北京航空航天大学学报 ›› 2021, Vol. 47 ›› Issue (5): 1068-1076.doi: 10.13700/j.bh.1001-5965.2020.0095

• 论文 • 上一篇    下一篇

基于时空图卷积网络的无人机网络入侵检测方法

陈卓, 吕娜, 陈坤, 张彦晖, 高维廷   

  1. 空军工程大学 信息与导航学院, 西安 710077
  • 收稿日期:2020-03-17 发布日期:2021-05-28
  • 通讯作者: 吕娜 E-mail:lvnn2007@163.com
  • 作者简介:陈卓,男,硕士研究生。主要研究方向:机器学习、通信网络;吕娜,女,博士,教授,博士生导师。主要研究方向:通信网络、航空数据链、机器学习。
  • 基金资助:
    国家自然科学基金(61703427,61701521)

UAV network intrusion detection method based on spatio-temporal graph convolutional network

CHEN Zhuo, LYU Na, CHEN Kun, ZHANG Yanhui, GAO Weiting   

  1. College of Information and Navigation, Air Force Engineering University, Xi'an 710077, China
  • Received:2020-03-17 Published:2021-05-28

摘要: 无人机网络相比地面网络具有节点快速移动、拓扑结构变换频繁和通信链路不可靠的特点,传统的针对地面网络的入侵检测方法难以适用。针对无人机网络的时空动态特性进行建模,提出了一种无人机网络的入侵检测方法——基于注意力机制的时空图卷积网络(ATGCN)。将图卷积网络和门控递归单元组合为时空图卷积网络,从复杂多变的数据中提取网络的时空演变特征,通过注意力机制提取和入侵检测最相关的特征,输入支持向量机进行分类预测。多个数据集的实验分析表明:所提方法能够适应无人机网络的动态性和不稳定性,相比传统检测方法准确率高且误报率低,具有良好的鲁棒性和适应性。

关键词: 无人机网络, 入侵检测, 图卷积网络, 门控递归单元, 注意力机制

Abstract: Compared with ground networks, UAV networks have the characteristics of fast moving nodes, frequent topology changes, and unreliable communication links. Traditional intrusion detection methods are difficult to apply. Aimed at the spatio-temporal dynamic characteristics of UAV networks, an intrusion detection method:Attention-based Spatio-Temporal Graph Convolutional Network (ATGCN) is proposed, which combines graph convolutional network and gated recursive unit into spatio-temporal graph convolutional network. The spatio-temporal graph convolutional network extracts the spatio-temporal evolution characteristics of the network from complex and changeable data, attention mechanism is used to extract the features most relevant to intrusion detection, and the support vector machine is used as the last layer of the model for classification to identify network attacks. The experimental analysis of multiple datasets shows that the proposed method can adapt to the dynamics and instability of UAV networks, has higher accuracy and lower false positive rate than traditional detection methods, and has good robustness and adaptability.

Key words: UAV network, intrusion detection, graph convolutional network, gated recursive unit, attention mechanism

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