基于增量式孤立森林的液体火箭发动机异常检测方法

张万旋; 薛薇; 张楠

doi:10.13224/j.cnki.jasp.20220119

基于增量式孤立森林的液体火箭发动机异常检测方法

doi: 10.13224/j.cnki.jasp.20220119

中国航天科技集团有限公司北京航天动力研究所，北京 100076

详细信息

作者简介:
张万旋（1994-），男，博士生，主要研究方向为液体火箭发动机系统设计与故障诊断。

中图分类号: V434
计量
- 文章访问数: 64
- HTML浏览量: 19
- PDF量: 35
- 被引次数: 0
出版历程
- 收稿日期: 2022-03-09

Anomaly detection method of liquid rocket engine based on incremental isolation forest

Beijing Aerospace Propulsion Institute，China Aerospace Science and Technology Corporation，Beijing 100076，China

摘要

摘要:
为解决液体火箭发动机故障标签缺失条件下流数据无监督检测问题，以及满足不同发动机台次和不同工况的自适应检测需求，基于增量学习思想，提出了基于增量式孤立森林的异常检测算法。设计了多工况流数据检测条件下的在线更新策略、异常分数表达式，并通过更新停止策略避免故障数据对模型的污染。利用多台次试车数据对该模型进行验证，并与传统方法进行比较，结果表明，该算法能够对样本异常程度进行量化评价，能够有效检测早期缓变故障，其F₁指标较原始孤立森林算法提高了43%，检测及时性优于红线算法和自适应阈值算法。
- 孤立森林 /
- 自适应检测 /
- 增量学习 /
- 异常检测 /
- 流数据
Abstract:
In order to solve the problem of streaming data unsupervised detection of liquid rocket engine with the absence of fault label，and to enable adaptive detection of various engines and multiple working conditions，an anomaly detection algorithm of liquid rocket engine based on incremental isolation forest was proposed based on incremental learning.The online updating strategy and anomaly score expression for streaming data under various working condition were designed.And the update stop strategy was applied to avoid the pollution of fault on the model.Using a number of test data for analysis and comparison with traditional methods，the result showed that the algorithm can evaluate the anomaly degree of sample quantitatively，and detect the degree of fault effectively.The algorithm had 43% improvement on F₁ score over original isolation forest algorithm.And its detection timeliness was better than the red line algorithm and adaptive threshold algorithm.
- isolation forest /
- adaptive detection /
- incremental learning /
- anomaly detection /
- streaming data
注释:

1) 陈越

HTML

图 1 孤立树示意图

Figure 1. Sketch of isolation tree

下载: 全尺寸图片幻灯片

图 2 试验1归一化推力室压力和异常评分曲线

Figure 2. Normalized pressure of chamber and anomaly score curves of test 1

下载: 全尺寸图片幻灯片

图 3 试验2归一化推力室压力和异常评分曲线

Figure 3. Normalized pressure of chamber and anomaly score curves of test 2

下载: 全尺寸图片幻灯片

图 4 试验3归一化推力室压力和异常评分曲线

Figure 4. Normalized pressure of chamber and anomaly score curves of test 3

下载: 全尺寸图片幻灯片

图 5 试验4归一化推力室压力和异常评分曲线

Figure 5. Normalized pressure of chamber and anomaly score curves of test 4

下载: 全尺寸图片幻灯片

图 6 试验1原始孤立森林算法异常评分曲线

Figure 6. Anomaly score curve of test 1 with original isolation forest

下载: 全尺寸图片幻灯片

图 7 试验4原始孤立森林算法异常评分曲线

Figure 7. Anomaly score curve of test 4 with original isolation forest

下载: 全尺寸图片幻灯片

表 1 试车数据集

Table 1. Test data set

试验	现象	故障定位
1	93~96 s发生故障后恢复	涡轮泵密封环故障
2	突变型破坏	涡轮泵烧蚀故障
3	突变型破坏	涡轮盘故障
4	275 s后参数缓慢下降	涡轮泵叶片故障

