基于事件提取和改进MMHC的航空旅客运输事故征候贝叶斯网络建模*

为表征航空旅客运输事故征候演化机理，提出事故征候贝叶斯网络的建模方法。基于事故征候中致因事件、结果事件及分类标准的定义，以7 265起事故征候案例为样本，利用事件提取算法，识别事故征候叙述文本中的致因事件，利用改进的最大最小爬山算法实现网络建模；依据事件提取的测试集验证与结构学习的交叉验证，检验建模算法的准确性与有效性；基于证据敏感性指标，识别关键致因事件。结果表明:航空旅客运输事故征候贝叶斯网络模型包含94个节点和247条有向弧。空降冲突、严重设备故障、机组成员疾病及火灾烟雾是模型中高风险关联的致因事件，在安全监管过程中消除或减弱关键致因事件的发生能有效控制系统风险。

Bayesian network modeling of aviation passenger transport incident based on event extraction and improved MMHC

To characterize the evolution mechanism of aviation passenger transportation incidents,a Bayesian network modeling method for the aviation incidents was proposed.Based on the definition of the causal event,result event and classification criteria in the incident,taking 7265 incident cases as samples,the event extraction algorithm was used to identify the causal event in the incident narrative text,and the improved maximum and minimum hill climbing algorithm was applied to realize the network modeling.According to the test set validation of event extraction and the cross validation of structural learning,the accuracy and effectiveness of the modeling algorithm were verified.The key cause events were identified on the basis of evidence sensitive indexes.The results showed that the Bayesian network model of aviation passenger transportation incidents contained 94 nodes and 247 directed arcs.The airborne conflicts,serious equipment faults,crew member diseases,and fire smoke were the high risk associated causal events in the model.Eliminating or reducing the occurrence of key causal events in the safety supervision process can effectively control the system risk.