Research on model of subway operation accident's cause under small sample condition

doi:10.16265/j.cnki.issn1003-3033.2023.03.1134

Abstract

Abstract:

In order to overcome the defect that the traditional accident causation reasoning model was not suitable for the subway operation accident causation reasoning under the condition of small sample, the BN, Bootstrap sampling method and D-S evidence theory were combined to establish the subway operation accident causation reasoning model. Firstly, the accident samples of subway operation over the years were analyzed to determine the accident cause factors and BN nodes. Secondly, based on BN, the Bootstrap sampling method and K2 algorithm were used to learn BN structure. Then the D-S evidence theory was used to correct the BN parameters, and the BN model of subway operation accident causation was established under the condition of small sample. Finally, causal reasoning, diagnostic reasoning and sensitivity analysis were carried out. The results show that the constructed model can effectively predict accident reasoning, and the mean value of node prediction accuracy is 0.858. Through causal reasoning, the most common type of subway operation accident is operation interruption, followed by fire and train conflict. Combined with diagnostic reasoning, it is found that the fault of power supply system is the main cause of operation interruption.

Key words: small sample, metro operation accidents, causation reasoning model, Bayesian network(BN), Bootstrap sampling, D-S evidence theory

WU Haitao, LIU Yue, DU Huimin. Research on model of subway operation accident's cause under small sample condition[J]. China Safety Science Journal, 2023, 33(3): 134-140.

Figures/Tables 12

Fig.1

Tab.1

Fig.2

Fig.3

Fig.4

Tab.2

Conditional probability chart of node C20

节点序号			节点状态与概率
$C 28$			否					是
$C 11$			否		是			否			是
$C 18$		否		是		否	是		否	是		否	是
$C 20$	否	0.87		0.96		0.87	0.75		0.77	0.5		0.5	0.5
$C 20$	是	0.13		0.04		0.13	0.25		0.23	0.5		0.5	0.5

Tab.2

Tab.3

Comprehensive BPA function value of each focal element

焦元	$E 1 ⊕ E 2$	$E 1 — E 2$ $⊕ E 3$	$E 1 — E 3$ $⊕ E 4$	$E 1 — E 4$ $⊕ E 5$	$E 1 — E 5$ $⊕ E 6$	$E 1 — E 6$ $⊕ E 7$	$E 1 — E 7$ $⊕ E 8$	$m i 20$
${p 202}$	0.137 36	0.086 05	0.079 67	0.076 81	0.033 75	0.032 57	0.029 07	0.128 75
${p 205}$	0.142 44	0.142 44	0.120 17	0.110 92	0.040 04	0.038 39	0.034 27	0.151 10
${p 207}$	0.183 14	0.126 26	0.117 44	0.107 87	0.054 14	0.051 60	0.046 06	0.203 09
${p 208}$	0.212 42	0.197 31	0.191 77	0.181 97	0.116 27	0.112 24	0.109 02	0.480 69
${p 202, p 205}$	0	0	0.003 19	0.002 76	0.001 15	0.001 09	0.000 97	0.004 28
${p 202, p 207}$	0.020 07	0.007 50	0.006 68	0.005 78	0.000 73	0.000 69	0.000 62	0.002 73
${p 202, p 205, p 207}$	0	0	0	0	0	0	0.000 61	0.002 69
${p 202, p 205, p 207, p 208}$	0.258 32	0.096 57	0.085 95	0.074 32	0.009 37	0.007 61	0.006 18	0.027 25

Tab.3

Tab.4

Revised conditional probability chart of node C20

节点		节点及概率
$C 28$		否				是
$C 11$		否		是		否		是
$C 18$		否	是	否	是	否	是	否	是
$C 20$	否	0.87	0.90	0.87	0.75	0.86	0.50	0.77	0.3
$C 20$	是	0.13	0.10	0.13	0.25	0.14	0.50	0.23	0.7

Tab.4

Tab.5

Causal reasoning results without adding evidence

排序	节点序号	发生概率/%
1	$C 27$	32.0
2	$C 1$	25.1
3	$C 26$	21.2
4	$C 25$	8.9
5	$C 5$	5.5
6	$C 4$	4.8
7	$C 28$	4.3
8	$C 12$	3.7
9	$C 7$	3.1
10	$C 6$	1.9
11	$C 2$	1.3
12	$C 8$	1.1
13	$C 13$	0.8
14	$C 10$	0.6

