基于改进VAE-WGAN的飞机不稳定进近检测

丁聪; 李啸宇; 王文涛; 张琦; 张小贝

doi:10.13700/j.bh.1001-5965.2024.0365

基于改进VAE-WGAN的飞机不稳定进近检测

doi: 10.13700/j.bh.1001-5965.2024.0365

丁聪^{1, 2},
李啸宇²,
王文涛²,
张琦^1, ,,
张小贝¹

1.
上海大学通信与信息工程学院，上海 200444
2.
中国商用飞机有限责任公司上海飞机设计研究院，上海 201210

基金项目:

民用飞机专项科研项目(MJZ2-6N21-2)

详细信息

通讯作者:
E-mail：qzhang9@shu.edu.cn

中图分类号: V328.2；TP183
计量
- 文章访问数: 293
- HTML全文浏览量: 125
- PDF下载量: 20
- 被引次数: 0
出版历程
- 收稿日期: 2024-05-29
- 录用日期: 2024-08-30
- 网络出版日期: 2024-10-11
- 整期出版日期: 2026-07-31

Unstable approach detection of aircraft based on modified VAE-WGAN

1.
School of Communication & Information Engineering，Shanghai University ，Shanghai 200444，China
2.
Shanghai Aircraft Design and Research Institute，Commercial Aircraft Corporation of China Ltd.，Shanghai 201210，China

Funds:

Special Scientific Research Project for Civil Aircraft (MJZ2-6N21-2)

More Information

Corresponding author: E-mail：qzhang9@shu.edu.cn

摘要

摘要:
随着航空业的快速发展，保证飞机的安全飞行显得尤为重要。为针对性开展航空领域异常事件检测，以飞机进近阶段的不稳定进近事件为目标，提出一种飞行级异常检测方法。该方法融合了变分自编码器(VAE)和Wasserstein生成对抗网络(WGAN)，并使用帕累托分布来模拟异常情况的概率分布，简称PVAE-WGAN。将VAE和WGAN的生成器共享，以正态分布和帕累托分布中随机采样的隐变量作为生成器输入，输出的重构样本分别作为正负样本。使用Wasserstein距离作为模型拟合正负样本的分布与真实分布之间的度量，使生成器和鉴别器都获得区分异常的能力，从而实现对不稳定进近事件的精确检测。以真实飞行数据记录器(FDR)记录的数据为例，对所提方法进行训练与测试，与其他适用于不稳定进近检测的多维时间序列异常检测方法相比有显著提升。所提方法的F₁值可达0.935，相对其他方法的F₁值平均提升12.95%。
- 不稳定进近 /
- 异常检测 /
- 变分自编码器 /
- 生成对抗网络 /
- 帕累托分布
Abstract:
With the rapid development of the aviation industry, the safe flight of aircraft has become particularly important. A flight-level anomaly identification method based on unstable approach events during aircraft approach is presented to identify anomalous events in the aviation field. The method, called PVAE-WGAN, combines variational auto-encoders (VAE) and Wasserstein generative adversarial networks (WGAN), and uses a Pareto distribution to simulate the probability distribution of anomalous cases. The generators of VAE and WGAN are shared, and the hidden variables randomly sampled from the normal distribution and Pareto distribution are taken as input to the generator. The reconstructed output samples are taken as positive and negative samples, respectively. The Wasserstein distance is used as a measure between the distribution of positive and negative samples fitted by the model and the true distribution, so that both the generator and the discriminator gain the ability to distinguish anomalies, thereby achieving accurate detection of unstable approach events. The method was trained and tested using the real flight data recorder (FDR) data as an example, and it was found to be much better than existing multi-dimensional time series anomaly identification techniques that are appropriate for unstable approach detection. The F₁ score of the proposed method in this paper can reach 0.935, which is an average increase of 12.95% compared with other methods.
- unstable approach /
- anomaly detection /
- variational auto-encoder /
- generative adversarial network /
- Pareto distribution

HTML全文

图 1 GAN结构

Figure 1. Structure of GAN

下载: 全尺寸图片幻灯片

图 2 稳定进近标准检查点

Figure 2. Stable approach standard checkpoint

下载: 全尺寸图片幻灯片

图 3 不稳定进近检测的整体架构

Figure 3. The overall structure of unstable approach detection

下载: 全尺寸图片幻灯片

图 4 PVAE-WGAN的网络架构

Figure 4. The network structure of PVAE-WGAN

下载: 全尺寸图片幻灯片

图 5 异常分数结果

Figure 5. Anomaly score result

下载: 全尺寸图片幻灯片

图 6 航班序号234的飞行参数

Figure 6. Flight parameter of flight ID 234

下载: 全尺寸图片幻灯片

图 7 航班序号153的飞行参数

Figure 7. Flight parameter of flight ID 153

下载: 全尺寸图片幻灯片

表 1 飞行变量的超限标准

Table 1. Exceed criteria for flight variable

惯性垂直速度/（m·s⁻¹）	校正空速/ (m·s⁻¹)	俯仰角/rad	横滚角/rad	航道偏差	下滑道偏差
> 7.62	56.59 ～ 77.17	−0.17 ～ 0.17	−0.26 ～ 0.26	> 3	> 3
注：本文定义航道偏差、下滑偏差超限判据为3倍标准差，表中简写为3。

下载: 导出CSV

表 2 不同模型的性能对比

Table 2. Performance comparison of different models

方法	精确率	召回率	F₁值
Isolation Forest	0.671	0.855	0.752
DBSCAN^[5]	0.765	0.935	0.846
LOF^[6]	0.810	0.855	0.832
WGAN-GP^[8]	0.830	0.786	0.807
CVAE^[9]	0.875	0.890	0.883
VAE-LSTM^[10]	0.792	0.941	0.860
PVAE-WGAN	0.962	0.909	0.935
注：加粗数字表示最优值。

