贝叶斯网络在CFIT风险评价中的应用

针对事故树分析法(FTA)在风险评价中的局限性,在可控飞行撞地(CFIT)事故树的基础上,建立贝叶斯网络(BN).运用推理运算对贝叶斯网络进行定量分析,通过分析计算数据,寻找主要事故致因,并提出对应的改进措施.再将改进措施引入到贝叶斯网络中,评价相关措施的有效性.结果表明,改进措施后,高度设置错误的后验概率最大,将成为预防CFIT的工作重点.最后指出贝叶斯网络方法是对传统的基于故障树分析的风险评价方法的有益改进.

On the application of Bayesian network to hazard assessment in CFIT

This paper is concerned about its authors' research on the application of Bayesian network to hazard assessment in the controlled flight into terrain (CFIT) . As is known, fault tree approach has been traditionally used in hazard assessment. However, since it has had its limitations in analyzing the likely reasons leading to the accidents, we suggest here applying Bayesian network to make inference by using its quantitative algorithm. In so doing, the fault tree of the CFIT can be mapped into a Bayesian network, which contains a conditional probability distribution table so as to assess the system by inferring the Bayesian network. Further analysis of the data gained, we have found out that there exist some key factors that may lead to the CFTT, such as the setting error of the flying height, the losing of height alertness, the failure of correcting operating measures, or the failure in effective checking and examination of some other data in flight control. According to the above consideration, this paper has brought about due improved measures, such as crossing/overlapping examinations, buildup of alertness training, enhanced ground proximity warning system ( EGPWS), etc. While introducing improved measures into Bayesian networks, we have made careful consideration of the effectiveness of related measures so as for Bayesian network to be able to make a full and thorough hazard assessment to CFIT. For its distinguished nature, Bayesian network enables us to assess the effectiveness of the related improved measures, which help to reduce the likeliness of CFIT. Owing to the advantages of Bayesian network in comparison with the fault tree analysis, it can not only deal with the uncertain information but also make beforehand inference ( prediction) as well as the back-sequential inference (diagnosis), which can help us to work out a highly probable mode that may result in the system failure. Furthermore, the network can help to make account of the changing conditions of nodes induced by the variation of any other nodes of networks, which the fault tree approach can't. Therefore, it can be said in a word that the Bayesian network approach can be taken as a good substitute for fault tree approach for hazard assessment with promising perspective of application.