Tab.5

Tab.6

Causal reasoning results of train conflict accident under different evidence conditions

排序	列车冲突 $C 26$ 事故发生路径	事故发生概率/%
1	$C 11 → C 26$	62.6
2	$C 17 → C 26$	51.2
3	$C 16 → C 26$	31.7
4	$C 19 → C 26$	30.7
5	$C 28 → C 26$	23.4
6	$C 18 → C 26$	21.4
7	$C 3 → C 26$	21.3
8	$C 22 → C 26$	20.0
9	$C 24 → C 26$	13.0
10	$C 14 → C 26$	12.4
11	$C 20 → C 26$	11.7

Tab.6

Tab.7

Tab.8

References 17

[1]	XU Hui, JIAO Liudan, CHEN Shulin, et al. An innovative approach to determining high-risk nodes in a complex urban rail transit station: a perspective of promoting urban sustainability[J]. Sustainability, 2018, 10(7): DOI:10.3390/su10072456. doi: 10.3390/su10072456
[2]	ZHENG Feng, XU Wenping. Analysis of Beijing subway operation accident network based on complex network theory[C]. Conference Proceedings of the 7^th International Symposium on Project Management, 2019: 187-194.
[3]	SONG Liangliang, LI Qiming, LIST G F, et al. Using an AHP-ISM based method to study the vulnerability factors of urban rail transit system[J]. Sustainability, 2017, 9(6): 1-16. doi: 10.3390/su9010001
[4]	CAKIR E, SEVGILI C, FISKIN R. An analysis of severity of oil spill caused by vessel accidents[J]. Transportation Research Part D: Transport and Environment, 2021, 90: DOI:10.1016/j.trd.2020.102662. doi: 10.1016/j.trd.2020.102662
[5]	周志鹏, 诸泽宇. 基于事件提取和改进MMHC的航空旅客运输事故征候贝叶斯网络建模[J]. 中国安全生产科学技术, 2021, 17(2): 152-158.
	ZHOU Zhipeng, ZHU Zeyu. Bayesian network modeling of aviation passenger transport incident based on event extraction and improved MMHC[J]. Journal of Safety Science and Technology, 2021, 17(2): 152-158.
[6]	鲁义, 伍江乐, 邵淑珍, 等. 基于贝叶斯网络的危化品道路运输事故推理模型[J]. 中国安全科学学报, 2022, 32(3): 174-182. doi: 10.16265/j.cnki.issn1003-3033.2022.03.024
	LU Yi, WU Jiangle, SHAO Shuzhen, et al. Prediction model for road transport accidents of hazardous chemicals based on Bayesian network[J]. China Safety Science Journal, 2022, 32(3):174-182. doi: 10.16265/j.cnki.issn1003-3033.2022.03.024
[7]	FRIEDMAN N, GOLDSZMIDT M, WYNER A. On the application of the Bootstrap for computing confidence measures on features of induced Bayesian networks[C]. Artificial Intelligence and Statistics 99, Proceedings, 1999: 197-202.
[8]	李海龙, 侯艳, 柯朝甫, 等. 基于bootstrap方法的贝叶斯网络结构学习算法在构建基因调控网络中的应用[J]. 中国卫生统计, 2015, 32(2): 217-220.
	LI Hailong, HOU Yan, KE Zhaopu, et al. The application of Bayes network structure learning algorithm based on bootstrap method to the construction of gene regulatory network[J]. Chinese Journal of Health Statistics, 2015, 32(2):217-220.
[9]	SHAFER G. A mathematical theory of evidence[M]. Princeton: Princeton University Press, 1976:85-150.
[10]	WEI Shuhuan, CHEN Yanqiao, GENG Junbao. Bayesian network parameter learning method based on AHP/D-S evidence theory[C]. ACM International Conference Proceeding Series, 2020:DOI:10.1145/3424978.3425149. doi: 10.1145/3424978.3425149
[11]	XU Tingxue, LI Zhiqiang, GU Junyuan, et al. Missile condition assessment based on multi-state Bayesian network[J]. Binggong Xuebao/Acta Armamentarii, 2018, 39(2): 391-398.
[12]	COOPER G F, HERSKOVITS E. A Bayesian method for the induction of probabilistic networks from data[J]. Machine Learning, 1992, 9(4): 309-347.
[13]	SHORTLIFFE E H, BEINLICH I A, SUERMONDT H J, et al. The ALARM monitoring system: a case study with two probabilistic inference techniques for belief networks[C]. AIME 89:Second European Conference on Artificial Intelligence in Medicine, 1989: 247-256.
[14]	杜元伟, 石方园, 杨娜. 基于证据理论/层次分析法的贝叶斯网络建模方法[J]. 计算机应用, 2015, 35(1): 140-146,151. doi: 10.11772/j.issn.1001-9081.2015.01.0140
	DU Yuanwei, SHI Fangyuan, YANG Na. Construction method for Bayesian network based on Dempster-Shafer/analytic hierarchy process[J]. Journal of Computer Applications, 2015, 35(1):140-146,151. doi: 10.11772/j.issn.1001-9081.2015.01.0140
[15]	马小薇. 基于复杂网络的地铁事故致因机理研究[D]. 成都: 西南交通大学, 2019.
	MA Xiaowei. Research on the causation mechanism of metro accidents based on complex network theory[D]. Chengdu: Southwest Jiaotong University, 2019.
[16]	SCHWARZ G. Estimating the dimension of a model[J]. Annals of Statistics, 1978, 6(2):461-464.
[17]	李贤功, 葛家家, 胡婷, 等. 煤矿顶板事故致因分析的贝叶斯网络研究[J]. 中国安全科学学报, 2014, 24(7): 10-14.
	LI Xiangong, GE Jiajia, HU Ting, et al. Bayesian network modeling for causation analysis of coal mine roof accident[J]. China Safety Science Journal, 2014, 24(7):10-14.

