小样本条件下列车通信网络攻击样本生成方法

岳川; 王立德; 闫海鹏

doi:10.3969/j.issn.0258-2724.20210557

小样本条件下列车通信网络攻击样本生成方法

doi: 10.3969/j.issn.0258-2724.20210557

北京交通大学电气工程学院，北京 100044

基金项目: 中国国家铁路集团有限公司科技研究开发计划（N2020J007）

详细信息

作者简介:
岳川（1993—），男，博士研究生，研究方向为列车牵引控制与网络，E-mail：16117394@bjtu.edu.cn

通讯作者:
王立德（1960—），男，教授，研究方向为列车牵引控制与网络，E-mail：ldwang@bjtu.edu.cn

中图分类号: U285.8
计量
- 文章访问数: 316
- HTML全文浏览量: 132
- PDF下载量: 51
- 被引次数: 0
出版历程
- 收稿日期: 2021-07-14
- 修回日期: 2021-12-31
- 网络出版日期: 2023-08-25
- 刊出日期: 2022-04-01

Attack-Sample Generation Method for Train Communication Network Under Few-Shot Condition

School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China

摘要

摘要:
基于深度学习的列车通信网络入侵检测需要充足的训练样本支撑，然而实际可获取的列车通信网络攻击样本非常少. 本文将生成对抗网络（GAN）应用于攻击样本生成任务中，对GAN的采样策略、约束条件与损失函数进行改进，设计了基于卷积神经网络的生成器与判别器，提出一种基于改进GAN的攻击样本生成方法. 基于此方法开展了样本生成实验与入侵检测实验. 结果表明，该方法能够生成有效的攻击样本，改进模型的训练效果，模型的F₁分数平均增加了4.23%.
- 列车通信网络 /
- 入侵检测 /
- 深度学习 /
- 样本生成 /
- 生成对抗网络
Abstract:
Deep learning-based intrusion detection for the train communication network requires sufficient training samples, but there are few available attack samples in the actual scenario. Generative adversarial network (GAN) thus operates to generate attack samples. Also, the sampling strategy, constraint condition and loss function of GAN are improved; and a generator based on convolutional neural network and a discriminator are designed. Then an improved GAN-based method is proposed for attack sample generation. Sample generation experiments and intrusion detection experiments are conducted to test the proposed method, indicating that it can generate effective attack samples. When applying these generated samples in the training process of the intrusion detection model, the average F₁ score increase by 4.23%, which means that the detection capability is effectively improved.
- train communication network /
- intrusion detection /
- deep learning /
- sample generation /
- generative adversarial network

HTML全文

图 1 GMCW-GAN的整体结构

Figure 1. Overall structure of GMCW-GAN

下载: 全尺寸图片幻灯片

图 2 生成器网络结构

Figure 2. Network structure of generator

下载: 全尺寸图片幻灯片

图 3 判别器网络结构

Figure 3. Network structure of discriminator

下载: 全尺寸图片幻灯片

图 4 ETCN半实物实验平台

Figure 4. Hardware-in-the-loop experimental platform of ETCN

下载: 全尺寸图片幻灯片

图 5 入侵检测模型的结构

Figure 5. Network structure of intrusion detection model

下载: 全尺寸图片幻灯片

图 6 真实样本与生成样本的三维空间表示

Figure 6. 3D representation of real and generated samples

下载: 全尺寸图片幻灯片

图 7 TADS、TADS-F、TADS-L数据集分别训练的DNN入侵检测模型的F₁分数

Figure 7. F₁ score of DNN intrusion detection models trained by TADS, TADS-F, and TADS-L

下载: 全尺寸图片幻灯片

图 8 不同生成方法对入侵检测F₁分数的影响对比

Figure 8. Comparison of data generation methods affecting F₁ score in intrusion detection

下载: 全尺寸图片幻灯片

图 9 2项类内改进方法对入侵检测的影响对比

Figure 9. Comparison of two in-class improvements affecting intrusion detection

下载: 全尺寸图片幻灯片

表 1 TADS的样本分布

Table 1. Sample distribution of TADS

攻击类别	样本数量/个	比例/%
IP 扫描	10648	44.89
端口扫描	9562
漏洞扫描	11257
篡改	9301
重放	8872
ARP 欺骗	9525
正常报文	72643	55.11

