Online anomaly detection method integrating LSTM and MGD for suspension system of maglev trains
-
摘要: 针对磁浮列车悬浮系统的在线异常检测问题，提出了一种基于长短时记忆(LSTM)神经网络与多元高斯分布(MGD)的检测方法；建立学习悬浮系统正常运行的LSTM时间序列预测模型，得到了正常情况下的预测误差；基于间隙、电流和加速度的预测误差，建立了反映正常情况下预测误差分布特性的MGD模型；以对数概率密度作为检测指标，设计了在线检测逻辑，并以F1分数作为检测效果衡量指标设置阈值；为了验证所提方法的有效性，利用磁浮列车运营线数据模拟在线数据，采用所提方法对过轨道接缝异常、砸轨异常和加速度信号异常进行了检测与分析。研究结果表明：在检测以上3类异常时，所提方法的F1分数分别达到了100.00%、97.85%和83.33%，所提方法的检测指标在正常和异常情况下对比明显，可以反映出悬浮系统产生异常到调整好的具体时间段，并且算法平均耗时约2 s；相较于基于超球体高斯分布方法，所提方法检测率平均提高了1.9%，其中对于持续时间短的过轨道接缝异常的检测率提高了9.4%。可见，所提方法可以对悬浮系统状态数据进行异常在线检测。Abstract: In view of online anomaly detection of suspension system of maglev trains, a detection method based on long short-time memory (LSTM) neural network and multivariate Gaussian distribution (MGD) was proposed. The LSTM time series prediction model for learning the normal operation of the suspension system was established, and the prediction errors under normal conditions were obtained. Based on the prediction errors of the gap, current, and acceleration, the MGD model reflecting the distribution characteristics of the prediction errors under normal conditions was built. The log probability density was used as the detection indicator, the online detection logic was designed, and the threshold was set using F1 score as the evaluation criterion of the detection effect. To verify the effectiveness of the proposed method, the operation line data of the maglev train were used to simulate online data, and the proposed method was applied to detect and analyze track crossing joint, rail smashing, and acceleration signal anomalies. Research results show that the F1 scores of the proposed method for detecting the above three anomalies are 100.00%, 97.85%, and 83.33%, respectively. The detection indicators of the proposed method are significantly different between normal and abnormal conditions, reflecting the specific time period from anomaly occurrence of the suspension system to normal adjustment, and the algorithm takes about 2 s on average. Compared with the method based on the Gaussian distribution of hyperspheres, the proposed method achieves an average improvement of 1.9% in detection rate. Specifically, it achieves a 9.4% increase in detecting short-duration track crossing joint anomaly. Therefore, the proposed method can achieve online anomaly detection of the suspension system's state data.
-
Key words:
- maglev train /
- suspension system /
- anomaly detection /
- LSTM neural network /
- time series prediction model /
- MGD
-
图 4 间隙、电流、加速度预测误差分布
Figure 4. Distributions of prediction errors of gap, current, and acceleration
图 6 磁浮列车悬浮数据采集系统设计
Figure 6. Design of levitation data acquisition system for maglev trains
图 16 基于训练集预测误差的MGD模型
Figure 16. MGD model based on prediction errors of training datasets
图 19 过轨道接缝异常测试集预测误差曲线
Figure 19. Prediction error curves of track crossing joint anomaly test datasets
图 20 过轨道接缝异常测试集检测结果
Figure 20. Detection result of track crossing joint anomaly test datasets
图 21 砸轨异常测试集预测误差曲线
Figure 21. Prediction error curves of rail smashing anomaly test datasets
图 23 加速度信号异常测试集预测误差曲线