下载: 导出CSV

表 3 消融实验

Table 3. Ablation experiments

方法	精确率	召回率	F₁值
VAE-WGAN	0.839	0.945	0.889
PVAE	0.885	0.836	0.859
PVAE-WGAN	0.962	0.909	0.935

下载: 导出CSV

参考文献(21)

[1]	International Civil Aviation Organization. Safety management manual(DOC9859)[R/OL]. Montreal: International Civil Aviation Organization, 2009[2024-04-16] https://www.afeonline.com/shop/icaodoc-9859.html.
[2]	Airplanes B C. Statistical summary of commercial jet airplane accidents[R]. Virginia: Boeing, 2023: 8-14.
[3]	Sembiring J, Liu C W, Koppitz P, et al. Energy management for unstable approach detection[C]//Proceedings of the IEEE International Conference on Aerospace Electronics and Remote Sensing Technology. Piscataway: IEEE Press, 2019: 1-6.
[4]	鲁志东, 张曙光, 戴闰志, 等. 大型民机进近着陆段异常能量风险判据[J]. 航空学报, 2021, 42(6): 91-104. Lu Z D, Zhang S G, Dai R Z, et al. Abnormal energy risk criteria of large civil airplanes in approach and landing[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(6): 91-104(in Chinese).
[5]	Sheridan K, Puranik T G, Mangortey E, et al. An application of DBSCAN clustering for flight anomaly detection during the approach phase[C]//Proceedings of the AIAA Scitech 2020 Forum. Reston: AIAA, 2020: 1851.
[6]	Jasra S K, Valentino G, Muscat A, et al. Hybrid machine learning-statistical method for anomaly detection in flight data[J]. Applied Sciences, 2022, 12(20): 10261.
[7]	陈农田, 满永政, 李俊辉. 基于QAR数据的民机高高原进近着陆风险评估方法[J]. 北京航空航天大学学报, 2024, 50(1): 77-85. Chen N T, Man Y Z, Li J H. Risk assessment method for civil aircraft approach and landing at high plateau based on QAR data[J]. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(1): 77-85(in Chinese).
[8]	张鹏, 田子都, 王浩. 基于改进生成对抗网络的飞参数据异常检测方法[J]. 浙江大学学报(工学版), 2022, 56(10): 1967-1976. Zhang P, Tian Z D, Wang H. Flight parameter data anomaly detection method based on improved generative adversarial network[J]. Journal of Zhejiang University (Engineering Science), 2022, 56(10): 1967-1976 (in Chinese).
[9]	Memarzadeh M, Matthews B, Avrekh I. Unsupervised anomaly detection in flight data using convolutional variational auto-encoder[J]. Aerospace, 2020, 7(8): 115.
[10]	Rong C T, Ouyang S X, Sun H B. Anomaly detection in QAR data using VAE-LSTM with multihead self-attention mechanism[J]. Mobile Information Systems, 2022, 2022(1): 8378187.
[11]	Goodfellow I J, Pouget-abadie J, Mirza M, et al. Generative adversarial nets[C]//Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 2. New York: ACM, 2014: 2672-2680.
[12]	Schlegl T, Seeböck P, Waldstein S M, et al. Unsupervised anomaly detection with generative adversarial networks to guide marker discovery[C]//Proceedings of the Information Processing in Medical Imaging. Berlin: Springer, 2017: 146-157.
[13]	Arjovsky M, Chintala S, Bottou L. Wasserstein generative adversarial networks[C]//Proceedings of the 34th International Conference on Machine Learning - Volume 70. New York: ACM, 2017: 214-223.
[14]	Gulrajani I, Ahmed F, Arjovsky M, et al. Improved training of Wasserstein GANs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 5769-5779.
[15]	Kingma D P, Welling M. Auto-encoding variational Bayes[EB/OL]. (2013-12-20)[2024-04-16]. https://arxiv.org/abs/1312.6114.
[16]	Arnold B C. Pareto distribution[EB/OL]. (2014-12-24)[2024-4-16]. https://doi.org/10.1002/9781118445112.stat01100.
[17]	卢晓光, 许忠睿, 张喆, 等. ECOD算法在飞机不稳定进近检测中的应用[J]. 安全与环境学报, 2024, 24(5): 1872-1878. Lu X G, Xu Z R, Zhang Z, et al. Application of ECOD algorithm for detection of aircraft unstable approach[J]. Journal of Safety and Environment, 2024, 24(5): 1872-1878(in Chinese).
[18]	Kinders G. Guidance for identifying unstable approach with flight data[EB/OL]. (2022-05-01)[2024-04-16]. https://www.easa.europa.eu/en/downloads/136957/en.
[19]	Garcia E J, Mulvihill M L, Kharab M S, et al. Capturing multivariate time series interactions to detect high-risk instability during approach[C]//Proceedings of the AIAA AVIATION 2023 Forum. Reston: AIAA, 2023: 3548.
[20]	彭宇, 何永福, 王少军, 等. 飞行数据异常检测技术综述[J]. 仪器仪表学报, 2019, 40(3): 1-13. Peng Y, He Y F, Wang S J, et al. Flight data anomaly detection: a survey[J]. Chinese Journal of Scientific Instrument, 2019, 40(3): 1-13(in Chinese).
[21]	NASA. DASHlink-sample flight data[EB/OL]. (2012-12-29)[2024-04-16]. https://c3.ndc.nasa.gov/dashlink/projects/85/.