Track circuit fault diagnosis method for massive imbalanced data

doi:10.16265/j.cnki.issn1003-3033.2022.05.2389

Abstract

Abstract:

In order to address deviation of decision-making boundary of track circuit diagnosis model due to imbalanced monitoring data and slow training speed caused by massive data, a fault diagnosis method based on data resampling and ensemble learning algorithm was proposed. Firstly, imbalanced data were processed by feature synthesis and resampling including random down-sampling and Synthetic Minority Oversampling Technique (SMOTE). Secondly, a fault diagnosis module for massive monitoring data was constructed based on LightGBM algorithm which could be trained efficiently, training and diagnosis flow was designed, and key parameters were selected by grid search and cross-validation. Finally, Macro-F₁, which was not affected by imbalanced data, was introduced as an evaluation indicator of the model. The results show that the comprehensive performance of each diagnosis model for imbalanced data can be improved by feature synthesis and data resampling. Compared with other algorithms, LightGBM is the best in terms of comprehensive performance and training time, ensuring superiority and rapidity when faced with massive data.

Key words: imbalanced data, track circuit, fault diagnosis, ensemble learning, light gradient boosting machine(LightGBM)

XING Yulong, WANG Jian, SHANGGUAN Wei, PENG Cong, ZHU Linfu. Track circuit fault diagnosis method for massive imbalanced data[J]. China Safety Science Journal, 2022, 32(5): 112-118.

Figures/Tables 11

Fig.1

Fig.2

Tab.1

Tab.2

Tab.3

Fig.3

Fig.4

Tab.4

Tab.5

Fig.5

Fig.6

References 12

[1]	孙上鹏. 无绝缘轨道电路故障诊断方法研究[D]. 北京: 北京交通大学, 2014.
	SUN Shangpeng. Research on fault diagnosis for railway jointless track circuits[D]. Beijing: Beijing Jiaotong University, 2014.
[2]	BRUIN T D, VERBERT K, ROBERT B. Railway track circuit fault diagnosis using recurrent neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2017, 28(3):523-533. doi: 10.1109/TNNLS.2016.2551940
[3]	朱文博, 王小敏. 基于组合决策树的无绝缘轨道电路故障诊断方法研究[J]. 铁道学报, 2018, 40(7):79-84.
	ZHU Wenbo, WANG Xiaomin. Research on fault diagnosis of railway jointless track circuit based on combinatorial decision tree[J]. Journal of the China Railway Society, 2018, 40(7):79-84.
[4]	王梓丞, 张亚东, 郭进, 等. 基于区间二型神经模糊系统的轨道电路故障诊断[J]. 西南交通大学学报, 2021, 56(1):190-196.
	WANG Zicheng, ZHANG Yadong, GUO Jin, et al. Fault diagnosis for track circuit based on interval type-2 neural-fuzzy system[J]. Journal of Southwest Jiaotong University, 2021, 56(1):190-196.
[5]	GALAR M, FERNANDEZ A, BARRENECHEA E, et al. A review on ensembles for the class imbalance problem: bagging-, boosting-, and hybrid-based approaches[J]. IEEE Transactions on Systems Man & Cybernetics Part C, 2012, 42(4):463-484.
[6]	邵良杉, 李相辰. 不平衡数据下的采空区自然发火预测研究[J]. 中国安全科学学报, 2017, 27(6): 61-66. doi: 10.16265/j.cnki.issn1003-3033.2017.06.011
	SHAO Liangshan, LI Xiangchen. Prediction of coal spontaneous combustion in caving zone with unbalanced data[J]. China Safety Science Journal, 2017, 27(6): 61-66. doi: 10.16265/j.cnki.issn1003-3033.2017.06.011
[7]	邵良杉, 周玉. 基于PSO-ELM-Boosting模型的底板破坏深度预测[J]. 中国安全科学学报, 2018, 28(4): 24-29. doi: 10.16265/j.cnki.issn1003-3033.2018.04.005
	SHAO Liangshan, ZHOU Yu. Prediction of destroyed floor depth based on SO-ELM-Boosting model[J]. China Safety Science Journal, 2018, 28(4): 24-29. doi: 10.16265/j.cnki.issn1003-3033.2018.04.005
[8]	YOAV F, ROBERT E S. A decision theoretic generalization of on-line learning and an application to boosting[J]. Journal of Computer and System Sciences, 1997, 55(1):119-139. doi: 10.1006/jcss.1997.1504
[9]	LIAW A, WIENER M. Classification and regression by randomforest[J]. R News, 2002, 2/3:18-22.
[10]	FRIEDMAN J H. Greedy function approximation: a gradient boosting machine[J]. Annals of Statistics, 2001, 29(5):1189-1232. doi: 10.1214/aos/1013203450
[11]	CHEN Tianqi, CARLOS G. XGBoost: a scalable tree boosting system[C]. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 785-794.
[12]	KE Guolin, MENG Q, FINLEY T, et al. Lightgbm: a highly efficient gradient boosting decision tree[C]. Advances in Neural Information Processing Systems, 2017: 3146-3154.