下载: 导出CSV

表 2 TADS-F的样本分布

Table 2. Sample distribution of TADS-F

攻击类别	样本数量/个	比例/%
IP 扫描	500	3.97
端口扫描	500
漏洞扫描	500
篡改	500
重放	500
ARP欺骗	500
正常报文	72643	96.03

下载: 导出CSV

表 3 TADS-L的样本分布

Table 3. Sample distribution of TADS-L

攻击类别	样本数量/个	比例/%
IP 扫描	10000	45.23
端口扫描	10000
漏洞扫描	10000
篡改	10000
重放	10000
ARP 欺骗	10000
正常报文	72643	54.77

下载: 导出CSV

表 4 不同生成方法对入侵检测模型虚警率的影响对比

Table 4. Comparison of data generation methods affecting F_PR of intrusion detection models %

方法	ROS	SMOTE	GAN	文献[8]	文献[9]	本文方法
虚警率	12.5	12.1	12.2	12.4	12.2	12.4

下载: 导出CSV

参考文献(16)

[1]	简捷. 基于以太网的列车通信网络多业务调度优化策略研究[D]. 北京: 北京交通大学, 2020.
[2]	FENG J, LU X, YANG W, et al. Survey of development and application of train communication network[C]//Proceedings of the 2015 International Conference on Electrical and Information Technologies for Rail Transportation. Berlin: Springer, 2016: 843-854.
[3]	KUROSE J F, ROSS K W. Computer networking: a top-down approach. 7th edition[M]. [S.l.]: Cenveo Publishing Services, 2017.
[4]	赵凯琳,靳小龙,王元卓. 小样本学习研究综述[J]. 软件学报,2021,32(2): 349-369. ZHAO Kailin, JIN Xiaolong, WANG Yuanzhuo. Survey on few-shot learning[J]. Journal of Software, 2021, 32(2): 349-369.
[5]	TAN X, SU S, HUANG Z, et al. Wireless sensor networks intrusion detection based on SMOTE and the random forest algorithm[J]. Sensors, 2019, 19(203): S19010203.1-S19010203.15. doi: 10.3390/s19010203
[6]	刘金平,周嘉铭,刘先锋,等. 基于聚类簇结构特性的自适应综合采样法在入侵检测中的应用[J]. 控制与决策,2021,36(8): 1920-1928. LIU Jinping, ZHOU Jiaming, LIU Xianfeng, et al. Toward intrusion detection via cluster structure-based adaptive synthetic sampling approach[J]. Control and Decision, 2021, 36(8): 1920-1928.
[7]	GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets[J]. Advances in Neural Information Processing Systems, 2014, 27: 2672-2680.
[8]	ANDRESINI G, APPICE A, DE ROSE L, et al. GAN augmentation to deal with imbalance in imaging-based intrusion detection[J]. Future Generation Computer Systems, 2021, 123: 108-127. doi: 10.1016/j.future.2021.04.017
[9]	HUANG S, LEI K. IGAN-IDS: an imbalanced generative adversarial network towards intrusion detection system in ad-hoc networks[J]. Ad Hoc Networks, 2020, 105: 102177.1-102177.11. doi: 10.1016/j.adhoc.2020.102177
[10]	REYNOLDS D A. Gaussian mixture models[J]. EncycLopedia of Biometrics, 2009, 741: 659-663.
[11]	ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein generative adversarial networks[C]// Proceedings of the 3th International Conference on Machine Learning. Sydney: [s.n.], 2017: 214-223.
[12]	VILLANI C. Optimal transport: old and new[M]. Berlin: Springer, 2009.
[13]	MAAS A L, HANNUN A Y, NG A Y. Rectifier nonlinearities improve neural network acoustic models[C]// Proceedings of the 30th International Conference on Machine Learning. Atlauta: ICML, 2013: 3-9.
[14]	YUE C, WANG L, WANG D, et al. An ensemble intrusion detection method for train ethernet consist network based on CNN and RNN[J]. IEEE Access, 2021, 9: 59527-59539. doi: 10.1109/ACCESS.2021.3073413
[15]	GIBERT D, MATEU C, PLANES J. The rise of machine learning for detection and classification of malware: research developments, trends and challenges[J]. Journal of Network and Computer Applications, 2020, 153: 102526.1-102526.22. doi: 10.1016/j.jnca.2019.102526
[16]	VAN DER MAATEN L, HINTON G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2009, 9(11): 2579-2605.