基于WGAN-XGBoost的ADS-B异常数据检测模型

doi:10.16265/j.cnki.issn1003-3033.2025.08.1566

摘要/Abstract

摘要： 为保障航空运行安全、提升空域管理效率并增强系统对欺骗与干扰的防御能力,提出一种融合Wasserstein生成对抗网络(WGAN)和极限梯度提升(XGBoost)算法的异常数据检测模型。首先,通过WGAN学习预处理后ADS-B数据的内在分布特征,生成异常数据,将其与原始数据融合作为训练数据集;其次,基于XGBoost算法训练混合数据集,构建异常分类检测器;最后,通过试验与朴素贝叶斯、逻辑回归、感知机等基准模型开展性能对比。结果表明:与其他机器学习异常数据检测器相比,XGBoost 异常数据检测器在准确率、精确率、召回率、F₁ 分数等6项指标上均表现更优,其中准确率和精确率均超过0.999;模型检测 243 792 条数据的总耗时为 2.070 2 s,平均每条数据检测耗时 0.008 5 ms,在检测性能与时间成本间实现最优平衡,且经真实异常事件验证,具备良好的实用性与适用性。

关键词: Wasserstein生成对抗网络(WGAN), 极限梯度提升(XGBoost), 广播式自动相关监视(ADS-B), 异常数据检测, 朴素贝叶斯

Abstract:

To enhance aviation operation safety, improve airspace management efficiency, and enhance the defense capabilities of system against spoofing and interference, an anomaly data detection model was proposed based on WGAN-XGBoost. Firstly, WGAN was utilized to learn the intrinsic distribution of the preprocessed ADS-B data, generating abnormal data for augmenting and balancing the training dataset. Then, XGBoost algorithm was employed to train the mixed dataset, building the final abnormal classification detector. Finally, the performance comparisons were conducted through experiments with benchmark models such as Naive Bayes, Logistic Regression, and Perceptron. The results show that the performance of XGBoost is superior to that of all comparison models including accuracy, precision, recall, and F₁ score, with accuracy and precision both exceeding 0.999. The total detection time for 243 792 data points is 2.070 2 s, with an average detection time of 0.008 5 ms per data point. It achieves the optimal balance between detection performance and time cost and has been validated by real abnormal events, demonstrating good practicality and applicability.

Key words: Wasserstein generative adversarial network (WGAN), extreme gradient boosting (XGBoost), automatic dependent surveillance-broadcast (ADS-B), abnormal data detection, naive Bayesian

中图分类号:

X924.2安全监测技术与设备

李怀谦, 陈雨昊, 付宇翔, 沈嘉意. 基于WGAN-XGBoost的ADS-B异常数据检测模型[J]. 中国安全科学学报, 2025, 35(8): 188-195.

LI Huaiqian, CHEN Yuhao, FU Yuxiang, SHEN Jiayi. ADS-B abnormal data detection model based on WGAN-XGBoost[J]. China Safety Science Journal, 2025, 35(8): 188-195.

