基于粗糙集与遗传算法的航空器空中相撞事故因素分析

为高效准确挖掘航空器空中相撞事故与相应事故致因因素间的规律,利用事故数据,提出一种基于粗糙集和遗传算法的分析模型。首先运用粗糙集理论对SHELL模型提供的第一层次因素进行重要性程度排序,并确定其权重;然后分析事故中各环节的具体因素,结合遗传算法对这些子因素进行约简,约简时重点考虑重要度高的主因素,确定最终的事故因素及其决策规则;最后,实例分析得出导致事故的主要因素为机组违规、管制差错与违规、气象和空域。实例分析结果表明,该模型与算法能适应航空事故数据特点,能客观描述几个主要因素与事故发生之间的关联关系。

Analysis for the influential factors leading to air vehicle collision accidents based on the rough sets and genetic algorithm

The paper is to propose a novel analysis model for the influential factors leading to the air collision accidents based on the relevant historic data for such accidents so as to disclose and categorize the patterns and the internal causes for the events.In doing so,we have made an exploration of such influential factors at the following two levels via the combination of the rough set theory and the genetic algorithm.The first level is to create the condition attributes,which can be defined by the SHELL model,including the factors like those of human interference,the dysfunctions of hardware,software and environment.In proceeding with the study,we have taken into account the causes leading to the relevant historic air-collision accidents given in the statistical report and the corresponding ratios of each cause,while consulting the analyses of the authoritative experts in this area and select the key sub-factors,that is,the condition attributes at the level.The second level is to rank the four major factors provided by the SHELL model based on the specifically chosen rough set theory,and assign respective weights to each factor on account of the disparity in the severity of different accidents,so as to determine the respective significance of the condition attributes according to the accident type it has resulted in.And,the third level is to analyze the air collision causes based on the rough set theory and genetic algorithm.This step can be realized through analyzing the three steps below.The first one is to synthesize the sub-factors whose condition attributes should be of primary importance.The second step is to maximize the dependency between the decision attributes and the reduction(C).And the third one is to maximize the number of such sub-factors whose condition attribute weights should be 1 included in a few top reasons.And,last of all,we have taken 20 air collision accidents and the potential accidents as the case study sample.Through simulation experiments done with the model we have proposed,the results we have gained show that the major factors of such air collision accidents can be illustrated as follows:the operation rule violation by the crew-members,the air-control violation errors,the abrupt disastrous climatic and weather hits,the aircraft its own structural problems,as well as the wrong decision-made by the pilots in accordance with the corresponding changing situations.As compared with the previous study methods,the given model can be qualified enough to deal with the problems,such as the incomplete accidents data and the subjective interferes of the pilots,the failure in identifying and determining the major influential factors in dealing with the potential relation among the factors concerned.