事故发生过程简述	事故致因
1995-10-28,阿塞拜疆巴库地铁,一列车因电气故障起火,产生大量浓烟和毒气,造成558人死亡,269人受伤	电气设备故障→ 火灾→中毒
2009-12-22,上海地铁1号线发生侧撞事故,造成列车脱轨,1号线全面停运	列车冲突→脱轨→运营中断
2011-07-5,北京地铁4号线动物园站上行扶梯发生设备故障导致踩踏事件,造成1人死亡,2人重伤,26人轻伤	电梯故障→摔伤→踩踏
2017-09-15,英国伦敦帕森格林地铁站因恐怖分子袭击发生爆炸,造成至少29人受伤	恐怖袭击→爆炸
2021-07-20,郑州暴雨导致地铁5号线一列车被洪水围困,造成14人遇难	自然灾害→窒息

节点	致因节点诊断推理结果
C₁	C₁₄ (24.7%)、C₁₉ (22.4%)、C₂₆ (21.5%)、C₂₀ (21.3%)、C₁₈ (12.3%)、C₂₅ (10.1%)、C₁₁ (9.2%)
C₂₆	C₁₉ (24.5%)、C₁₄ (14.8%)、C₂₀ (12.6%)、C₁₆ (11.3%)、C₁₁ (11.2%)
C₂₅	C₁₁ (33.6%)、C₁₉ (27.9%)、C₂₆ (24.1%)、C₁₄ (14.0%)、C₃ (11.6%)
C₅	C₉ (76.1%)、C₁ (32.2%)、C₁₉ (26.4%)、C₂₆ (24.0%)
C₄	C₁ (29.5%)、C₂₆ (22.2%)、C₁₄ (20.5%)、C₁₉ (16.7%)、C₇ (15.5%)
C₂₇	C₂₀ (23.2%)、C₁₉ (23.1%)、C₁ (20.2%)、C₁₄ (14.6%)、C₁₈ (11.7%)、C₁₆ (11.2%)、C₉ (9.1%)、C₂₈ (8.5%)、C₂₄ (8.0%)、C₁₂ (7.4%)、C₁₁ (6.8%)、C₃ (5.9%)

节点	影响事故节点的关键因素
C₁	C₁₄、C₂₀、C₁₈、C₂₄、C₇、C₁₁、C₂₅、C₂₆等
C₂₆	C₁₇、C₁₁、C₁₆、C₁₉、C₃、C₂₄等
C₂₅	C₁₁、C₃、C₁₉、C₂₆等
C₅	C₉、C₁₀、C₂₃、C₂₄、C₂₈等
C₄	C₇、C₁₄
C₂₇	C₂₈、C₁₂、C₅、C₃、C₂₃、C₂₄、C₂₀、C₁₉、C₁₈、C_17等