【作者】 黄玉奇；

【导师】 彭冬亮；

【作者基本信息】 杭州电子科技大学 ， 模式识別与智能系统， 2012， 硕士

【摘要】 信息融合是军事领域中研究的热点问题，信息融合技术包括多个级别：较低级别融合表现为对目标位置和属性的估计，高级别融合则表现为对战场进行态势估计与威胁估计。因此，态势估计是在一级融合获得的战场目标信息、环境信息等的基础之上，进行战场态势的评估，并确定其对敌、我方的有利程度。在态势估计的过程中，需要综合大量的、多种来源、多种格式以及不确定性的知识，并从这些知识中推理出对战场态势的预测。因此，态势知识的表示和不确定知识的推理是态势估计研究的重要问题。本文结合基于×××估计技术国防预研基金项目和×××目标×××系统项目，对这两个方面进行了研究，主要研究工作和成果如下：首先，从态势估计的定义、概念以及态势估计的功能模型、信息流程、知识的表示方式、知识的不确定推理模式等方面进行了详细的描述，为本文中态势估计知识表示和推理的研究提供了理论基础并明确了实际应用的方向。其次，在介绍贝叶斯网络构建、网络学习、网络推理的基础上，分析了贝叶斯网络在态势估计应用中的基本流程，基本思路和应用方向。并针对以往态势估计的过程中，贝叶斯网络结构和节点概率由专家知识来确定而导致的与实践的偏差，以及不能动态适应战场变化的问题，提出了自适应贝叶斯网络的方法。该方法通过对低层次融合数据周期性的参数学习，得到贝叶斯网络参数，以期提高态势估计的适应性和精确性，并更好的反映战场规律变化。最后给出了一个假想实例，证明了自适应贝叶斯网络在态势估计中的良好应用。然后，针对军事领域信息的多源化、格式的多样化导致的信息多样性，引入本体理论，构建态势估计本体用于态势估计的知识描述，实现知识的共享和重用，并方便计算机的自动处理。另外，针对本体在不确定性知识描述方面的不足，本文对态势估计本体进行了概率扩展。并引入了一种将态势估计本体转化为贝叶斯网络的方法，使态势估计本体能够转换为贝叶斯网络进行推理。最后，给出了一个软件框架用于二者的转换。实验结果表明该系统框架切实可行，能够为后续基于本体的态势估计研究提供一个有实际意义的应用平台。最后，本文给出了一个能够实现态势估计完整流程的软件系统。该系统主要有五个模块组成：场景仿真模块、信息融合模块、通信模块、贝叶斯网络推理模块、态势结果显示模块。该系统明晰了态势估计的主要流程，为后续工作提供了一个基础应用平台。更多还原

【Abstract】 Information fusion is a hot issue in the field of the military, the informationfusion technology includes multiple levels: low level fusion performs as estimatingthe position and attribute of the target, while high level fusion behaves to reckonongoing situation and threats of the battlefield. Therefore, the situation assessment isbased on the information of military targets and environment getting from first levelfusion, and confirms who is favored between enemy and us. During the process ofsituation assessment, many multiple-sourced, diversified format, and uncertainknowledge needs to be integrated. And then the prediction of situation can bereasoned from above. So, how to indicate and reason the uncertain knowledge aboutsituation is a significant problem to research. This paper is grounded on the×××assessment technology and the project×××battlefield information fusion system totake research on the two aspects, and main research work and achievements are asfollows:First, the paper describes situation assessment from definition, main concept,functional model, procedure of information, the way to express knowledge, method ofuncertainty reasoning and so on. All above provides a theory basis and actualapplication direction for the research for knowledge representation and reasoning inthe situation assessment.Secondly, on the basis of the theory of Bayesian network construction, learningand reasoning, the paper analyzes the basic procedure, main thoughts and applieddirection of Bayesian network in the application of situation assessment. According tothe process of situation assessment before, structure and probability of node inBayesian network are determined by experts’knowledge, which brings out certaindeviation between data and practice. As a result, the network is unsuited to dynamicbattlefield. So this paper puts forward the method of adaptive Bayesian network. Thismethod studies the training data from the low level fusion periodically, to get theBayesian network parameters reflecting the changes of the battlefield and to improvethe adaptability and accuracy of the situation assessment. At last the paper gives ahypothetical example, to proof that the adaptive Bayesian network is good at situationassessment.Thirdly, because of the mul-soruce and the variety format of the information inthe situation assessment, this paper introduces the theory of ontology. And construct situation assessment ontology for description of knowledge, to realize the knowledgesharing and reuse, and make the knowledge to be easy for automatic processing.However, as ontology has shortcomings of describing for uncertain knowledge, thispaper expands probability for the situation assessment ontology, and introduces asoftware framework which can convert situation assessment ontology into Bayesiannetwork. The experimental results show that the system framework is practical andfeasible, can provide an application framework for the subsequent research insituation assessment.Finally, this paper gives a software system which can realize the whole processof situation assessment. The system mainly includes: Scene emulation module,information fusion module, communication module, Bayesian network reasoningmodule, situation results displaying module. This system cleared the main process ofsituation assessment, and provides a basic application platform for the follow-upwork.更多还原