图/表 11

图1

图2

图3

图4

图5

图6

表1

图7

图8

表2

图9

参考文献 25

[1]	暴佳伟, 田小平, 刘宇娜, 等. ADS-B安全问题研究综述[J]. 现代防御技术, 2022, 50(5):28-35. doi: 10.3969/j.issn.1009-086x.2022.05.005
	BAO Jiawei, TIAN Xiaoping, LIU Yuna, et al. Summary of research on security issues of ADS-B protocol[J]. Modern Defence Technology, 2022, 50(5): 28-35.
[2]	COSTIN A, FRANCILLON A. Ghost in the air (traffic): on the security of ADS-B[C]. Proceeding in 21^st USENIX Security Symposium (USENIX Security 12), 2012: 1-16.
[3]	MUELLER R K. Quality of reported NACP in surveillance and broadcast services systems[C]. Proceeding in 28^th IEEE/AIAA Digital Avionics Systems Conference, 2009: DOI:10.1109/DASC.2009.5347431.
[4]	沈笑云, 唐鹏, 张思远, 等. ADS-B统计数据的位置导航不确定类别质量分析[J]. 航空学报, 2015, 36(9): 3128-3136. doi: 10.7527/S1000-6893.2015.0004
	SHEN Xiaoyun, TANG Peng, ZHANG Siyuan, et al. Quality analysis of navigation uncertain category for position based on ADS-B statistical data[J]. Acta Aeronautica et Astronautica Sinica, 2015, 36(9):3128-3136. doi: 10.7527/S1000-6893.2015.0004
[5]	LI Nisi, LIN Lin, LI Fan. ADS-B anomaly data detection using SVDD-based LSTM encoder-decoder algorithm[C]. Proceeding in 3^rd IEEE International Conference on Civil Aviation Safety and Information Technology,2021:1295-1300.
[6]	王振昊, 王布宏. 基于SVDD的ADS-B异常数据检测[J]. 河北大学学报:自然科学版, 2019, 39(3):323-329.
	WANG Zhenhao, WANG Buhong. ADS-B anomaly data detection based on SVDD[J]. Journal of Hebei University: Natural Science Edition, 2019, 39(3):323-329.
[7]	LUO Peng, WANG Buhong, LI Tengyao, et al. ADS-B anomaly data detection model based on VAE-SVDD[J]. Computers & Security, 2021, 104: DOI: 10.1016/j.cose.2021.102213.
[8]	丁建立, 邹云开, 王静, 等. 基于深度学习的ADS-B异常数据检测模型[J]. 航空学报, 2019, 40(12):167-177.
	DING Jianli, ZHOU Yunkai, WANG Jing, et al. ADS-B anomaly data detection model based on deep learning[J]. Acta Aeronautica et Astronautica Sinica, 2019, 40(12):167-177.
[9]	刘浪, 时宏伟. 基于注意力机制的CNN-LSTM的ADS-B异常数据检测[J]. 计算机系统应用, 2023, 32 (4): 94-103.
	LIU Lang, SHI Hongwei. ADS-B anomaly detection based on attention mechanism for CNN-LSTM[J]. Computer Systems & Applications, 2023, 32(4):94-103.
[10]	ÇEVIK N, AKLEYLEK S. SoK of machine learning and deep learning based anomaly detection methods for automatic dependent surveillance-broadcast[J]. IEEE Access, 2024, 12:643-662.
[11]	SABUHI M, ZHOU Ming, BEZEMER C P, et al. Applications of generative adversarial networks in anomaly detection: a systematic literature review[J]. IEEE Access, 2021, 9:3-29.
[12]	YUE Meng, ZHENG Han, CUI Haoran, et al. GAN-LSTM-based ADS-B attack detection in the context of air traffic control[J]. IEEE Internet of Things Journal, 2023, 10(14): DOI: 10.1109/jiot.2023.3252809.
[13]	郑圣彬, 谢加良, 张东晓. 基于LSTM-WGAN的时间序列数据异常检测[J]. 福建师范大学学报:自然科学版, 2024, 40 (2): 36-45.
	ZHENG Shengbin, XIE Jialiang, ZHANG Dongxiao. Time series data anomaly detection based on LSTM-WGAN[J]. Journal of Fujian Normal University: Natural Science Edition, 2024, 40(2): 36-45.
[14]	李佳晟. 基于多元特征的ADS-B异常检测关键技术研究[D]. 成都: 电子科技大学, 2023.
	LI Jiasheng. Research on key technologies of multivariate feature based ADS-B anomaly detection[D]. Chengdu: University of Electronic Science and Technology of China, 2023.
[15]	官成功, 高勇. 基于生成对抗网络的ADS-B信号降噪[J]. 通信技术, 2021, 54 (3): 568-574.
	GUAN Chenggong, GAO Yong. Generative adversarial nets for ADS-B signal denoising[J]. Communications Technology, 2021, 54(3):568-574.
[16]	KIM C, PARK S, HWANG H J. Local stability of Wasserstein GANs with abstract gradient penalty[J]. Neural Networks and Learning Systems, 2022, 33(9):4527-4537.
[17]	CHEN Shichuan, ZHENG Shilian, YANG Lifeng, et al. Deep learning for large-scale real-world ACARS and ADS-B radio signal classification[J]. IEEE Access, 2019, 7:256-264.
[18]	CHEN Tianqi, GUESTRIN C. XGBoost: a scalable tree boosting system[C]. Proceedings of the 22^nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 785-794.
[19]	MANESH M R, KAABOUCH N. Analysis of vulnerabilities, attacks, countermeasures and overall risk of the automatic dependent surveillance-broadcast (ADS-B) system[J]. International Journal of Critical Infrastructure Protection, 2017,19:16-31.
[20]	ALI B S, TAIB N A. A study on geometric and barometric altitude data in automatic dependent surveillance broadcast (ADS-B) Messages[J]. The Journal of Navigation, 2019, 72(5):40-58.
[21]	周志华. 机器学习[M]. 北京: 清华大学出版社, 2016:44-56.
[22]	SHERWIN J S, CHARTIER J. Parameter optimization of logistic regression classifiers[J]. BMC Neuroscience, 2013, 14(S1): DOI: 10.1186/1471-2202-14-S1-P62.
[23]	周长春, 陈勇刚. 基于人工神经网络的城市火灾事故的预测方法[J]. 中国安全科学学报, 2005, 15 (5):21-23.
	ZHOU Changchun, CHEN Yonggang. Prediction method of city fire accidents based on artificial neural network[J]. China Safety Science Journal, 2005, 15(5):21-23.
[24]	王旭. 人工神经元网络原理与应用[M]. 沈阳: 东北大学出版社, 2000:20-45.
[25]	徐翔, 陈龙, 李松阳, 等. 航空发动机燃烧室机匣烧穿火焰特性试验研究[J]. 清华大学学报:自然科学版, 2024, 64(6):984-991.
	XU Xiang, CHEN Long, LI Songyang, et al. Experimental study on flame characteristics of combustor casing burn-through in aero-engine[J]. Journal of Tsinghua University: Science and Technology, 2024, 64(6): 984-991.

评价指标	XGBoost	朴素贝叶斯	逻辑回归	感知机
准确率	0.999 4	0.947 2	0.824 8	0.814 2
精确率	1.000 0	0.824 5	0.451 5	0.214 2
召回率	0.996 2	0.868 9	0.229 1	0.042 4
F₁分数	0.998 1	0.846 1	0.303 9	0.070 8
ROC-AUC	1.000 0	0.986 4	0.865 9	0.781 1
PR-AUC	0.999 9	0.942 9	0.439 6	0.325 4

模型	总检测时间/s	平均每次检测时间/ms	异常检测次数
XGBoost	2.070 2	0.008 5	2437 92
朴素贝叶斯	0.157 7	0.000 6	2437 92
逻辑回归	0.757 7	0.003 1	2437 92
感知机	0.547 8	0.002 2	2437 92