【作者】 郑文恩；

【导师】 孙尧；

【作者基本信息】 哈尔滨工程大学 ， 导航、制导与控制， 2004， 博士

【摘要】 未来海上战争的主要作战样式之一是“网络中心战(Net-centric Warfare, NCW)”，即使用计算机、高速数据链和网络软件等，将海军作战舰艇、作战飞机和岸基军事力量联结成一个高度集中的计算机／通讯网络。在该网络中，各作战单元将在高速和连续的基础上共享大量的关键信息，将会大大提高海军作战的反应速度、精度和有效性．网络中心战系统是一个分布式实时系统，具有动态性、异构性、适应性和协同性，对传统的分布实时计算技术提出了巨大的挑战．Agent(智能单元)具有自主性、适应性、社会性、智能性、移动性和反应性的特点，因此本文将分布实时多Agent技术作为实现网络中心战的关键技术。 本文以海军兵力网络中心战系统为应用背景，研究构造一个分布实时多Agent系统的理论与实践。目前Agent理论研究侧重Agent逻辑理论及其形式化表达，通用Agent技术较少考虑实时性，已有的系统在实时性与智能性的权衡方面考虑的较少，难以满足分布式实时智能系统的需要。本论文重点围绕着多Agent系统满足实时系统硬实时和软实时的要求进行了研究，论文的主要工作和成果有： 通过将Agent逻辑理论和实时智能系统相结合，提出了单一实时Agent的结构和控制模型。基于Agent的BDI逻辑理论，采用混合结构，提出了实时混合层次Agent结构。在Agent控制中，采用了任意时间算法生成缺省控制方案，较好地满足了多Agent系统的实时性；同时采用TAEMS表达实时任务结构模型，提出了实时Agent方案质量Quality的定义，制定了产生所有可选择方案的三种策略，即最高质量策略、最小执行时间策略和最高可能质量策略，基于任务质量和完成时间的折衷算法--tradeoff算法设计了Agent任务规划器，并给出了任务可调度性分析算法；给出了单个实时Agent的控制实现框架，在实现时采用计划缓存和学习进行优化，满足了实时Agent的控制运行的需要。 深入研究了分布实时Agent的通信理论。首先给出了Agent基于知识的通信语言KQML(Knowledge Query and Manipulation Language, KQML)的软件结构，研究了实时Agent通信的解决方法，在KQML消息的原语层增加 哈尔滨工程大学博士学位论文一.......叫甲甲，.，网.时间约束，提出了通信质量QoS(Quality of Serviee，QoS)的定义，给出了QoS的语义，提出了基于通信质量的KQML的实时扩展方法，解决了多Ageni系统的实时通信问题;另外研究了Ageni通信安全性的原理，对KQML原语参数进行了安全性扩展，定义了实现安全机制的KQML两个新原语。 研究了多Agent系统中，在实时约束条件下Ageni间的协调和协作方法，建立了多Agent实时协调模型。提出一种新的Agent思维状态模型BGC，改进了.Rao和Georgeff提出的BDI模型。采用BGC模型，提出了一种新的基于命令的MAS的组织结构，建立了指挥网模型，给出了满足多Agent间的实时协调和协作的新方法，提出了Agent组织的形成方法。运用对策论中多人合作对策的理论和方法，建立了非对抗的Agent社会动力学模型，描述了Agent组织形成的推动力以及Agent组织的理性。 研究了多Agent系统中的不确定推理方法。首先指出了多Agent系统中一种新的不确定性—未确知性，将未确知数学理论引入到不确定性处理中，并利用未确知有理数的加法，解决了多Agent决策融合问题，实现了智能的加法，通过与已有证据理论方法进行比较，证明了该方法的有效性和先进性;提出了建立处理多Agent系统中的随机性、模糊性、灰性和未确知性的通用模型的新思路。给出了未确知有理数的函数定义，根据该定义提出了基于连续认知结构的Agent的不确定推理方法，解决了由不确定证据推出具有高可靠性的结论的问题。 利用分布实时多Agent技术建立了网络中心战的原型系统和仿真环境，建立了通用实时Agent模板结构，提出了Agent描述语言，采用面向Agent的程序设计方法开发了该原型系统应用软件。 最后本文指出了进一步的研究工作，展望了实时多Agent系统研究未来的发展方向。更多还原

【Abstract】 The main campaign form of future sea war is net-centric warfare, that is navy warship, fighter plane and coastal military foreces join in a high-centralized computer and communication network, in which every unit shares large amount key information, this promotes greatly the response speed, accurateness and validity of navy battle. Net-centric warfare system is a distributed real-time system that has characteristics of independence, heterogeneousness, adaptability and synergetic, it is a great challenge to traditional distributed real-time computation. Agent (intelligent unit) has characteristics of independence, adaptation, sociality, intelligence, mobility and reactivity, so it is considered that distributed real-time multi-agent is the key technology of net-centric warfare system.This dissertation is a study of theories and technologies to build a distributed real-time multi-agent system in the background of navy force net-centric warfare system. At present, agent study is focused on logic of agent and its formalization, but current agent technology seldom cares about real-time, the ready system pays little attention to balance between real-time and intelligence and can not meet the demands of distributed multi-agent system, this dissertation surrounds the hard-time and soft-time requirement of multi-agent system, the main work and achivement as the followings:The structure and control model of a real-time agent is put forward through integration of agent logics theory and real-time intelligent techonology. A real-time mixed multi-layered agent structure based on agent’s BDI logics and mixed structure is given. Better work is done in meeting the requirement of multi-agent real-time system by introducing control plan using Anytime Arithmetic in agent control, also TAEMS is adopted to express the real-time task structure, and the quality of real-time agent plan is defined, and three kinds of policy of choosing plan is given, which is Highest Quality Policy, Minimum Execution Time Policy and Most Probably Quality Policy; agent task schedulerusing tradeoff arithmetic which is based on a tradeoff between task’s quality and its time is designed; implement frame of an single agent is given, the methods are optimized by using plan cache and online learning, this further meets the real-time control requirement of a agent.Communication theory of distributed real-time agent is lucubrated in this dissertation. It first brings about the software structure of agent communication based on KQML, and the means by which agent to achieve real time communication, namly to add time constraint at the meta-layer of communication language; gives definition of QoS (Quality of Service, QoS) in communication, and puts forward the real time extention methods of KQML based on QoS, then solves the problem of real-time communication in Multi-agent communication. Further, presents principle of safe agent communication, extends safety parameters in KQML, defines tow new sentences of KQML for safety communication.The cooperation and coordination method of Agents under real-time constriant are studied, a multi-agent coordination model is presented. A new agent mental status model BGC (Belief, Goal and Command, BGC) is put forward, this is a betterment of Rao and Georgef’s agent BDI (Belief, Desire and Intention, BDI) theory; brings about an new organization structure based on command using BGC model, builds an model of command net; gives a new coorperation and coordination technique satifying real-time multi-agent system, and development way of agent organization. Further builds a social dynamics of non-conflict multi-agent society using multi-person cooperative game theories and technologies, describes the original impetus of shaping agent organization.Uncertain reasoning methods in multi-agent system are studied in this paper. It introduces a new uncertainty in multi-agent system, namely unascertainty, and brings unascertained mathematics in the processing of uncertainty, further applies unascertained numbers to settle the problem o更多还原