北京航空航天大学学报 ›› 2016, Vol. 42 ›› Issue (7): 1486-1493.doi: 10.13700/j.bh.1001-5965.2015.0445

• 论文 • 上一篇    下一篇

图模型与学习算法结合的贝叶斯网络自动建模

沈琳1, 于劲松1,2, 唐荻音1, 刘浩1,3   

  1. 1. 北京航空航天大学 自动化科学与电气工程学院, 北京 100083;
    2. 北京航空航天大学 先进航空发动机协同创新中心, 北京 10008;
    3. 解放军95809部队, 93分队, 沧州 061736
  • 收稿日期:2015-07-02 出版日期:2016-07-20 发布日期:2016-07-29
  • 通讯作者: 于劲松,Tel.: 010-82338693 E-mail: yujs@buaa.edu.cn E-mail:yujs@buaa.edu.cn
  • 作者简介:沈琳 女,硕士研究生。主要研究方向:故障预测与健康管理技术。Tel.: 010-82338693 E-mail: shenlin9177@163.com;于劲松 男,副教授,硕士生导师。主要研究方向:预测与健康管理技术、自动测试系统。Tel.: 010-82338693 E-mail: yujs@buaa.edu.cn

Automatic learning of Bayesian network structure using graph model and learning algorithm

SHEN Lin1, YU Jinsong1,2, TANG Diyin1, LIU Hao1,3   

  1. 1. School of Automation Science and Electrical Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100083, China;
    2. Collaborative Innovation Center of Advanced Aero-Engine, Beijing University of Aeronautics and Astronautics, Beijing 10008;
    3. Unit 93, Army 95809 of PLA, Cangzhou 061736, China
  • Received:2015-07-02 Online:2016-07-20 Published:2016-07-29

摘要: 针对纯数据驱动的贝叶斯网络结构学习算法的准确度和效率较低的问题,提出了一种融合多信号流图模型与K2学习算法的贝叶斯网络自动建模方法。该方法利用多信号流图模型能够描述信号之间传递与依赖关系的能力,结合K2学习算法在结构学习中的优势,实现了专家知识与数据驱动方法有效融合的贝叶斯网络结构自动学习算法。通过与常用网络结构学习算法的对比实验证明,该融合算法显著降低了结构学习对学习范围和训练数据规模的要求,具有更高的学习准确度和运算效率。采用真实系统实例阐述了该融合算法的应用过程,验证了算法的实用性。

关键词: 贝叶斯网络, 结构学习, 多信号流图, K2算法, 故障诊断

Abstract: In order to improve the accuracy and efficiency of the data-driven approaches in learning Bayesian network structure, expert knowledge is usually implemented in the learning algorithm. To deal with the lack of effective ways to combine the expert knowledge and the data-driven learning approaches in the existing methods, this paper proposes an automatic learning method for Bayesian network structure learning, which combines multi-signal flow graphs and learning algorithm K2. The method inserts expert knowledge into data-driven learning methods, using the information of relationships between signals from multi-signal flow graphs and the structure learning algorithm K2, to achieve automatic learning of Bayesian network structure. Numerical analysis, compared with other typical network structure learning algorithms, proves that the proposed method significantly lowers the structure learning requirements for learning scale and training data size and provides a higher learning accuracy and computation efficiency. The application of the proposed method is illustrated using a real engineering system and verified the practicability of the algorithm at the same time.

Key words: Bayesian networks, structure learning, multi-signal flow graphs, K2 algorithm, fault diagnosis

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发