下载: 导出CSV

表 2 检测参数

Table 2. Detection parameters

参数	符号	参数	符号
氢泵后压力	$p_{e f}$	氧泵后压力	$p_{e o}$
燃气发生器氢喷前压力	$p_{g i f}$	燃气发生器氧喷前压力	$p_{g i o}$
推力室氢喷前压力	$p_{c i f}$	推力室氧喷前压力	$p_{c i o}$
氢泵转速	$n_{f}$	氧泵转速	$n_{o}$
氢副气蚀管入口压力	$p_{w i f}$	氧副气蚀管入口压力	$p_{w i o}$
氢涡轮入口压力	$p_{i t f}$	氧涡轮入口压力	$p_{i t o}$
推力室压力	$p_{c}$	燃气发生器压力	$p_{g}$

下载: 导出CSV

表 3 检测结果评价（F₁指标）

Table 3. Detection result evaluation （F₁ score）

试验	增量式孤立森林算法	原始孤立森林算法
1	0.847	0.428
2	0.961	0.884
3	0.917	0.819
4	0.496	0.282
综合	0.812	0.564

下载: 导出CSV

表 4 检测时间对比

Table 4. Comparison of detection result

试验	增量式孤立森林算法	自适应阈值算法	红线算法
1	92.92	93.31	/
2	188.91	188.92	191.15
3	302.13	302.15	302.19
4	285.26	290.74	290.83
注：“/”表示无法有效检测。

下载: 导出CSV

参考文献(25)

[1]	李新鹏,高欣,阎博,等.基于孤立森林算法的电力调度流数据异常检测方法[J].电网技术,2019,43（4）:1447⁃1456. LI Xinpeng,GAO Xin,YAN Bo,et al.An approach of data anomaly detection in power dispatching streaming data based on isolation forest algorithm[J].Power System Technology,2019,43（4）:1447⁃1456.（in Chinese）
[2]	刘鑫.无监督异常检测方法研究及其应用[D].西安:电子科技大学,2018. LIU Xin.Research on unsupervised anomaly detection algorithm and application[D].Xi'an:University of Electronic Science and Technology of China,2018.（in Chinese）
[3]	王思齐.智能异常检测及其应用[D].长沙:国防科技大学,2019. WANG Siqi.Intellegent anomaly detection and its applications[D].Changsha:National University of Defense Technology,2019.（in Chinese）
[4]	RAMASWAMY S,RASTOGI R,SHIM K.Efficient algorithms for mining outliers from large data[J].ACM Sigmod Record,2000,29（2）:427⁃438.
[5]	ESTER M,KRIEGEL J,SANDER J,et al.A density⁃based algorithm for discovering clusters in large spatial databases with noise[C]∥Proceedings of 2nd International Conference on Knowledge Discovery and Data Mining.Menlo Park,US:AAAI Press,1996:226⁃231.
[6]	SCHÖLKOPF B,WILLIANMSON R,SMOLA A,et al.Support vector method for novelty detection[C]∥Advances in Neural Information Processing Systems 12:Proceedings of the 1999 Conference.Cambrtge,US:MIT Press,2000:582⁃588.
[7]	KRIEGEL H P,SCHUBERT M,ZIMEK A.Angle⁃based outlier detection in high⁃dimensional data[C]∥Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.Las Vegas,US:ACM,2008:444⁃452.
[8]	BREUNIG M M,KRIEGEL H P,SANDER N J,et al.LOF:identifying density⁃based local outliers[C]∥Proceedings of the 2000 ACM SIGMOD international conference on Management of data.New York:ACM,2000:1⁃12.
[9]	HAWMAN M W.Health monitoring system for the SSME⁃program overview[R].AIAA 90⁃1987,1990.
[10]	IVERSON D L.Inductive system health monitoring[C]∥Proceedings of the International Conference on Artificial Intelligence.Las Vegas,US:ICAI,2004：605⁃611.
[11]	胡雷.面向飞行器健康管理的新异常检测方法研究[D].长沙:国防科技大学,2010. HU Lei.Research on novelty detection methods oriented to flight vehicle health management[D].Changsha:National University of Defense Technology,2010.（in Chinese）
[12]	SCHWABACHER M,NIKUNJ C O,MATTHEWS B,et al.Unsupervised anomaly detection for liquid⁃fueled rocket propulsion health monitoring[J].Journal of Aerospace Computing Information and Communication,2009,6:464⁃482.
[13]	PEVNY T.Loda:lightweight on⁃line detector of anomalies[J].Machine Learning,2016,102:275⁃304.
[14]	LIU F T,TING K M,ZHOU Z H.Isolation forest[C]∥2008 Eighth IEEE International Conference on Data Mining.Pisa,Italy:IEEE Press,2008:413⁃422.
[15]	王诚,狄萱.孤立森林算法研究及并行化实现[J].计算机技术与发展,2021,31（6）:13⁃18. WANG Cheng,DI Xuan.Research and parallelization of isolation forest algorithm[J].Computer Technology and Development,2021,31（6）:13⁃18.（in Chinese）
[16]	朱恒伟,王克昌,陈启智.液体火箭发动机地面试车故障检测的自适应阈值算法[J].推进技术,2000,21（1）:1⁃4. ZHU Hengwei,WANG Kechang,CHEN Qizhi.Adaptive thresholds algorithm for fault detection of liquid rocket engine in ground test[J].Journal of Propulsion Technology,2000,21（1）:1⁃4.（in Chinese）
[17]	LI Z,ZHAO Y,BOTTANICOLA,et al.COPOD:copula⁃based outlier detection[C]∥2020 IEEE International Conference on Data Mining.Sorrento,Italy:ICDM,2020:1118⁃1123.
[18]	丁智国.流数据在线异常检测方法研究[D].上海:上海大学,2015. DING Zhiguo.Research on online anomaly detection for streaming data[D].Shanghai:Shanghai University,2015.（in Chinese）
[19]	琚安康.基于多源异构数据的定向网络攻击检测关键技术研究[D].郑州:战略支援部队信息工程大学,2020. JU Ankang.Research on key technologies of targeted cyber attacks detection based on multi⁃source heterogeneous data[D].Zhengzhou:Information Engineering University,2020.（in Chinese）
[20]	贾庆轩.基于机器学习的电力调度自动化系统业务异常检测方法研究[D].北京:北京邮电大学,2020. JIA Qingxuan.Research on business anomaly detection method of power dispatching automation system based on machine learning[D].Beijing:Beijing University of Posts and Telecommunications,2020.（in Chinese）
[21]	GUHA S,MISHRA N,ROY G,et al.Robust random cut forest based anomaly detection on streams[C]∥Proceedings of the 33rd International Conference on Machine Learning.New York:Proceedings of Machine Learning Research,2016:2712⁃2721.
[22]	李国成,陆俊,王赟,等.基于Bagging二次加权集成的孤立森林窃电检测算法[J].电力系统自动化,2022,46（2）:92‑100. LI Guocheng,LU Jun,WANG Yun,et al.Isolated⁃forest electricity theft detection algorithm based on bagging secondary weighted ensemble[J].Automation of Electric Power Systems,2022,46（2）:92⁃100.（in Chinese）
[23]	谢炎昆.基于孤立森林算法的用水异常监测研究[D].北京:北京邮电大学,2018. XIE Yankun.Research on water use anomaly monitoring based on isolated forest algorithm[D].Beijing:Beijing University of Posts and Telecommunications,2018.（in Chinese）
[24]	陈春旭.基子机器学习的电力调度自动化系统业务异常检测方法研究[D].北京:北京邮电大学,2020. CHEN Chunxu.Research on business anomaly detection method of power dispatching automation systems based on machine learning[D].Beijing:Beijing University of Posts and Telecommunications,2018.（in Chinese）
[25]	周志华.机器学习[M].北京:清华大学出版社,2016:403‑408.