编号	故障模式	样本数量	占比/%
E₀	正常	100 000	98.95
E₁	发送器故障	199	0.20
E₂	发送端电缆模拟网络盘故障	44	0.04
E₃	发送端SPT数字信号电缆故障	55	0.05
E₄	发送端F₁型调谐单元故障	62	0.06
E₅	发送端空心线圈故障	64	0.06
E₆	发送端F₂型调谐单元故障	58	0.06
E₇	接收端F₁型调谐单元故障	22	0.02
E₈	接收端空心线圈故障	25	0.02
E₉	接收端F₂型调谐单元故障	14	0.01
E₁₀	接收端SPT数字信号电缆故障	46	0.05
E₁₁	接收端电缆模拟网络盘故障	62	0.06
E₁₂	接收端衰耗器、接收器故障	94	0.09
E₁₃	补偿电容故障	320	0.32

编号	特征参数名称	正常样本示例
T₁	发送功出电压/V	146.920
T₂	发送功出电流/A	0.254
T₃	发送端电缆模拟网络电缆侧电压/V	41.292
T₄	发送端F₁型调谐单元电流/A	6.372
T₅	发送端空心线圈电流/A	2.210
T₆	发送端F₂型调谐单元电流/A	3.843
T₇	接收端F₁型调谐单元电流/A	2.241
T₈	接收端空心线圈电流/A	0.813
T₉	接收端F₂型调谐单元电流/A	1.414
T₁₀	接收端电缆模拟网络电缆侧电压/V	7.733
T₁₁	接收端电缆模拟网络设备侧电压/V	0.603
T₁₂	主轨道接收(轨出)电压/V	0.472

特征编号	组合方法	参数物理意义	正常状态样本示例
T₁₃	T₁·T₂	发送器功出功率/W	37.318
T₁₄	T₃/T₁	发送端电缆模拟网络对信号的衰减	0.281
T₁₅	T₁₁/T₁₀	接收端电缆模拟网络对信号的衰减	0.078
T₁₆	T₁₂/T₁₁	衰耗器对轨道信号电平调整	0.782

参数	初始值	搜索范围	步长	最优值
η	150	[150,500]	50	300
μ	-1	[3,10]	1	8
θ	0.1	[0.01,0.3]	0.05	0.01
σ	255	[205,305]	10	255
φ	31	[15,63]	4	31

序号	是否特征合成	是否数据重采样	故障诊断模型	Macro-F₁	训练时间/s
1	√	√	LightGBM	0.952 2	7.89
2	√	×	LightGBM	0.922 3	7.95
3	×	√	LightGBM	0.920 2	7.14
4	×	×	LightGBM	0.880 8	7.50
5	√	√	XGBoost	0.949 1	94.17
6	√	×	XGBoost	0.928 6	34.66
7	×	√	XGBoost	0.915 2	39.68
8	×	×	XGBoost	0.887 5	27.43
9	√	√	GBDT	0.946 1	1 144.78
10	√	√	AdaBoost	0.583 0	61.09
11	√	√	RF	0.939 2	26.05
12	√	√	SVM	0.721 3	13.13
13	√	×	SVM	0.220 6	8.24
14	×	√	SVM	0.595 1	10.08
15	×	×	SVM	0.153 5	5.51
16	√	√	MLP	0.907 1	132.82
17	√	×	MLP	0.608 9	128.63
18	×	√	MLP	0.864 2	192.37
19	×	×	MLP	0.599 8	107.60