基于自然语言处理的航行通告风险识别方法

doi:10.16265/j.cnki.issn1003-3033.2025.S1.0006

摘要/Abstract

摘要： 为实现海量航行通告文本数据的高效精准风险识别,满足现代航班运行对航行通告风险解析的需求,提升民航领域风险识别能力,提出一种融合知识增强的语义表示(ERNIE)模型、卷积神经网络(CNN)以及双向长短期记忆神经网络(BiLSTM)的航行通告风险识别模型;借助ERNIE模型进行词向量的训练,利用CNN捕捉文本局部结构的特性,并结合BiLSTM理解文本的深层次上下文关联;并开展对比试验进行验证。结果表明:相较于其他同类模型,该方法对国内中文和国际英文航行通告的识别精度分别高达92.01%和93.85%。该成果可以为航空公司在航行情报的风险解析和安全管理提供一定的数据支撑。

关键词: 自然语言处理(NLP), 航行通告, 风险识别, 知识增强的语义表示(ERNIE), 卷积神经网络(CNN), 双向长短期记忆网络(BiLSTM)

Abstract:

To achieve efficient and accurate risk identification of massive notice to airmen text data, meet the needs of modern flight operations for notice to airmen risk analysis, and enhance risk identification capabilities in the civil aviation field, a notice to airmen risk identification model integrating the pre-trained ERNIE, CNN, and BiLSTM model is proposed. The ERNIE model is used for word vector training, CNN is utilized to capture the characteristics of the local structure of the text, and the BiLSTM is combined to understand the in-depth contextual associations of the text; comparative experiments are conducted for verification. The results show that compared with other similar models, the recognition accuracy of this method for domestic Chinese and international English notice to airmens is as high as 92.01% and 93.85% respectively. This achievement can provide certain data support for airlines in risk analysis of flight intelligence and safety management.

Key words: natural language processing (NLP), notice to airmen, risk identification, enhanced representation through knowledge integration (ERNIE), convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM)

中图分类号:

X913安全系统学

瞿也丰, 辜汝桐, 黄文强, 陈东玲, 邓李明. 基于自然语言处理的航行通告风险识别方法[J]. 中国安全科学学报, 2025, 35(S1): 33-39.

QU Yefeng, GU Rutong, HUANG Wenqiang, CHEN Dongling, DENG Liming. Risk identification method for notice to airmen based on natural language processing[J]. China Safety Science Journal, 2025, 35(S1): 33-39.

图/表 7

表1

图1

表2

表3

表4

表5

表6

参考文献 15

[1]	刘冰. 航行情报动态信息处理系统的研究与设计[D]. 成都: 电子科技大学, 2016.
	LIU Bing. Research and design of ais dynamic information processing system[D]. Chengdu: University of Electronic Science and Technology, 2016.
[2]	JOULIN A, GRAVE E, BOJANOWSKI P, et al. Bag of tricks for efficient text classification[J]. Proceedings of the 15^th Conference of the European Chapter of the Association for Computational Linguistics,2017: 427-431.
[3]	MIKOLOV T, CHEN K, CORRADO G, et al. Efficient estimation of word representations in vector space[J]. Computer Science, 2013: DOI:10.48550/arXiv.1301.3781.
[4]	徐菁, 闫尊昊, 杨松森, 等. 基于EMI-CNN的建筑施工模板支撑体系节点健康监测[J]. 中国安全科学学报, 2022, 34(7): 83-90.
	XU Jing, YAN Zunhao, YANG Songsen, et al. Health monitoring of joints in construction formwork support systems based on EMI-CNN[J]. China Safety Science Journal, 2024, 34(7): 83-90. doi: 10.16265/j.cnki.issn1003-3033.2024.07.0163
[5]	SUSAN M D, SHINI R. Comparison of word embeddings in text classification based on RNN and CNN[J]. IOP Conference Series: Materials Science and Engineering, 2021, 1187(1):DOI:10.1088/1757-899X/1187/1/012029.
[6]	BANIATA H L, KANG S. Transformer text classification model for arabic dialects that utilizes inductive transfer[J]. Mathematics, 2023, 11(24):DOI:10.3390/math11244960.
[7]	KAUR J S. LSTM based deep learning approach to detect online violent activities over dark web[J]. Multimedia Tools and Applications, 2023, 83(14): 42 379-42 390.
[8]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[J]. ArXiv, 2017: DOI: 10.48550/arXiv.1706.03762.
[9]	张羽, 赵碧柳, 刘红勇. 基于情绪分析的事故风险感知偏差研究[J]. 中国安全科学学报, 2022, 32(8) : 16-22. doi: 10.16265/j.cnki.issn1003-3033.2022.08.1730
	ZHANG Yu, ZHAO Biliu, LIU Hongyong. Study on accident risk perception bias based on emotion analysis[J]. China Safety Science Journal, 2022, 32(8) : 16-22. doi: 10.16265/j.cnki.issn1003-3033.2022.08.1730
[10]	张治鹏, 毛煜升, 张李义. 基于领域ERNIE和BiLSTM模型的酒店评论观点原因分类研究[J]. 数据分析与知识发现, 2022, 6(9): 65-76. doi: 10.11925/infotech.2096-3467.2021.1303
	ZHANG Zhipeng, MAO Yusheng, ZHANG Liyi. Classifying reasons of hotel reviews with domain ERNIE and BiLSTM model[J]. Data Analysis and Knowledge Discovery, 2022, 6(9): 65-76. doi: 10.11925/infotech.2096-3467.2021.1303
[11]	邓益鹏, 罗银辉. 航行通告中自然语言理解算法研究[J]. 电脑知识与技术, 2021, 17(11):206-209.
	DENG Yipeng, LUO Yinhui. Research on natural language understanding algorithm in NOTAM[J]. Computer Knowledge and Technology, 2021, 17(11):206-209.
[12]	李华锋, 曾崇, 胡海青, 等. 基于机器学习的航行通告文本分类研究[J]. 民航学报, 2022, 6(4):6-9.
	LI Huafeng, ZENG Chong, HU Haiqing, et al. Research on text classification of NOTAM based on machine learning[J]. Journal of Civil Aviation, 2022, 6(4): 6-9.
[13]	BAIGANG M, YI F, YU S. NOTAM text analysis and classification based on attention mechanism[J]. Journal of Physics: Conference Series, 2022, 2171(1): DOI: 10.1088/1742-6596/2171/1/012042.
[14]	李华锋, 辜汝桐, 吴东岳. 基于机器学习的民航信息岗工作负荷预测研究[J]. 民航管理, 2024(1):88-93.
	LI Huafeng, GU Rutong, WU Dongyue. Research on workload prediction of civil aviation information-related posts based on machine learning[J]. Civil Aviation Management, 2024(1):88-93.
[15]	董兵, 罗创, 郝宽公, 等. 基于大样本数据的不规范航行通告识别方法[J]. 科学技术与工程, 2024, 24(23): 9 973-9 979.
	DONG Bing, LUO Chuang, HAO Kuangong, et al. A method for identifying irregular NOTAMs based on large sample data[J]. Science Technology and Engineering, 2024, 24(23):9 973-9 979.

