基于文本挖掘的民航飞行风险评价指标研究  

Study of flight risk assessment indicator based on text mining

作　　者：汪磊 安佳宁 史少铭 WANG Lei;AN Jianing;SHI Shaoming(School of Safety Science and Engineering,Civil Aviation University of China,Tianjin 300300,China;Flight Department,Shandong Airlines,Jinan 250014,China)

机构地区：[1]中国民航大学安全科学与工程学院,天津300300 [2]山东航空股份有限公司飞行部,济南250014

出　　处：《安全与环境学报》2025年第3期825-834,共10页Journal of Safety and Environment

基　　金：民航局安全能力建设资金项目(KJZ49420210076)。

摘　　要：为定量评价民航飞行风险,研究提出一种基于文本挖掘的民航飞行风险评价指标识别方法。该方法聚焦于冲偏出跑道、可控飞行撞地、空中失控3类典型核心风险事件,收集全球运输航空2008-2023年相关事故调查报告共210篇。利用词频与逆文档频率算法(Term Frequency-Inverse Document Frequency,TF-IDF)和潜在狄利克雷分布主题模型(Latent Dirichlet Allocation,LDA)提取语料中主题及关键词,参考航空公司飞行品质监控标准文件,归纳3类核心风险评价指标,并结合相关文献规范,构建民航飞行风险评价指标体系。采集某航空公司B737-800机型60条航班数据,对评价指标体系的合理性开展实例验证。结果显示:该方法能够客观高效地识别飞行风险指标,实现了对5名飞行员个体风险的量化排序。研究结果可应用于飞行风险评价,为后续建立风险量化模型奠定基础。To get insight into the quantitative evaluation of civil aviation flight risk,a novel method utilizing text-mining techniques to identify evaluation indicators is proposed.This study mainly focuses on three typical core risk events,including Runaway Excursion(RE),Controlled Flight Into Terrain(CFIT),and Loss of Control(LOC).To initiate this study,a comprehensive corpus consisting of 210 pertinent accident investigation reports was constructed,covering global air transport accidents from the year 2008 to 2023.Through the implementation of the Term Frequency-Inverse Document Frequency(TF-IDF)algorithm and the Latent Dirichlet Allocation(LDA)model,relevant topics and keywords associated with the targeted risk events were extracted from the corpus.In this way,it is possible to discover the underlying risk characteristics,which in turn helps to filter out representative indicators.Subsequently,the evaluation indicators system was established by referencing the Flight Operational Quality Assurance(FOQA)document employed by airline companies and incorporating relevant literature as well as industry specifications.To validate the rationality of the indicator system,Quick Access Recorder(QAR)data from 60 B737-800 flights operated by five individual pilots from Airline A was collected.By classifying the exceedance risk level for the monitored parameters of the risk indicators,individual risk values for the five pilots were quantified and analyzed in a targeted manner.The results demonstrate that this method is capable of identifying flight risk indicators effectively and objectively,enabling the quantitative ranking of individual pilots'flight risk.This capability holds significant implications for the development of targeted interventions,personalized training programs,and the enhancement of overall flight safety.By integrating the power of text mining and data analytics,the application of this method in flight risk evaluation serves as a cornerstone for the subsequent development of risk quantification models.As such,thi

关 键 词：安全工程 风险评价 文本挖掘 词频与逆文档频率 潜在迪利克雷分布 

分 类 号：X949[环境科学与工程—安全科学]