华南理工大学学报(自然科学版) ›› 2015, Vol. 43 ›› Issue (1): 34-40.doi: 10.3969/j.issn.1000-565X.2015.01.006

• 电子、通信与自动控制 • 上一篇    下一篇

参数在线学习的动态贝叶斯网络态势估计算法

袁德平1 郑娟毅2 史浩山1   

  1.  1. 西北工业大学 电子信息学院, 陕西 西安 710129 ; 2. 西安邮电大学 通信与信息工程学院, 陕西 西安 710121
  • 收稿日期:2014-08-13 修回日期:2014-10-17 出版日期:2015-01-25 发布日期:2014-12-01
  • 通信作者: 袁德平(1972-) ,男,博士生,高级工程师,主要从事信息融合、目标跟踪研究. E-mail:depingy@sina.com
  • 作者简介:袁德平(1972-) ,男,博士生,高级工程师,主要从事信息融合、目标跟踪研究.
  • 基金资助:

    国家自然科学基金资助项目( 61201194) ;陕西省科学技术研究发展计划资助项目( 2013K06-07)

Situation Assessment Algorithm for Online Parameters Learning in Dynamic Bayesian Networks

Yuan De - ping1 Zheng Juan - yi2 Shi Hao - shan1   

  1. 1. School ofElectronicsInformation , Northwestern Polytechnical University , Xi ’ an 710129 , Shaanxi , China ; 2. School ofTelecommu-nication and Information Engineering , Xi ’ an University of Posts and Telecommunications , Xi ’ an 710061 , Shaanxi , China 
  • Received:2014-08-13 Revised:2014-10-17 Online:2015-01-25 Published:2014-12-01
  • Contact: 袁德平(1972-) ,男,博士生,高级工程师,主要从事信息融合、目标跟踪研究. E-mail:depingy@sina.com
  • About author:袁德平(1972-) ,男,博士生,高级工程师,主要从事信息融合、目标跟踪研究.
  • Supported by:
    Supported by the National Natural Science Foundation of China ( 61201194 ) and the Science and Technology Research Development Project of Shaanxi Province ( 2013K06-07)

摘要: 为快速实现对战场态势的精确估计,提出了参数在线学习的动态贝叶斯网络方法:在基于专家知识确定的动态贝叶斯网络结构模型基础上,用前向递归方法对网络模型的参数进行估计 . 针对战场态势模型的观测值具有小样本的特性,以狄利克雷分布作为样本的先验分布,采用矩估计法对先验分布的超参数进行估计,以该先验分布的等价样本与观测值实现对网络参数的学习和对战场态势的估计 . 仿真实验结果表明,应用该方法实现态势估计具有较高的实时性和准确性 .

关键词: 贝叶斯网络, 态势估计, 狄利克雷分布, 参数学习

Abstract:

In order to assess battlefield situation accurately and quickly , an algorithm for online parameters learning is proposed on the basis of dynamitic Bayesian networks ( DBN ) . Forward recursion algorithm is used to estimate the parameters of network model after the structure model of dynamic Bayesian network is confirmed by expert knowledge. Dirichlet distribution is used as the prior distribution of samples according to the characteristics of small samples for the observation value of battlefield situation model , and moment estimation is adopt to estimate the hyper parameters of the prior distribution. Then , in combination with the equivalent samples value from the prior distribution , the observation value can be used to implement parameters learning and battlefield situation assessment. Simulated results indicate that the proposed algorithm is of good real-time performance and high accuracy for situation assessment.

Key words: Bayesian networks, situation assessment, Dirichlet distribution, parameters learning

中图分类号: