基于LSTM-GBSVDD模型的飞行轨迹异常检测方法

doi:10.12382/bgxb.2024.0489

摘要/Abstract

摘要：

为解决传统检测方法在处理复杂、动态以及数据长度实时变化的飞行轨迹数据时特征提取不准确、检测效率较低的问题,提出一种结合长短时记忆(Long Short-Term Memory,LSTM)网络和支持向量数据描述(Support Vector Data Description,SVDD)的无监督异常检测方法。利用LSTM网络提取可变长度飞行轨迹的关键特征,并将其转化为固定长度的序列表示;通过SVDD算法构建多维超球分类器,对正常飞行轨迹进行建模,从而识别潜在异常轨迹。为进一步提升模型性能,引入基于梯度的优化算法(Gradient-Based training algorithm,GB),实现LSTM与SVDD参数的联合训练,大幅度提高检测精度和计算效率。仿真实验结果表明,新提出的基于梯度优化的长短时记忆网络和支持向量数据描述模型(Long Short-Term Memory network and Support Vector Data Description model based on Gradient-Based training algorithm optimization, LSTM-GBSVDD)的飞行轨迹异常检测方法在处理复杂、多变的飞行轨迹异常检测任务中表现出较好的有效性和优越性,有较强的应用前景。

关键词: 飞行轨迹, 长短时记忆, 支持向量数据描述, 异常检测

Abstract:

The traditional detection methods have the disadvantages of inaccurate feature extraction and low detection efficiency when processing the complex and dynamic flight trajectory data with real-time change in data length. An proposed flight trajectory anomaly detection method using the gradient-based optimization of long short-term memory network and support vector data description model based on gradient training algorithm optimization (LSTM-GBSVDD)is proposed. The LSTM network is used to extract the key features of variable-length flight trajectories and convert them into a fixed-length sequence representation. A multidimensional hypersphere classifier is constructed using the SVDD algorithm, which is used to model the normal flight trajectories and identify the potentially abnormal flight trajectories. To further improve model performance, a gradient-based training algorithm (GB) is introduced to jointly train the parameters of LSTM and SVDD, which greatly improves the detection accuracy and computational efficiency. The simulated results show that the proposed flight trajectory anomaly detection method using the gradient-based optimization of long short-term memory network and support vector data description model based on gradient training algorithm optimization (LSTM-GBSVDD) has good effectiveness and superiority in dealing with complex and changeable flight trajectory anomaly detection tasks, and has good application prospects.

Key words: flight trajectory, long short-term memory, support vector data description, anomaly detection

中图分类号:

E926

李琳, 曾雅琴, 朱惠民, 孙世岩, 梁伟阁. 基于LSTM-GBSVDD模型的飞行轨迹异常检测方法[J]. 兵工学报, 2025, 46(5): 240489-.

LI Lin, ZENG Yaqin, ZHU Huimin, SUN Shiyan, LIANG Weige. Flight Trajectory Anomaly Detection Method Based on LSTM-GBSVDD Model[J]. Acta Armamentarii, 2025, 46(5): 240489-.

图/表 22

图1 LSTM-GBSVDD模型的飞行轨迹异常检测方法流程图

Fig.1 Flowchart of flight trajectory anomaly detection method based on LSTM-GBSVDD model

图2 故障模式1飞行器速度变化图

Fig.2 Failt Ⅰ: aircraft speed variation graph

图3 故障模式2飞行器速度变化图

Fig.3 Fault Ⅱ: aircraft velocity Change Graph

图4 故障模式1飞行器高度变化图

Fig.4 Fault Ⅰ: aircraft altitude change graph

图5 故障模式2飞行器高度变化图

Fig.5 Fault Ⅱ: aircraft altitude change graph

图6 故障模式1飞行器第10s故障后飞行过程攻角变化图

Fig.6 Plot of angle of attack change during flight after 10th second failure of Fault Ⅰ aircraft

图7 故障模式2飞行器第10s故障后飞行过程攻角变化图

Fig.7 Plot of angle of attack change during flight after 10th second failure of Fault Ⅱ aircraft

图8 故障模式1飞行器侧滑角变化图

Fig.8 Fault Ⅰ: aircraft side slip angle variation graph

图9 故障模式2飞行器侧滑角变化图

Fig.9 Fault Ⅱ: aircraft side slip angle variation graph

图10 LSTM-GBSVDD方法的速度异常点检测 (飞行器故障1)

Fig.10 Speed anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅰ)

图11 LSTM-GBSVDD方法的速度异常点检测 (飞行器故障2)

Fig.11 Speed anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅱ)

图12 LSTM-GBSVDD方法的高度异常点检测 (飞行器故障1)

Fig.12 High anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅰ)

图13 LSTM-GBSVDD方法的高度异常点检测 (飞行器故障2)

Fig.13 High anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅱ)

图14 LSTM-GBSVDD方法的攻角异常点检测 (飞行器故障1)

Fig.14 Angle of attack anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅰ)

图15 LSTM-GBSVDD方法的攻角异常点检测 (飞行器故障2)

Fig.15 Angle of attack anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅱ)

图16 LSTM-GBSVDD方法的侧滑角异常点检测 (飞行器故障1)

Fig.16 Side slip angle anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅰ)

图17 LSTM-GBSVDD方法的侧滑角异常点检测 (飞行器故障2)

Fig.17 Sideslip angle anomaly detection based on LSTM-GBSVDD model (aircraft Fault Ⅱ)

表1 两种故障模式下不同方法异常检测200次平均值仿真实验的精确度(Prec)、F1、AUC和训练时间

Table 1 Precision, F1, AUC score and Train Time for 200 mean simulation trials of anomaly detection of different methods in two failure modes

参数	算法	故障模式1			故障模式2
参数	算法	SVDD	LSTM-GBSVM	LSTM-GBSVDD	SVDD	LSTM-GBSVM	LSTM-GBSVDD
	Prec	0.651±0.027	0.972±0.008	0.987±0.003	0.672±0.041	0.954±0.013	0.963±0.002
速度	F1	0.534±0.031	0.901±0.013	0.913±0.001	0.577±0.032	0.906±0.010	0.925±0.001
	AUC	0.781±0.024	0.955±0.006	0.996±0.002	0.792±0.036	0.982±0.014	0.998±0.003
	训练时间/s	0.187±0.040	9.594±0.079	1.085±0.005	0.198±0.032	9.438±0.197	1.065±0.004
	Prec	0.357±0.019	0.967±0.012	0.964±0.003	0.451±0.022	0.963±0.015	0.972±0.004
高度	F1	0.467±0.026	0.924±0.006	0.927±0.001	0.437±0.046	0.908±0.017	0.923±0.001
	AUC	0.734±0.018	0.943±0.011	0.957±0.004	0.832±0.054	0.936±0.022	0.942±0.004
	训练时间/s	0.193±0.045	9.984±0.186	1.043±0.008	0.174±0.35	9.805±0.193	1.135±0.003
	Prec	0.579±0.032	0.962±0.005	0.973±0.002	0.584±0.027	0.959±0.016	0.968±0.002
攻角	F1	0.771±0.034	0.968±0.009	0.989±0.005	0.649±0.035	0.903±0.010	0.917±0.003
	AUC	0.863±0.022	0.987±0.006	0.994±0.003	0.778±0.059	0.945±0.027	0.957±0.008
	训练时间/s	0.191±0.032	9.768±0.194	1.086±0.005	0.198±0.031	9.742±0.198	1.065±0.003
	Prec	0.632±0.029	0.956±0.008	0.992±0.004	0.679±0.035	0.954±0.043	0.968±0.001
侧滑角	F1	0.576±0.017	0.919±0.007	0.925±0.002	0.543±0.026	0.960±0.035	0.973±0.004
	AUC	0.789±0.036	0.937±0.013	0.966±0.002	0.738±0.052	0.975±0.039	0.989±1.063
	训练时间/s	0.198±0.031	9.438±0.165	1.097±0.005	0.194±0.030	9.672±0.191	1.063±0.006

图18 基于LSTM-GBSVDD算法的飞行轨迹数据集ROC曲线对比(飞行器故障1和故障2)

Fig.18 Comparison of ROC curves of flight trajectory dataset based on LSTM-GBSVDD model (aircraft Failts Ⅰ and Ⅱ)

图19 故障模式1下F1分数和AUC分数方法检测异常的精确度对比

Fig.19 Comparison of the accuracies of F1 and AUC scoring methods for detecting anomalies in Failt Ⅰ

图20 故障模式2下F1分数和AUC分数方法检测异常的精确度对比

Fig.20 Comparison of the accuracies of F1 and AUC scoring methods for detecting anomalies in Failt Ⅱ

图21 LSTM-GBSVDD异常检测方法训练时间对比

Fig.21 Comparison of training times of anomaly detection algorithms

参考文献 22

[1]	SLIMENE M B, OUALI M S. Anomaly detection method of aircraft system using multivariate time series clustering and classification techniques[J]. IFAC-PapersOnLine, 2022, 55(10): 1582-1587.
[2]	CHEVROT A, VERNOTTE A, LEGEARD B. CAE: contextual auto-encoder for multivariate time-series anomaly detection in air transportation[J]. Computers & Security, 2022, 116: 102652.
[3]	COELHO E, SILVA L, MURÇA M C R. A data analytics framework for anomaly detection in flight operations[J]. Journal of Air Transport Management, 2023, 110:102409.
[4]	ZHANG W N, HU M H, DU J H. An end-to-end framework for flight trajectory data analysis based on deep autoencoder network[J]. Aerospace Science and Technology, 2022, 127:107726.
[5]	ZHAO W Z, LI L S, ALAM S, et al. An incremental clustering method for anomaly detection in flight data[J]. Transportation Research Part C: Emerging Technologies, 2021, 132:103406.
[6]	KIM J, MAVRIS D. Flight data clustering for offline evaluation of real-time trajectory optimization framework[J]. Decision Analytics Journal, 2023, 7: 100221.
[7]	张堃, 杜睿怡, 时昊天, 等. 基于Mogrifier-BiGRU的飞行器轨迹预测[J]. 兵工学报, 2024, 45(2):373-384. doi: 10.12382/bgxb.2022.0750
	ZHANG K, DU R Y, SHI H T, et al. Aircraft trajectory prediction based on Mogrifier-BiGRU[J]. Acta Armamentarii, 2024, 45(2):373-384. (in Chinese) doi: 10.12382/bgxb.2022.0750
[8]	李楠, 强懿耕, 樊瑞, 等. 基于异常因子的航空器飞行轨迹异常检测研究[J]. 安全与环境学报, 2021, 21(2):643-648.
	LI N, QIANG Y G, FAN R, et al. On the abnormal detection of the aircraft flight trajectory based on the abnormal factor statistics[J]. Journal of Safety and Environment, 2021, 21(2):643-648. (in Chinese)
[9]	SHAFIENYA H, REGAN A C. 4D flight trajectory prediction using a hybrid deep learning prediction method based on ADS-B technology: a case study of Hartsfield-Jackson Atlanta International Airport (ATL)[J]. Transportation Research Part C: Emerging Technologies, 2022, 144:103878.
[10]	DENG H L, LU Y, YANG T, et al. Unmanned aerial vehicles anomaly detection model based on sensor information fusion and hybrid multimodal neural network[J]. Engineering Applications of Artificial Intelligence, 2024, 132:107961.
[11]	LI D Y, GAO H M, YANG K, et al. Abnormal identification of oil monitoring based on LSTM and SVDD[J]. Wear, 2023, 526-527: 204793.
[12]	LUO P, WANG B H, LI T Y, et al. ADS-B anomaly data detection model based on VAE-SVDD[J]. Computers & Security, 2021, 104:102213.
[13]	ZHANG Z, DENG X G. Anomaly detection using improved deep SVDD model with data structure preservation[J]. Pattern Recognition Letters, 2021, 148: 1-6.
[14]	CAI L, YIN H P, LIN J D, et al. A relevant variable selection and SVDD-based fault detection method for process monitoring[J]. IEEE Transactions on Automation Science and Engineering, 2023, 20(4): 2855-2865.
[15]	LI X H, ZHANG H Y, MIAO Y B, et al. CAN bus messages abnormal detection using improved SVDD in internet of vehicles[J]. IEEE Internet of Things Journal, 2022, 9(5): 3359-3371.
[16]	ZHANG C Q, CUI L F, ZHANG Q Z, et al. Online anomaly detection for aeroengine gas path based on piecewise linear representation and support vector data description[J]. IEEE Sensors Journal, 2022, 22(23): 22808-22816.
[17]	邓诗卓, 滕达, 李晓红, 等. 基于球面正则化的支持向量描述视觉异常检测[J]. 仪器仪表学报, 2024, 45(3):315-325.
	DENG S Z, TENG D, LI X H, et al. Support vector description for visual anomaly detection based on spherical regularisation[J]. Journal of Instrumentation, 2024, 45(3): 315-325. (in Chinese)
[18]	张逸豪, 王振雷. 基于最大信息系数的分组支持向量数据描述故障检测[J]. 化工学报, 2023, 74(9): 3865-3878. doi: 10.11949/0438-1157.20230501
	ZHANG Y H, WANG Z L. Maximum information coefficient-based fault detection with grouped support vector data description[J]. Journal of Chemical Engineering, 2023, 74(9): 3865-3878. (in Chinese)
[19]	GUHA D, CHATTERJEE R, SIKDAR B. Anomaly detection using LSTM-based variational autoencoder in unsupervised data in power grid[J]. IEEE Systems Journal, 2023, 17(3): 4313-4323.
[20]	BRADLEY P. The use of the area under the ROC curve in the evaluation of machine learning algorithms[J]. Pattern Recognition, 1997, 30(7):1145-1159.
[21]	LUO P, WANG B H, TIAN J W. TTSAD: TCN-transformer-SVDD model for anomaly detection in air traffic ADS-B data[J]. Computers & Security, 2024, 141: 103840.
[22]	宫文峰, 陈辉, WANG Dan-wei. 基于改进LSTM-SVM的多传感器船舶旋转机械快速故障诊断方法[J]. 船舶力学, 2021, 25(9): 1239-1250.
	GONG W F, CHEN H, WANG D W. A fast fault diagnosis method for multi-sensor ship rotating machinery based on improved LSTM-SVM[J]. Ship Mechanics, 2021, 25(9): 1239-1250. (in Chinese)