航空安全事件关联分析方法研究

为有效预防航空安全事件的发生,对航空安全事件进行关联分析,提出了一种改进的FP-Growth算法实现航空安全事件多层关联规则挖掘。首先,采用Pawlak属性重要度约简算法对航空安全事件属性进行约简,结合航空安全领域本体中事件属性的概念层次结构对属性编码;通过引入跨层次频繁项和修补项改进FP-Growth算法,每层设置不同的支持度自底向上对编码后的多层次属性进行挖掘,最后得到事件原因、事件结果、运行阶段、事件类型等多种事件属性之间的多层关联规则。研究表明,该方法能够实现航空安全事件的关联分析,为减少航空安全事件的发生提供语义服务和决策支持。

On the correlation analysis method of the aviation safety incidents

This paper intends to offer the results of our improved associated analysis of the aviation safety events based on a FPGrowth algorithm so as to realize a multi-layer association rule exploration of the aviation safety events.And,for the said purpose,we have to improve the FP-Growth algorithm by introducing the cross-level frequent items and the patch items by setting the different supporting degrees at the level of all the layers.The improved FP-Growth algorithm enjoys indeed the advantage of reducing the database scanning time and the improved execution efficiency.The improved algorithm can mainly be divided into 2 steps,that is,the events FP-Tree construction and the FP-Tree for the events search.To achieve the above said 2 steps,it is necessary to generate the frequent item sets,the frequent items across the levels and the reparation items by setting the different supports for each layer according to the concept of the event-attribute hierarchical sets.And,next,the exploration event FPTree method can help to produce the frequent item sets and the strong association rules for the minimum confidence construction,for the goal of the rule generation is to extract all the full confidence rules from the frequent item sets gained from the previously prepared steps.And,as a matter of fact,to search for the association rules,we have,first of all,taken 1055 aviation safety incidents reported in the world aviation safety accident investigation reports as the data origin and manually extracted attributes of the data to form the aviation safety event attributes sets.And,then,we have to carry out the attribute reduction based on the Pawlak attribute importance reduction algorithm to remove the redundant attributes,and encode the filtered attributes in accordance with the conceptual hierarchy of the event attributes in the aviation safety domain ontology.And,at last,we have managed to work out the association rules and discover the implied association of the events via the improved FP-Growth algorithm.The association rules we have mined out and determined may include not only various types of events,such as the engine failure,heavy landing and flight safety,but also define and attribute the various types of events to the different conceptual levels,such as the event cause,the event result,the event types and their operation stages.The results of the above research indicate that the taking-place of the above events is closely related to the human interference factors,the environmental factors,the mechanical factors,as well as the other influential factors.Thus,consequently,the information exploration association rules can be expected to provide semantic services and decision-making supports for the improvement and heightening of the civil aviation safety probability.