【作者】 陈中祥；

【导师】 岳超源；

【作者基本信息】 华中科技大学 ， 系统工程， 2004， 博士

【摘要】 CGF 建模技术是近年来随着先进分布仿真和人工智能技术发展而兴起的一门新的仿真建模领域。它的突出特点是利用计算机来代替真实的作战训练人员和武器装备,再现其在真实系统中的角色和职能,从而能够极大地降低训练费用和组织联合训练的难度。CGF 建模包括环境建模、物理建模和行为建模。环境建模给 CGF 提供一个虚拟战场;物理建模为 CGF 提供逼真的外观,使其与真实兵力“形似”;而行为建模则是对 CGF 的内部思维活动建模,使其与真实兵力“神似”,是 CGF 建模技术的核心和难点。 建立 CGF 的行为模型不同于原有的物理建模,要面对大量不确定且难以形式化的因素,涉及行为的推理、决策和知识学习等方面的内容,实现难度很大。目前 CGF的行为模型集中在简单的反应式行为层次上,高级智能行为涉及不多,且多采用专家系统形式实现。同时 CGF 的行为不是由 CGF 自身控制,系统耦合紧。 基于以上的背景,本文结合一个具体的应用项目运用 BDI agent 技术从个体的层面上深入研究 CGF 主体的行为建模理论及技术。论文的主要思路可分为三个层次:一是将 CGF 主体的行为分为内在行为和外在行为,并认为 CGF 主体的外在行为输出是基于其内在行为的。二是将 CGF 主体的内在行为表示分为理性行为模型和行为差异模型,并认为 CGF 主体的内在行为是由 CGF 主体的理性行为模型决定,但受行为差异模型影响。三是将 CGF 主体的理性行为进一步分为感知行为、决策行为和学习行为。 论文的主要工作包括: 1.对计算机生成兵力和行为建模技术的基础理论进行了综述。 2.从总体上描述了 CGF 主体的行为产生机制和行为表示方法。通过信念、愿望和意图及其相互关系来分析说明 CGF 主体的内在行为的产生过程,形式化地阐述CGF 主体的外在行为输出的内在机理;提出一种表示外在行为的输出框架,并定义一组基于行为输出框架的行为运算,以此将一个意图分解为一系列元行为。 3.建立了 CGF 主体感知行为的描述性模型和可计算模型。这样既描述了感知行为的产生过程,同时保证建立在描述性模型之上的可计算模型能应用于自治的仿真环境中。 4.建立了 CGF 主体的决策行为模型,着重分析了决策的风险特性和时间特性。分析了加入感知行为后,对决策行为复杂性的影响;提出了一种不确定型决策方法 III<WP=4>用于处理不完全感知信息对决策行为的影响。5.在 CGF 主体的行为模型加入学习行为,分析了 CGF 主体学习行为框架,和在该框架下实现学习的两种方式:离线学习和在线学习。建立了各自的学习模型,并进行了形式化描述,给出了各自的学习算法。6.从个体的角度研究影响 CGF 主体理性行为的若干重要行为因素,并探讨其在作战仿真中的应用。根据行为科学和心理学的研究成果,分析了影响 CGF 主体行为的各种主要因素,探讨了应用上述定性分析的结果建立行为因素影响 CGF 主体行为输出的行为差异模型。7.研究了构造仿真系统中仿真成员的行为建模问题。以通信对抗仿真训练系统为背景,描述 CGF 主体的行为建模实现方法。本文认为,建立 CGF 主体的行为模型是构造 CGF 系统的核心和难点所在。Agent技术具有良好的自治性、自主性、反应性和社会性等特点,能够用来解决传统的人工智能所不能解决的环境适应性、信息不完全性等问题,比较适合建立 CGF 主体的行为模型。CGF 主体的内部行为是由信念、愿望、意图和知识等内在状态决定的,通过感知、决策、学习等内在行为产生的;CGF 主体的行为输出是由理性行为模型决定,但受各种内外部行为因素影响。随着对 CGF 主体行为建模的进一步深入研究,必将促进计算机生成兵力技术和仿真建模技术的进一步发展,并使计算机生成兵力成为军事仿真训练的有力工具。更多还原

【Abstract】 The technique of the CGF modeling is being a new field of modeling and simulationwith the development of Advanced Distributed Simulation and Artificial Intelligence. Itcan make computer behave like real combatant and military equipment, so computer canreplace them, and the outlay and difficulty of training are greatly reduced. The CGF modelconsists of environment model, physics model and behavioral model. A virtual realisticbattlefield is provided for the CGF agent by environment model. A realistic appearance isbuilt by physics model. But behavioral model refer to represent the mental thinkingactivity, is the core and nodus in constructing the CGF system. There are many uncertainty factors that are difficult to be formalized, when we modelthe CGF’s behavior, which is different from the physics model. Modeling the CGF’sbehavior is very difficult, because it involves in the reasoning, decision-making andlearning of behavior. At present, the behavioral model of the CGF is simple and reactive,lack of advanced intelligent. It is usually developed with Expert System. At the same time,the CGF’s behavior is not under the control of itself. The system is coupled tightly. Based on the analysis above, the paper investigates thoroughly technique andapplication of behavior modeling by the BDI agent from the level of entity, integratingwith an application. It can be divided into three levels. Firstly, the behavior of CGF agentis divided into mental behavior and outer behavior, and the outer behavior is based on themental behavior. Secondly, the mental behavior of the CGF agent is denoted by rationalbehavioral model and behavioral difference model. The mental behavior is determined byrational behavior model, and influenced by behavioral difference model. Thirdly, therational behavior of the CGF agent consists of perceiving, decision-making and learning. The main efforts are outlined as follows: 1. The theory foundations of CGF and behavior modeling are summarized. 2. The output mechanism and expression of CGF’s behavior are described as a whole.The output process of the CGF’s mental behavior is explained through the Belief, Desire,Intension and their relation. The inherent mechanism of the outer behavior is formalized.An behavioral framework to express the outer behavior is put forward. A set of behavioralcalculation are defined base on the framework, in order to decompose Intension into aseries of meta-behaviors. V<WP=6>3. The descriptive and prescriptive models of the perceiving behavior of the CGF arebuilt. The prescriptive model, which is based on the descriptive model, is ensured to beapplied into the autonomous simulated environment. 4. The decision of CGF is modeled. And the characteristics of risk and time ofdecision are analyzed. The impacts of introduction of the perceiving behavior on thecomplexity of decision behavior are analyzed. Subsequently, an uncertainty decisionmethod is given, which is used to deal with the impact of incomplete perceivinginformation on decision. 5. The learning behavior is added into the behavior model of the CGF, and theframework of learning of the CGF is brought forward. The alternative model of offlinelearning and online learning are built and formalized. And the alternative arithmetic isgiven. Several important behavioral factors, which influenced on the rational behavior of theCGF, are investigated from the point of view of the entity according the research ofbehavioral science and psychology. The issue of the application of the factors is discussedin the military simulation. Subsequently the behavioral difference model is given based onthe result of qualitative analysis above. 7. The behavioral modeling of simulation member in the constructed simulation isresearched. The realization of behavior modeling of the CGF is described. We conclude that modeling the behavior of the CGF is the core and nodus in theresea更多还原