基于主动学习的民航不安全事件风险因素识别与耦合

doi:10.16265/j.cnki.issn1003-3033.2026.08.1066

摘要/Abstract

摘要：

为探究民航不安全事件的风险因素耦合关系,基于22 439起民航不安全事件文本数据,在风险因素识别中引入主动学习(AL)策略,构建基于变换器的双向编码器表示模型(BERT)与双向门控循环单元(BiGRU)融合的风险因素识别模型,自动识别风险因素;在此基础上,应用N-K模型耦合分析民航不安全事件风险因素。研究结果表明:在18 000条训练集上,AL策略将人工标注成本降低70%(12 600条),而模型性能(精确率和召回率的调和平均数F₁=0.907 5)比使用全量数据训练的BERT-BiGRU风险因素识别模型下降1%;在风险耦合值较高的组合中,人、航空器、环境是引发民航不安全事件的关键风险因素;随着参与耦合的因素数量增多,风险耦合值普遍呈现上升态势,其中六要素耦合值最大,为0.371,在民航安全管理中需要避免多因素共同作用。

关键词: 主动学习(AL), 民航不安全事件, 风险因素识别, 风险耦合, N-K模型

Abstract:

To explore the coupling relationship between risk factors of aviation safety occurrences, AL strategies were introduced into risk factor identification based on 22 439 occurrence text records, and a risk factor identification model combining Bidirectional Encoder Representations from Transformers (BERT) and Bidirectional Gated Recurrent Unit (BiGRU) was constructed to achieve automatic identification of risk factors. On this basis, the N-K model was applied to perform a coupling analysis of aviation safety occurrence risk factors. The results show that with 18 000 training samples, the AL strategy reduces manual annotation costs by 70% (12 600 samples), while the model performance (F₁=0.907 5) only decreases by 1% compared with the fully data-trained BERT-BiGRU model. The results of the risk factor coupling analysis show that in combinations with high coupling risk values, human, aircraft, and the environment are the key risk factors that cause aviation safety occurrences. As the number of factors involved in the coupling increases, the risk coupling value generally shows an upward trend, with the coupling value for the six factors being the highest at 0.371. In civil aviation safety management, it is necessary to avoid the combined effects of multiple factors.

Key words: active learning(AL), aviation safety occurrence, risk factor identification, risk coupling, N-K model

中图分类号:

X949

李丽, 邢瑞杰, 包随义. 基于主动学习的民航不安全事件风险因素识别与耦合[J]. 中国安全科学学报, 2026, 36(8): 29-36.

Li Li, Xing Ruijie, Bao Suiyi. Identification and coupling study of risk factors for aviation safety occurrences based on active learning[J]. China Safety Science Journal, 2026, 36(8): 29-36.