Figure 23. Prediction error curves of acceleration signal anomaly test datasets
图 24 加速度信号异常测试集检测结果
Figure 24. Detection result of acceleration signal anomaly test datasets
表 1 常用激活函数及其用途
Table 1. Common activation functions and their applications
激活函数 用途 Rectified Linear Unit (ReLU) 用于隐藏层神经元输出 Sigmoid 用于隐藏层神经元输出 Softmax 用于多分类神经网络输出 Tanh 用于输入层和输出层 Linear 用于回归神经网络输出 
下载: 导出CSV
表 2 LSTM网络结构参数
Table 2. LSTM network structure parameters
参数名称 参数设置 Batch Size 32 Epochs 40 Dropout 0.1 Look Back 12 Look Ahead 1 Number of Hidden Layer 1 Number of Hidden Layer Unit 80 损失函数 均方误差 学习率 0.02
下载: 导出CSV
表 4 各数据集相关的数据情况
Table 4. Data situations of datasets
数据集类型 正常值数量 异常值数量 合计 Xtg、Xta、Xtc 25 553 0 25 553 Xvg、Xva、Xvc 8 114 0 8 114 Vg1、Va1、Vc1 7 834 29 7 863 Vg2、Va2、Vc2 1 831 1 310 3 141 Vg3、Va3、Vc3 233 91 324 Tg1、Ta1、Tc1 6 933 26 6 959 Tg2、Ta2、Tc2 766 1 304 2 070 Tg3、Ta3、Tc3 283 24 307
下载: 导出CSV
表 5 文献[7]中的异常检测方法对3类异常的检测结果
Table 5. Detection results of anomaly detection methods in reference [7] for three types of anomalies
异常类型 异常数量 超球体高斯分布 经验阈值 PCA SVDD 检测数量 检测率/% 检测数量 检测率/% 检测数量 检测率/% 检测数量 检测率/% 过轨道接缝异常 32 29 90.6 0 0.0 24 75.0 23 71.9 砸轨异常 104 104 100.0 104 100.0 104 100.0 104 100.0 加速度信号异常 41 39 95.1 0 0.0 34 82.9 32 78.0
下载: 导出CSV
表 6 本文方法对3类异常的检测率结果
Table 6. Detection results of proposed method for three types of anomalies
异常类型 异常数量 本文方法 检测数量 检测率/% 过轨道接缝异常 26 26 100.0 砸轨异常 1 304 1 300 99.7 加速度信号异常 24 22 91.7
下载: 导出CSV
-
[1] 马卫华, 罗世辉, 张敏, 等. 中低速磁浮车辆研究综述[J]. 交通运输工程学报, 2021, 21(1): 199-216. doi: 10.19818/j.cnki.1671-1637.2021.01.009MA Wei-hua, LUO Shi-hui, ZHANG Min, et al. Research review on medium and low speed maglev vehicle[J]. Journal of Traffic and Transportation Engineering, 2021, 21(1): 199-216. (in Chinese) doi: 10.19818/j.cnki.1671-1637.2021.01.009 [2] 邓自刚, 刘宗鑫, 李海涛, 等. 磁悬浮列车发展现状与展望[J]. 西南交通大学学报, 2022, 57(3): 455-474, 530.DENG Zi-gang, LIU Zong-xin, LI Hai-tao, et al. Development status and prospect of maglev train[J]. Journal of Southwest Jiaotong University, 2022, 57(3): 455-574, 530. (in Chinese) [3] 徐飞, 罗世辉, 邓自刚. 磁悬浮轨道交通关键技术及全速度域应用研究[J]. 铁道学报, 2019, 41(3): 40-49. doi: 10.3969/j.issn.1001-8360.2019.03.006XU Fei, LUO Shi-hui, DENG Zi-gang. Study on key technologies and whole speed range application of maglev rail transport[J]. Journal of the China Railway Society, 2019, 41(3): 40-49. (in Chinese) doi: 10.3969/j.issn.1001-8360.2019.03.006 [4] 梁潇, 陈峰, 傅庆湘. 160 km·h-1中速磁浮交通系统的关键技术问题[J]. 城市轨道交通研究, 2019, 22(9): 21-26.LIANG Xiao, CHEN Feng, FU Qing-xiang. Key technical issues on 160 km·h-1 medium-speed maglev transit system[J]. Urban Mass Transit, 2019, 22(9): 21-26. (in Chinese) [5] NASIRI Z R, HEKMATI A. A review of suspension and traction technologies in maglev trains[C]//IEEE. 34th International Power System Conference(PSC). New York: IEEE, 2019: 129-135. [6] PHAENKONGNGAM T, CHINNAWONG K, PATUMASUIT N, et al. Reviewing propulsion and levitation system for magnetic levitation train[C]//IEEE. 2021 International Electrical Engineering Congress. New York: IEEE, 2021: 185-188. [7] 王平, 梅子, 龙志强. 基于超球体高斯分布的悬浮系统异常检测[J]. 机车电传动, 2021(6): 9-17.WANG Ping, MEI Zi, LONG Zhi-qiang. Anomaly detection for suspension systems based on the Gaussian distribution of hyperspheres[J]. Electric Drive for Locomotives, 2021(6): 9-17. (in Chinese) [8] 王平, 梅子, 龙志强. 基于改进典型相关分析的中低速悬浮系统异常检测方法[J]. 同济大学学报(自然科学版), 2022, 50(2): 241-252.WANG Ping, MEI Zi, LONG Zhi-qiang. Anomaly detection method of middle-low speed suspension system based on improved canonical correlation analysis[J]. Journal of Tongji University (Natural Science), 2022, 50(2): 241-252. (in Chinese) [9] 罗建辉, 王平. 基于海林格距离和相关系数的中低速悬浮系统异常检测方法[J]. 铁道科学与工程学报, 2022, 19(10): 3096-3106.LUO Jian-hui, WANG Ping. Anomaly detection method of middle-low speed suspension system based on Hellinger distance and correlation coefficient[J]. Journal of Railway Science and Engineering, 2022, 19(10): 3096-3106. (in Chinese) [10] ZHANG Yun-zhou, CHAO Chuang, WU Jun, et al. Magnetic anomaly of long track detection method based on wavelet combining with fractal for high speed maglev transit[J]. Measurement and Control, 2022, 55(7/8): 717-728. [11] 周旭, 温韬, 龙志强. 基于漏检率的磁浮列车悬浮系统异常检测[J]. 西南交通大学学报, 2023, 58(4): 903-912.ZHOU Xu, WEN Tao, LONG Zhi-qiang. Anomaly detection of suspension system in maglev train based on missed detection rate[J]. Journal of Southwest Jiaotong University, 2023, 58(4): 903-912. (in Chinese) [12] 龙子凡, 司恩, 罗华军, 等. 一种磁浮列车异常振动监测与诊断方法[J]. 科学技术创新, 2022(10): 21-24.LONG Zi-fan, SI En, LUO Hua-jun, et al. A method for monitoring and diagnosing abnormal vibration of maglev train[J]. Scientific and Technological Innovation, 2022(10): 21-24. (in Chinese) [13] 吴峻, 李洪鲁, 张雨馨, 等. 中低速磁浮轨道疑似不平顺的Box-Whisker图筛选法[J]. 交通运输工程学报, 2023, 23(3): 68-76. doi: 10.19818/j.cnki.1671-1637.2023.03.004WU Jun, LI Hong-lu, ZHANG Yu-xin, et al. Screening method of suspected irregularity for medium-and-low-speed maglev tracks based on Box-Whisker plot[J]. Journal of Traffic and Transportation Engineering, 2023, 23(3): 68-76. (in Chinese) doi: 10.19818/j.cnki.1671-1637.2023.03.004 [14] CHEN Po-yu, YANG Shu-sen, MCCANN J A. Distributed real-time anomaly detection in networked industrial sensing systems[J]. IEEE Transactions on Industrial Electronics, 2015, 62(6): 3832-3842. doi: 10.1109/TIE.2014.2350451 [15] AHMAD S, LAVIN A, PURDY S, et al. Unsupervised real-time anomaly detection for streaming data[J]. Neurocomputing, 2017, 262: 134-147. doi: 10.1016/j.neucom.2017.04.070 [16] XIE Kun, LI Xiao-can, WANG Xin, et al. On-line anomaly detection with high accuracy[J]. IEEE/ACM Transactions on Networking, 2018, 26(3): 1222-1235. doi: 10.1109/TNET.2018.2819507 [17] LUO Jian, HONG Tao, YUE Meng. Real-time anomaly detection for very short-term load forecasting[J]. Journal of Modern Power Systems and Clean Energy, 2018, 6(2): 235-243. doi: 10.1007/s40565-017-0351-7 [18] FISCH A T M, BARDWELL L, ECKLEY I A. Real time anomaly detection and categorisation[J]. Statistics and Computing, 2022, 32(4): 55. doi: 10.1007/s11222-022-10112-3 [19] ZHOU Yan-jun, REN Huo-rong, LI Zhi-wu, et al. Anomaly detection based on a granular Markov model[J]. Expert Systems with Applications, 2022, 187: 115744. doi: 10.1016/j.eswa.2021.115744 [20] LIU Sheng-hua, ZHOU Bin, DING Quan, et al. Time series anomaly detection with adversarial reconstruction networks[J]. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(4): 4293-4306. doi: 10.1109/TKDE.2021.3140058 [21] 王德文, 杨力平. 智能电网大数据流式处理方法与状态监测异常检测[J]. 电力系统自动化, 2016, 40(14): 122-128. doi: 10.7500/AEPS20150828001WANG De-wen, YANG Li-ping. Stream processing method and condition monitoring anomaly detection for big data in smart grid[J]. Automation of Electric Power Systems, 2016, 40(14): 122-128. (in Chinese) doi: 10.7500/AEPS20150828001 [22] 陈兴蜀, 江天宇, 曾雪梅, 等. 基于多维时间序列分析的网络异常检测[J]. 工程科学与技术, 2017, 49(1): 144-150.CHEN Xing-shu, JIANG Tian-yu, ZENG Xue-mei, et al. Network anomaly detector based on multiple time series analysis[J]. Advanced Engineering Sciences, 2017, 49(1): 144-150. (in Chinese) [23] 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 Xin-peng, 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) [24] 时磊. 基于LSTMs-Autoencoder的流数据异常检测算法[J]. 仪表技术与传感器, 2021(10): 120-125. doi: 10.3969/j.issn.1002-1841.2021.10.024SHI Lei. Anomaly detection algorithm in streaming data based on LSTMs-Autoencoder[J]. Instrument Technique and Sensor, 2021(10): 120-125. (in Chinese) doi: 10.3969/j.issn.1002-1841.2021.10.024 [25] 毛文涛, 田思雨, 窦智, 等. 一种基于深度迁移学习的滚动轴承早期故障在线检测方法[J]. 自动化学报, 2022, 48(1): 302-314.MAO Wen-tao, TIAN Si-yu, DOU Zhi, et al. A new deep transfer learning-based online detection method of rolling bearing early fault[J]. Acta Automatica Sinica, 2022, 48(1): 302-314. (in Chinese) [26] 陈仲磊, 伊鹏, 陈祥, 等. 基于集成学习的系统调用实时异常检测框架[J]. 计算机工程, 2023, 49(6): 162-169, 179.CHEN Zhong-lei, YI Peng, CHEN Xiang, et al. Real-time anomaly detection framework via system calls based on integrated learning[J]. Computer Engineering, 2023, 49(6): 162-169, 179. (in Chinese) [27] GUE I H V, UBANDO A T, TSENG M, et al. Artificial neural networks for sustainable development: a critical review[J]. Clean Technologies and Environmental Policy, 2020, 22: 1449-1465. doi: 10.1007/s10098-020-01883-2 [28] SHARMA S, SHARMA S, ATHAIYA A. Activation functions in neural networks[J]. International Journal of Engineering Applied Sciences and Technology, 2020, 4(12): 310-316. [29] SALEHINEJAD H, SANKAR S, BARFETT J, et al. Recent advances in recurrent neural networks[J]. arXiv, 2017, DOI: 10.48550/arXiv.1801.01078. [30] YU Yong, SI Xiao-sheng, HU Chang-hua, et al. A review of recurrent neural networks: LSTM cells and network architectures[J]. Neural Computation, 2019, 31(7): 1235-1270. doi: 10.1162/neco_a_01199 [31] LIU Dian-yu, SUN Chuan-le, GAO Jun. Machine learning of log-likelihood functions in global analysis of parton distributions[J]. arXiv, 2022, DOI: 10.48550/arXiv.2201.06586. [32] DING Mei-mei, TIAN Hui. PCA-based network traffic anomaly detection[J]. Tsinghua Science and Technology, 2016, 21(5): 500-509. [33] 杨敏, 张焕国, 傅建明, 等. 基于支持向量数据描述的异常检测方法[J]. 计算机工程, 2005, 31(3): 39-42.YANG Min, ZHANG Huan-guo, FU Jian-ming, et al. Anomaly intrusion detection method based on SVDD[J]. Computer Engineering, 2005, 31(3): 39-42. (in Chinese) -
点击查看大图
计量
- 文章访问数: 111
- HTML全文浏览量: 15
- PDF下载量: 43
- 被引次数: 0
