supply a necessary theoretical basis for implementing autonomous agent. In addition, the research on agent theory at present focus on formal model that is a bit far from practice. Our framework can serve as a platform in which, given a special problem, programmer can reduce a theoretical model to practicable one. For all we know, relevant research has not been appeared in literatures. /(4/) In IASC framework, we explore how to construct agent that can reflect the following features: pro-activeness, reactivity, social ability and cooperation, and how to depict real-time process and adaptivity of agent. Furthermore, a cooperation model based on joint intentions is implemented. /(5/) An agent-oriented programming language IAPL and its operational semantics are introduced, and some primitives, including manipulation of mental states and communication between agents, are designed. On-line execution and off-line planning are integrated in running manner of IAPL. /(6/) A prototype support system of IAPL /(lASS/) is designed and implemented, which may be used as a framework to develop multi-agent system integrated development environment.

关 键 词： 理论 结构 面向 程序设计语言 心智状态 行动理论

分 类 号： [TP312]

领　　域： [自动化与计算机技术] [自动化与计算机技术]