图/表 10

图1

图2

表1

表2

图3

表3

表4

表5

表6

图4

参考文献 22

[1]	Lázaro F L, Madeira T, Melicio R, et al. Identifying human factors in aviation accidents with natural language processing and machine learning models[J]. Aerospace, 2025, 12(2): DOI:10.3390/aerospace12020106.
[2]	马建文, 王波, 黄子硕. 基于N-K模型的水上交通事故触发过程及风险因素耦合研究[J]. 重庆交通大学学报(自然科学版), 2022, 41(3): 25-30.
	Ma Jianwen, Wang Bo, Huang Zishuo. Triggering process and risk factor coupling of marine traffic accident based on N-K model[J]. Journal of Chongqing Jiaotong University(Natural Sciences), 2022, 41(3): 25-30.
[3]	Qiao Wanguan. Analysis and measurement of multifactor risk in underground coal mine accidents based on coupling theory[J]. Reliability Engineering & System Safety, 2021, 208: DOI:10.1016/J.RESS.2021.107433.
[4]	方俊, 郭佩文, 朱科, 等. 基于N-K模型的地铁隧道施工安全风险耦合演化分析[J]. 中国安全科学学报, 2022, 32(6):1-9. doi: 10.16265/j.cnki.issn1003-3033.2022.06.2102
	Fang Jun, Guo Peiwen, Zhu Ke, et al. Coupling evolution analysis of subway tunnel construction safety risk based on N-K model[J]. China Safety Science Journal, 2022, 32(6):1-9. doi: 10.16265/j.cnki.issn1003-3033.2022.06.2102
[5]	李一可, 张洪海, 石宗北, 等. 基于N-K-FRAM的空中交通运行安全风险耦合机制[J]. 中国安全科学学报, 2024, 34(5): 175-185. doi: 10.16265/j.cnki.issn1003-3033.2024.05.1552
	Li Yike, Zhang Honghai, Shi Zongbei, et al. Coupling mechanism of air traffic operation safety risk based on N-K-FRAM[J]. China Safety Science Journal, 2024, 34(5): 175-185. doi: 10.16265/j.cnki.issn1003-3033.2024.05.1552
[6]	潘丹, 李永周, 罗帆. 基于N-K模型的飞机运行安全风险耦合研究[J]. 安全与环境学报, 2022, 22(2): 606-614.
	Pan Dan, Li Yongzhou, Luo Fan. Aircraft operational safety risk coupling based on N-K model[J]. Journal of Safety and Environment, 2022, 22(2): 606-614.
[7]	Dehkordi M K, Sattari F, Lefsrud L. Creating an incident investigation framework for a complex socio-technical system: application of multi-label text classification and Bayesian network structure learning[J]. Reliability Engineering & System Safety, 2025, 260: DOI:10.1016/J.RESS.2025.110971.
[8]	Dong Tianxi, Yang Qiwei, Ebadi N, et al. Identifying incident causal factors to improve aviation transportation safety: proposing a deep learning approach[J]. Journal of Advanced Transportation, 2021, 2021(1): 1-15.
[9]	Whang S E, Lee J G. Data collection and quality challenges for deep learning[J]. Proceedings of the VLDB Endowment, 2020, 13(12): 3429-3432. doi: 10.14778/3415478.3415562
[10]	刘颖, 庞羽良, 张伟东, 等. 基于主动学习的图像分类技术:现状与未来[J]. 电子学报, 2023, 51(10): 2960-2984. doi: 10.12263/DZXB.20230397
	Liu Ying, Pang Yuliang, Zhang Weidong, et al. Active learning-based image classification technology: status and future[J]. Acta Electronica Sinica, 2023, 51(10): 2960-2984. doi: 10.12263/DZXB.20230397
[11]	Liu Lei, Lei Wentao, Wan Xiang, et al. Semi-supervised active learning for COVID-19 lung ultrasound multi-symptom classification[C]. Proceedings of the 2020 IEEE 32^nd International Conference on Tools with Artificial Intelligence (ICTAI), IEEE, 2020:1045-1049.
[12]	Demir G, Moslem S, Duleba S. Artificial intelligence in aviation safety: systematic review and biometric analysis[J]. International Journal of Computational Intelligence Systems, 2024, 17(1): 1-30. doi: 10.1007/s44196-023-00380-w
[13]	Zhang Yifeng, Xi Qi, Wang Jing, et al. Architecture and algorithm design for civil aviation data real-time analysis system[J]. IEEE Access, 2024, 12: 66 382-66 395. doi: 10.1109/ACCESS.2023.3346903
[14]	包随义, 李丽. 航空安全六要素及模型研究[J]. 安全, 2022, 43(10): 16-20,28.
	Bao Suiyi, Li Li. Research on six elements and model of aviation safety[J]. Safety & Security, 2022, 43(10): 16-20,28.
[15]	Devlin J, Chang M W, Lee K, et al. Bert: pre-training of deep bidirectional transformers for language understanding[C]. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2019: 4171-4186.
[16]	Cho K, Van Merriënboer B, Gulcehre C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[C]. Association for Computational Linguistics, 2014: 1724-1734.
[17]	Lin Tsungyi, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]. Proceedings of the IEEE International Conference on Computer Vision. IEEE, 2017:2980-2988.
[18]	Wang Qunbo, Zhang Hangu, Zhang Wentao, et al. Deep active learning for multi label text classification[J]. Scientific Reports, 2024, 14(1): DOI:10.1038/S41598-024-79249-7.
[19]	University of Wisconsin Madison. Active learning literature survey[R], 2009.
[20]	Frenay B, Verleysen M. Classification in the presence of label noise: a survey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2014, 25(5): 845-869. doi: 10.1109/TNNLS.2013.2292894 pmid: 24808033
[21]	Godbole S, Sarawagi S. Discriminative methods for multi-labeled classification[C]. Proceedings of the 8^th Pacific-Asia Conference on Knowledge Discovery and Data Mining, 2004: 22-30.
[22]	Kauffman S A. The origins of order: self-organization and selection in evolution[M]. New York: Oxford University Press, 1992: 61-100.

