北京航空航天大学学报 ›› 2017, Vol. 43 ›› Issue (7): 1450-1459.doi: 10.13700/j.bh.1001-5965.2016.0956

• 论文 • 上一篇    下一篇

基于深度置信网络的近距空战态势评估

张彬超, 寇雅楠, 邬蒙, 左家亮   

  1. 空军工程大学 航空航天工程学院, 西安 710038
  • 收稿日期:2016-12-20 修回日期:2017-02-15 出版日期:2017-07-20 发布日期:2017-07-28
  • 通讯作者: 寇雅楠,E-mail:49841256@qq.com E-mail:49841256@qq.com
  • 作者简介:张彬超 男,硕士研究生。主要研究方向:机器学习与智能空战;寇雅楠 女,博士,副教授,硕士生导师。主要研究方向:航空兵作战训练与效能评估。
  • 基金资助:
    航空科学基金(20155896026)

Close-range air combat situation assessment using deep belief network

ZHANG Binchao, KOU Yanan, WU Meng, ZUO Jialiang   

  1. College of Aeronautics and Astronautics Engineering, Air Force Engineering University, Xi'an 710038, China
  • Received:2016-12-20 Revised:2017-02-15 Online:2017-07-20 Published:2017-07-28
  • Supported by:
    Aeronautical Science Foundation of China (20155896026)

摘要: 针对传统态势评估方法权值确定困难、大规模数据处理和特征提取能力不足的问题,结合当前空战数据特征,将深度置信网络(DBN)应用于近距空战态势评估。通过密度峰值算法对空战特征数据进行聚类分析,并结合态势函数和专家判读进行修正,建立标准空战态势样本库;以重构误差和测试错误率为基础,建立网络拓扑结构和最优参数确定方法,提高模型的训练效率,并通过样本数据,对模型进行训练和验证。实验表明,模型态势分类正确率达到92.7%,模型运行时间满足应用需求,实例评估结果与客观态势一致性强。

关键词: 深度置信网络(DBN), 态势评估, 半监督学习, 网络拓扑结构, 密度峰值聚类

Abstract: Considering the difficulty in parameter setting, weakness of traditional situation assessment methods in processing and feature extraction of big data, feature of air combat data, applications of deep belief network (DBN) to close-range air combat situation assessment are discussed. A sample library of combat situation was constructed. The data were clustered using density peaks algorithm, and the results were revised by specialists of air combat and traditional functions. Then the model of deep belief network was constructed. According to the standard of test and reconstruction error, the network topology structure and optimal parameters were determined. The model was trained by the data from the sample library. Experimental results show that the model's situation classification accuracy reaches to 92.7%, and its running time meets the application requirements. Analysis of the practical example verified the feasibility of the DBN model.

Key words: deep belief network (DBN), situation assessment, semi-supervised learning, network topology structure, density peaks clustering

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