基于生成对抗网络的航天异常事件检测方法

张克明; 蔡远文; 任元

doi:10.13700/j.bh.1001-5965.2018.0682

基于生成对抗网络的航天异常事件检测方法

doi: 10.13700/j.bh.1001-5965.2018.0682

1.
航天工程大学研究生院, 北京 101416
2.
航天工程大学宇航科学与技术系, 北京 101416

基金项目:

国家自然科学基金 51475472

国家自然科学基金 61803383

国家自然科学基金 51605489

详细信息

作者简介:
张克明  男, 博士研究生, 高级工程师。主要研究方向:异常事件检测、信号识别

蔡远文  男, 博士, 教授。主要研究方向:航天器测试与发射

任元  男, 博士, 副教授。主要研究方向:导航、制导与控制

通讯作者:
任元, E-mail: renyuan_823@aliyun.com

中图分类号: TP391.4
计量
- 文章访问数: 1062
- HTML全文浏览量: 69
- PDF下载量: 641
- 被引次数: 0
出版历程
- 收稿日期: 2018-11-22
- 录用日期: 2019-02-16
- 网络出版日期: 2019-07-20

Space anomaly events detection approach based on generative adversarial nets

1.
Graduate School, Space Engineering University, Beijing 101416, China
2.
Department of Aerospace Science and Technology, Space Engineering University, Beijing 101416, China

Funds:

National Natural Science Foundation of China 51475472

National Natural Science Foundation of China 61803383

National Natural Science Foundation of China 51605489

More Information

Corresponding author: RE NYuan, renyuan_823@aliyun.com

摘要

摘要:
航天环境复杂，技术难度大，风险高，安全可靠性要求苛刻。航天异常事件样本少，且难以获取，有针对性地开展异常事件检测（AED）很有必要。为预防航天事故，尽早发现可能导致故障的异常事件，深入研究了最新人工智能和生成对抗网络（GAN）技术，提出了一种基于生成对抗网络的航天异常事件检测方法。使用正生成对抗网络模拟生成正常事件样本，训练反生成对抗网络模拟生成异常事件样本，设计合理算法训练测试，计算输入事件与正生成对抗网络生成的模拟正常事件欧氏距离，以及输入事件与反生成对抗网络生成的模拟异常事件的欧氏距离差，实现对异常事件的精确检测。通过在美国国家标准与技术研究所数据库（MNIST）数据集全部使用正常数据训练，并对异常事件检测性能进行了试验验证，试验结果表明：在MNIST数据集下，精确率和召回率综合评价指标（F1）及精确率和召回率曲线下面积（PRC）等关键技术指标比变分自动编码器（VAE）方法相应指标性能至少分别提升了31%和11%。在真实环境下采集的模拟航天音频数据试验，异常事件检测性能良好，进一步证实了所提方法真实可用。
- 生成对抗网络(GAN) /
- 异常检测 /
- 学习算法 /
- 深度学习 /
- 航天应用
Abstract:
Anomaly events detection (AED) is quite important in space field for the complex space environment, difficult technology, high risk and strictly safe and reliable requirements. Since there are few space anomaly events samples and they are hard to obtain, it is necessary to carry out targeted AED. In order to prevent space accidents and find anomaly events that may lead to fault as soon as possible, a novel approach for space anomaly events detection based on generative adversarial nets (GAN) is proposed in this paper. Normal event samples are generated by normal GAN, anomaly event samples are generated by anomaly GAN. We proposed a reasonable algorithm to calculate the divergence of Euclidean distance between input events and simulated normal events generated by normal GAN, and Euclidean distance between input events and simulated abnormal events generated by anomaly GAN.As a result, abnormal events is detected accurately. The method is trained and tested using the Mixed National Institute of Standards and Technology (MNIST) database. The test results show that the key technical indexes, such as precision rate and recall rate of comprehensive evaluation index (F1) and precision recall curve (PRC), are at least 31% and 11% higher than the traditional variational autoencoder (VAE) method. In addition, we evaluated the method by collected data in real environment which simulated space audio data. The abnormal event detection performance is very good, which proved that the proposed method could detect anomaly event in real environments.
- generative adversarial nets (GAN) /
- anomaly detection /
- learning algorithms /
- deep learning /
- space application

HTML全文

图 1 生成对抗网络结构示意图

Figure 1. Schematic diagram of structure of generative adversarial nets

下载: 全尺寸图片幻灯片

图 2 航天异常事件检测结构组成

Figure 2. Structure constitution of space anomaly events detection

下载: 全尺寸图片幻灯片

表 1 基于MNIST数据集异常检测性能

Table 1. Anomaly detection performance on MNIST

异常数字	F1	精确率	虚警率	AUC
0	0.746	0.745	0.329	0.878
1	0.701	0.548	0.430	0.758
2	0.783	0.712	0.357	0.552
3	0.779	0.712	0.362	0.531
4	0.787	0.716	0.351	0.499
5	0.801	0.705	0.331	0.616
6	0.786	0.786	0.353	0.699
7	0.775	0.523	0.368	0.271
8	0.787	0.772	0.351	0.618
9	0.784	0.680	0.356	0.463