类别	关键参数	参数值
模型架构	嵌入维度	768
	GRU隐藏层维度	256×2
	随机失活	0.2
	优化器	AdamW
超参数	最大序列长度	512
	批大小	128
	学习率	3×10^-5
焦点损失函数	聚焦参数γ	6
焦点损失函数	类别平衡因子α	0.25, 0.44, 0.87, 0.41, 2.12, 1.92

风险因素	P	R	F₁	宏F₁值
人	0.99	0.98	0.98	0.916 7
机	0.96	0.91	0.94
设	0.80	0.92	0.86
环	0.96	0.94	0.95
技	0.98	0.89	0.93
规	0.85	0.82	0.84

风险因素	P	R	F₁	宏F₁值
人	0.99	0.97	0.98	0.907 5
机	0.95	0.91	0.93
设	0.77	0.90	0.83
环	0.91	0.97	0.94
技	0.96	0.96	0.96
规	0.78	0.82	0.80

事件代码	概率	事件代码	概率	事件代码	概率	事件代码	概率
000000	0.013 5	100100	0.233 5	000101	8.91×10^-5	100101	0.001 2
100000	0.402 0	100010	0.000 8	000011	4.46×10^-5	100011	0.000 2
010000	0.145 0	100001	0.004 0	111000	0.000 8	011100	8.91×10^-5
001000	0.000 9	011000	0.001 4	110100	0.006 6	011010	4.46×10^-5
000100	0.032 0	010100	0.010 1	110010	0.000 3	010011	8.91×10^-5
000010	0.000 3	010010	0.001 6	110001	0.000 7	111100	8.91×10^-5
000001	8.91×10^-5	010001	0.000 9	101100	0.010 3	111001	4.46×10^-5
110000	0.076 3	001100	0.000 5	101001	0.000 1	110101	0.000 1
101000	0.055 8	000110	4.46×10^-5	100110	0.000 1	100111	4.46×10^-5

类型	变化风险概率
单因素	P_0.....=0.206 8	P_1.....=0.793 2	P_.0....=0.755 6	P_.1....=0.244 4	P_..0...=0.929 9
	P_..1...=0.070 1	P_...0..=0.705 2	P_...1..=0.294 8	P_....0.=0.996 3	P_....1.=0.003 7
	P_.....0=0.992 3	P_.....1=0.007 8	—	—	—
双因素	P_00....=0.047 5	P_01....=0.159 3	P_10....=0.708 1	P_11....=0.085 1	P_0.0...=0.203 8
双因素	P_0.1...=0.002 9	P_1.0...=0.726 1	P_1.1...=0.067 2	P_0..0..=0.164	…
3因素	P_000...=0.046 1	P_001...=0.001 4	P_010...=0.157 8	P_011...=0.001 6	P_100...=0.641 9
3因素	P_101...=0.066 2	P_110...=0.084 1	P_111...=0.000 9	P_00.0..=0.014 9	…
4因素	P_0000..=0.014	P_0001..=0.032 1	P_0010..=0.000 9	P_0011..=0.000 5	P_0100..=0.147 7
4因素	P_0101..=0.010 1	P_0110..=0.001 5	P_0111..=8.91×10^-5	P_1000..=0.407	…
5因素	P_00000.=0.013 6	P_00001.=0.000 4	P_00010.=0.032	P_00011.=4.46×10^-5	P_00100.=0.000 9
5因素	P_00101.=0	P_00110.=0.000 5	P_00111.=0	P_01000.=0.146	…
6因素	P₀₀₀₀₀₀=0.013 5	P₀₀₀₀₀₁=8.91×10^-5	P₀₀₀₀₁₀=0.000 3	P₀₀₀₀₁₁=4.46×10^-5	P₀₀₀₁₀₀=0.032
6因素	P₀₀₀₁₀₁=8.91×10^-5	P₀₀₀₁₁₀=4.46×10^-5	P₀₀₀₁₁₁=0	P₀₀₁₀₀₀=0.000 9	…