风险类型	类型说明
机场关闭	机场因故停止接受航班起降
跑道关闭	机场特定跑道因故停止接受航班起降
航路通告	航路、限制区因故禁止航班飞越
航行资料汇编修订	航行资料汇编修订与发布的提示性通告
运行标准变更	影响航空器着陆进近等运行标准变更
机场不接受备降	机场因故停止接受航班备降
障碍物建立与变更	机场等区域地面障碍物的设立或位置变动
导航设施不使用	导航设施因故不可用
仪表着陆系统不工作	仪表着陆系统因故不可用
滑行道关闭	机场特定滑行道因故不可用
鸟群通告	报告机场等区域影响飞行安全的鸟群提示
卫星导航可用性预测	接收机辅助卫星导航预测可用性的提示性通告

风险通告类型	数量	平均文本长度
机场关闭	168	17.6
跑道关闭	1 848	28.97
航路通告	2 343	148.81
航行资料汇编修订	828	36.18
运行标准变更	365	191.89
机场不接受备降	111	94.03
障碍物建立与变更	153	359.4
导航设施不使用	978	47.24
仪表着陆系统不工作	1 496	25.62
滑行道关闭	3 041	94.42
鸟群通告	137	125.97
卫星导航可用性预测	2 412	78.32

风险通告类型	数量	平均文本长度
机场关闭	4 744	87.12
跑道关闭	15 656	37.35
航路通告	10 592	413.55
航行资料汇编修订	640	38.15
运行标准变更	3 704	248.87
机场不接受备降	2 296	76.44
障碍物建立与变更	6 360	165.56
导航设施不使用	4 760	116.3
仪表着陆系统不工作	7 832	34.26
滑行道关闭	25 360	55.86
鸟群通告	240	90.46
卫星导航可用性预测	23 152	674.17

模型参数	值
卷积核大小	2,3,4
学习率	3×10^-5
批大小	128
隐藏层	384
激活函数	ReLU
优化器	Adam
暂退率	0.5
训练轮数	5

模型	A_c	ω_P	ω_R	${\omega }_{{F}_{1}}$
BERT	87.22	88.7	87.22	87.58
BERT-CNN	88.73	89.64	88.63	88.85
BERT-BiLSTM	88.28	89.47	88.28	88.11
ERNIE	88.29	89.88	88.29	88.11
ERNIE-CNN	91.26	91.27	92.08	91.49
ERNIE-BiLSTM	89.17	90.04	89.17	90.14
ERNIE-CNN-BiLSTM	92.01	92.45	91.98	91.94