基于事理图谱的飞行冲突事故演化研究

Research on the evolution of flight conflict accidents based on event logic graph

为识别飞行冲突事故的诱发事件与关键演化链路,提出了一种基于事理图谱的事故分析方法。首先依据系统理论过程分析(System Theoretic Process Analysis,STPA)方法辨识飞行冲突场景潜在风险事件,归纳出描述各组件交互过程的场景分析框架;再利用自然语言处理方法从事故文本中提取结构化事件,基于语义相似度匹配建立与风险事件的映射,组成事件演化链条,构建事理图谱;通过对飞行冲突场景进行仿真,运用统计指标定量分析网络拓扑结构,揭示事件对事故演化过程的影响力。结果表明,构建的事理图谱可视化诊断出机组突破管制员指令、机组操作不当导致偏航、航空器机上通讯导航设备故障、管制员纠正指令发布过晚、空中交通预警和防撞系统(Traffic Alert and Collision Avoidance System,TCAS)发出异常措施通告/活动通告(Resolution Advisory/Traffic Advisory,RA/TA)警告、航空器出现飞行汇聚趋势6个诱发事件与3条关键演化链路,其中机组突破管制员指令与管制员指令发布过晚为重点管控事件。

This paper proposed an accident analysis method that can identify the inducing events and key evolutionary links of flight conflict accidents based on the Event Logic Graph.To this end,we identified 37 potentially unsafe control behaviors and their corresponding risk events for four processes in a flight conflict scenario based on Systems-Theoretic Process Analysis(STPA).Thus,the scene analysis framework describing the interaction process of each component was summarized.The lexical,syntactic,and grammatical information contained in the event text is extracted using natural language processing methods and stored in a structured form to evolve the event.We mapped evolutionary events to risk events based on semantic similarity and formed an event evolutionary chain.In this process,we used events as the nodes and the event evolution relationship as the edges and constructed the Event Logic Graph of flight conflict events.Finally,we simulated flight conflict scenarios and quantitatively analyzed the topological properties of the network in terms of node degree,node strength,and clustering coefficient of the Event Logic Graph.In this way,we obtained different effects of different events on the evolution of the overall flight conflict event.And explored the evolutionary patterns of successive events and models of event evolution.Then we excavated the induced events and key evolution links in the evolution process of flight conflict.The results show that the constructed Event Logic Graph visually diagnosed six triggering events and three key evolutionary chains:crew breaking controller instructions,crew mismanagement leading to yaw,aircraft in-flight communication and navigation equipment failure,the controller issuing instructions too late,TCAS(Traffic Alert and Collision Avoidance System)issuing abnormal RA/TA(Resolution Advisory/Traffic Advisory)warnings,and aircraft showing flight convergence trend.Among them,the key control events were the crew breaking controller instructions and the controller issuing instructions too late.