下载: 导出CSV

表 2 本文方法与VAE方法异常检测性能对比

Table 2. Anomaly detection performance comparison between proposed method and VAE method

异常数字	本文方法		VAE^[12]
异常数字	F1	PRC	F1	PRC
0	0.746	0.745	0.537	0.517
1	0.701	0.548	0.205	0.063
2	0.783	0.712	0.598	0.644
3	0.779	0.716	0.332	0.251
4	0.787	0.705	0.381	0.337
5	0.801	0.786	0.427	0.325
6	0.786	0.846	0.433	0.432
7	0.775	0.522	0.212	0.148
8	0.787	0.772	0.490	0.499
9	0.784	0.679	0.210	0.104

下载: 导出CSV

表 3 真实环境数据检测性能

Table 3. Data detection performance in real environment

迭代次数	F1	精确率	召回率	AUC
1	0.778	0.747	0.813	0.617
2	0.963	0.940	0.987	0.978
3	0.981	0.987	0.975	0.984

下载: 导出CSV

参考文献(27)

[1]	诸彤宇, 王奇, 高梦丹.离群点挖掘技术在交通事件检测中的应用[J].计算机科学与探索, 2014, 8(1):111-120. http://d.old.wanfangdata.com.cn/Periodical/jsjkxyts201401012 ZHU T Y, WANG Q, GAO M D.Research on traffic incident detection with outlier mining technology[J].Journal of Frontiers of Computer Science and Technology, 2014, 8(1):111-120(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/jsjkxyts201401012
[2]	高小霞, 霍纬纲, 冯兴杰.基于模糊关联分类器的民机超限事件诊断方法[J].北京航空航天大学学报, 2014, 40(10):1366-1371. https://bhxb.buaa.edu.cn/CN/abstract/abstract13046.shtml GAO X X, HUO W G, FENG X J.Civil aircraft's exceedance event diagnosis method based on fuzzy associative classifier[J].Journal of Beijing University of Aeronautics and Astronautics, 2014, 40(10):1366-1371(in Chinese). https://bhxb.buaa.edu.cn/CN/abstract/abstract13046.shtml
[3]	李康强.基于广义能量算子的复杂时变调制信号分析方法及其在机械故障诊断中的应用研究[D].北京: 北京科技大学, 2018: 39-141. LI K Q.Generalized energy operator based complicated time-verying modulation signal analysis method for machinery fault diagnosis[D].Beijing: University of Science and Technology Beijing, 2018: 39-141(in Chinese).
[4]	冯英, 武建文, 王承玉, 等.基于振动信号识别的断路器故障诊断研究[J].高压电器, 2017, 53(2):1-7. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=gydq201702001 FENG Y, WU J W, WANG C Y, et al.Research of fault diagnosis of circuit breaker based on vibratin signal recognition[J].High Voltage Apparatus, 2017, 53(2):1-7(in Chinese). http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=gydq201702001
[5]	李晓峰, 杨春山, 丁树春.基于信息熵的城市隧道实时交通事件检测聚类[J].计算机技术与发展, 2013, 23(10):212-215. http://d.old.wanfangdata.com.cn/Periodical/wjfz201310053 LI X F, YANG C S, DING S C.Entropy-based city tunnel real-time traffic incident detection clustering[J].Computer Technology and Development, 2013, 23(10):212-215(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/wjfz201310053
[6]	张先飞, 郭志刚, 刘嵩, 等.基于触发词指导的自相似度聚类事件检测[J].计算机科学, 2010, 37(3):212-220. doi: 10.3969/j.issn.1002-137X.2010.03.051 ZHANG X F, GUO Z G, LIU S, et al.Self-similarity clustering event detection based on triggers guidance[J].Computer Science, 2010, 37(3):212-220(in Chinese). doi: 10.3969/j.issn.1002-137X.2010.03.051
[7]	BAY S D, SCHWABACHER M.Mining distance-based outliers in near linear time with randomization and a simple pruning rule[C]//Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York: ACM, 2003: 29-38.
[8]	IVERSON D L.Inductive system health monitoring[C]//Proceedings of the International Conference on Artificial Intelligence, ICAI'04.Las Vegas: CSREA Press, 2004: 605-611.
[9]	BUDALAKOTI S, SRIVASTAVA A N, OTEY M E.Anomaly detection and diagnosis algorithms for discrete symbol sequences with applications to airline safety[J].IEEE Transactions on Systems, Man, and Cybernetics, Part C:Applications and Reviews, 2009, 39(1):101-113. doi: 10.1109/TSMCC.2008.2007248
[10]	DAS S, MATTHEWS B L, SRIVASTAVA A N, et al.Multiple kernel learning for heterogeneous anomaly detection: Algorithm and aviation safety case study[C]//Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York: ACM, 2010: 47-56.
[11]	SMART E, BROWN D, DENMAN J.Combining multiple classifiers to quantitatively rank the impact of abnormalities in flight data[J].Applied Soft Computing, 2012, 12(8):2583-2592. doi: 10.1016/j.asoc.2012.03.059
