基于动态贝叶斯网络的空天态势评估方法研究

项目名称： 基于动态贝叶斯网络的空天态势评估方法研究

项目编号： No.61472441

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 杨海燕

作者单位： 中国人民解放军空军工程大学

项目金额： 87万元

中文摘要： 空天态势评估是空天一体化作战指挥决策的基础，是提高作战能力的重要途径。本项目以空天战场为研究对象，将态势评估与作战任务、态势感知动态结合，基于Holon理论构建空天态势评估的体系结构。在体系结构中，针对评估系统复杂、评估信息量大、输入信息不完备、推理过程动态不确定的问题，以动态贝叶斯网络（DBN）为研究工具，构建多模式DBN的空天态势评估模型，围绕DBN模式识别、转移网络学习、多维度推理展开理论研究和算法开发。通过将主动学习与半监督学习相结合，提高分类的准确度，快速确定DBN的模式类别；引入传递变量类构建时变评分函数，将多种智能搜索算法与蒙特卡罗仿真相融合，提高不同模式DBN的学习效率与精度；结合随机抽样与模糊推理的思想，并引入群智能算法，提出层次化的DBN快速推理算法，解决不确定、多维度推理的困难。本项目的开展及其预期成果将为促进空天作战指挥自动化的发展提供了理论依据和技术支撑。

中文关键词： 空天态势评估；动态贝叶斯网络；模式分类；机器学习；不确定推理

英文摘要： Aerospace integrative warfare is the main battle mode in the future. Aerospace situation assessment provides a great support for command & decision in the combat, which is an important way to improve the combat capability. In this project, the aerospace battlefield is as the object of research.Based on Holon theory, the adaptive situation assessment architecture is constructed for aerospace integrative warfare by combining situation assessment with situation awareness and combat mission.In order to solve practical problem, including the complex system modeling, a huge amount of evaluation information,incomplete input information,dynamic and uncertain inference, aerospace situation assessment model is builded based on the dynamic bayesian network (DBN), which is named multi model DBN (MMDBN).Theoretical research and algorithm development are carried out for MMDBN around pattern recognition ,transfer network learning and multi dimension inference.Pattern categories of DBN are quickly determined through the classification and learning methods. In order to improve the classification accuracy, online multi classifier will be developed by combining active learning with semi supervised learning.The frame of online learning is constructed by defining time-varying score function of a kind of transmission variable. Many intelligent search algorithm and simulation are fused to improve the learning efficiency and accuracy of the optimal structure for MMDBN. Many kinds of inference mechanism are analyzed, including the random sampling, fuzzy inference, swarm intelligence optimization algorithm. Hierarchical fast inference algorithm of multi model DBN will be developed to solve the problem of uncertain inference and multi dimension inference.The development and the prospective achievements of this project will provide theoretical basis and technical support for command automation of aerospace integrative warfare.

英文关键词： Aerospace Situation Assessment;Dynamic Bayesian Network;Pattern Classification;Machine Learning;Uncertain Inference