基于文本挖掘的空管不正常事件风险预测研究  

Risk Factors Analysis and Prediction of Air Traffic Controller's Abnormal EventsBased on Text Mining

作　　者：毛继志[1] 吴欣蓬 吴磊 汤新民[2] 郭鸿滨[1] MAO Ji-zhi;WU Xin-peng;WU Lei;TANG Xin-min;GUO Hong-bin(China Institute of Aeronautical Radio and Electronics,Shanghai 200241,China;College of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;Air Traffic Management Bureau,Civil Aviation Administration of China,Beijing 100022,China)

机构地区：[1]中国航空无线电电子研究所,上海200241 [2]南京航空航天大学民航学院,江苏南京211100 [3]中国民用航空局空中交通管理局,北京100022

出　　处：《航空计算技术》2020年第1期1-8,12,共9页Aeronautical Computing Technique

基　　金：国家自然科学基金项目资助(61773202);四川省科技计划项目资助(2018JZ0030);中国航空无线电电子研究所航空电子系统综合技术国防科技重点实验室基金项目资助(6142505180407).

摘　　要：人为因素是导致航空事故的主要原因。在空管系统安全问题频发的背景下,以空管不正常事件为研究对象,采用Python语言对空管不正常事件记录进行文本挖掘分析。建立概念向量空间模型,解决一义多词问题。提出将空管风险模式抽象为主题,采用LDA(潜在狄利克雷分配)算法围绕风险主题提取风险致因因素。相比于TF IDF算法,LDA能挖掘出更多潜在风险致因因素,且与专家评审意见基本一致,证明了方法的可靠性,实现了风险致因因素的自动提取。提出将LDA提取的风险致因因素与HFACS模型进行整合,形成以人为因素为中心的风险贝叶斯预测网络。考虑到不正常事件为不完全样本,使用EM算法优化贝叶斯网络参数。通过Netica软件对测试记录进行预测,验证了方案的有效性,同时也证明了文本挖掘结果的正确性与客观性。Human factors are the main causes of aviation accidents.Under the background that safety problems frequently occur in ATC(air traffic control)system,the air traffic controller′s abnormal events are taken as the study object.The abnormal events records are analyzed through the Python language and text mining.Firstly,the concept vector space model is established to solve the problem of polysemy.And then,risk patterns in ATC are regarded as topics in order to use LDA(Latent Dirichlet Assignment)algorithm to make the topic model,and extract the risk factors.Experiments show that LDA topic model can discover more potential risk factors than TF IDF algorithm,and the results are basically consistent with the experts′opinion,which proves the reliability of this method and the success of automatic extraction of risk factors.Finally,the risk factors extracted by LDA are classified and integrated by the HFACS model,to build Bayesian network based human factors.Considering that the abnormal events are incomplete samples,the EM algorithm is used to optimize the parameter of Bayesian network.The validity of this method is verified by Netica software simulation with the test records,and the correctness and objectivity of text mining results are also proved.

关 键 词：空管安全 文本挖掘 概念模型 LDA主题模型 风险致因因素 贝叶斯网络 风险预测 

分 类 号：V328[航空宇航科学与技术—人机与环境工程]