[12]	AN J, CHO S.Variational autoencoder based anomaly detection using reconstruction probability[J].Special Lecture on IE, 2015, 12:1-18. http://cn.bing.com/academic/profile?id=29a9d1208c0d4dff718113c0e9096102&encoded=0&v=paper_preview&mkt=zh-cn
[13]	LIM H, PARK J, LEE K, et al.Rare sound event detection using 1D convolutional recurrent neural networks[C]//Detection and Classification of Acoustic Scenes and Events Workshop 2017, 2017: 1-5.
[14]	胡绍林, 黄刘生.航天故障的成因分析与诊断技术[J].控制工程, 2003, 10(4):295-298. doi: 10.3969/j.issn.1671-7848.2003.04.003 HU S L, HUANG L S.Analysis and diagnosis of faults in spaceflight engineering[J].Control Engineering of China, 2003, 10(4):295-298(in Chinese). doi: 10.3969/j.issn.1671-7848.2003.04.003
[15]	谢敏, 楼鑫, 罗芊, 等.航天器故障诊断技术综述及发展趋势[J].软件, 2016, 37(7):70-74. doi: 10.3969/j.issn.1003-6970.2016.07.014 XIE M, LOU X, LUO Q, et al.Reviewed and developing trend of spacecraft fault diagnosis technology[J].Computer Engineering & Software, 2016, 37(7):70-74(in Chinese). doi: 10.3969/j.issn.1003-6970.2016.07.014
[16]	丁彩红, 黄文虎, 姜兴渭, 等.载人航天故障诊断技术的发展及其关键技术分析[J].强度与环境, 1999(2):20-24. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=QK199900521482 DING C H, HUANG W H, JIANG X W, et al.The development of spaceflight fault diagnostic techniques and the analysis towards its key skills[J].Structure & Environment Engineering, 1999(2):20-24(in Chinese). http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=QK199900521482
[17]	苏振华, 陆文高, 齐晶, 等.基于BP神经网络的卫星故障诊断方法[J].计算机测量与控制, 2015, 24(5):63-65. http://d.old.wanfangdata.com.cn/Periodical/jsjzdclykz201605019 SU Z H, LU W G, QI J, et al.A method of satellite fault diagnosis based on BP neural network[J].Computer Measurement & Control, 2015, 24(5):63-65(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/jsjzdclykz201605019
[18]	燕飞, 秦世引.基于RBF神经网络和M距离的卫星故障诊断[J].航天控制, 2006, 24(6):61-66. doi: 10.3969/j.issn.1006-3242.2006.06.014 YAN F, QIN S Y.Fault diagnosis for satellites based on RBF neural network and Mahalanobis distance[J].Aerospace Control, 2006, 24(6):61-66(in Chinese). doi: 10.3969/j.issn.1006-3242.2006.06.014
[19]	曾何俊.基于机器学习的卫星故障动态自适应建模关键技术研究[D].成都: 电子科技大学, 2018: 21-76. http://cdmd.cnki.com.cn/Article/CDMD-10614-1018991463.htm ZENG H J.Research on modeling key technology of machine learning methods for dynamical adaptation of satellite fault[D].Chengdu: University of Electronic Science and Technology of China, 2018: 21-76(in Chinese). http://cdmd.cnki.com.cn/Article/CDMD-10614-1018991463.htm
[20]	GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al.Generative adversarial nets[C]//International Conference on Neural Information Processing Systems, 2014: 2672-2680.
[21]	SCHLEGL T, SEEBOCK P, WALDSTEIN S M, et al.Unsupervised anomaly detection with generative adversarial networks to guide marker discovery[C]//Information Processing in Medical Imaging.Berlin: Springer, 2017: 146-157.
[22]	MICHELSANTI D, TAN Z H.Conditional generative adversarial networks for speech enhancement and noise-robust speaker verification[C]//Conference of the International Speech Communication Association 2017, 2017, 8: 2008-2012.
[23]	洪洋, 葛振华, 王纪凯, 等.深度卷积对抗生成网络综述[C]//第18届中国系统仿真技术及其应用学术年会, 2017, 5: 279-283. HONG Y, GE Z H, WANG J K, et al.An overview of deep convolution confrontation generation network[C]//18th Chinese Conference on System Simulation Technology & Application, 2017, 5: 279-283(in Chinese).
[24]	DUMOULIN V, BELGHAZI I, POOLE B, et al.Adversarially learned inference[C]//29th Conference on Neural Information Processing Systems(NIPS 2016), 2016, 6: 1-16.
[25]	ZENATI H, FOO C S, LECOUAT B, et al.Efficient gan-based anomaly detection[C]//International Conference on Learning Representations, 2018: 1-7.
[26]	YAMASHITA A, HARA T, KANEKO T.Inspection of visible and invisible features of objects with iImage and sound signal processing[C]//IEEE/RSJ International Conference on Intelligent Robots and Systems.Piscataway, NJ: IEEE Press, 2006: 3837-3842.
[27]	DONAHUE J, KRÄHENBVHL P, DARRELL T.Adversarial feature learning[C]//International Conference on Learning Representations, 2017, 4: 1-18.