基于增量模糊关联分类方法的飞行品质监控研究

Flight Operations Quality Assurance (FOQA) collects QAR(Quick Access Recorder) data generated by an aircraft moving through the air from one point to another, discovers various exceedance events from QAR data by pre-setted monitoring item and criteria. It also can adopt relevant measures in response to exceedance event, and avoid the occurrence of accident. However, Current practice of FOQA exists some shortcomings: It need domain expert to recognise exceedance event according to experience,which is a heavy workload; it can not dectect an abnormality which is not defined in monitoring item and criteria; and it does not analysis QAR(Quick Access Recorder) data deeply.In order to address these challenges, We propose a novel approach for building Incremental Fuzzy Associative Classifier(IFAC) on the basis of the characteristic of QAR data's magnanimity and high-dimension. The proposal will offer a new mode for monitoring and diagnosing exceedance event through the using of IFAC. We also will apply IFAC to detect fly abnormalities which are not defined in monitoring item and criteria. These abnormalities will provide the basis for supplementing monitoring item and criteria. Our work will enhence automaticity and accuracy of FOQA, realize more effective and proactive aviation safety management. The novelty of this project is that it, for the first time, presents the new method of building IFAC, and its application to FOQA.

飞行品质监控是指收集飞机运行过程中的QAR(Quick Access Recorder)数据，依据事先设定的监控项目和标准从中发现各种超限事件，采取针对性的改进措施，避免事故发生。但是，现行工作方式存在如下不足：需要领域工程师依据经验识别过滤超限事件，工作量大；无法检测到已有监控项目和标准集合之外的飞行异常；没有对QAR数据背后深层次问题进行系统分析。为此，本项目根据QAR数据集维数高、数据量大的特点，在前期研究的基础上提出构建增量模糊关联分类器的新方法，为监测、诊断超限事件提供一种新的方式；并利用其检测已有监控项目和标准集合之外的飞行异常，为补充完善监控项目和标准提供依据。本项目的研究将有助于提高民航飞行品质监控工作的自动性和准确性、有助于实现更加有效的主动航空安全管理。其创新性之处是首次提出构建增量模糊关联分类器的新方法，并将其应用于民航飞行品质监控工作。

飞行品质监控利用机载QAR(Quick Access Recorder)记录的飞行数据，依据事先设定的监控项目和标准发现超限事件。本项目针对QAR数据集多维非同质、数据量大的特点和现行飞行品质监控工作方式中的不足，提出了构建增量模糊关联分类器及其在民机超限事件检测中应用的研究目标。围绕研究目标，项目主要研究内容包括：基于SPFCM聚类和CHC遗传模型的增量学习三角形隶属函数方法；基于H-mine算法的增量模糊频繁项挖掘方法；基于eVQ聚类的增量模糊关联分类器构建方法；基于模糊关联分类器的超限事件检测方法；改进小波聚类算法及其在飞行异常检测中的应用。项目的研究从隶属函数学习、模糊关联分类规则生成、分类规则库更新方面提高了模糊关联分类器在动态数据集的构建效率，为检测民机飞行超限事件提供了新的方式，有助于实现更加有效的主动航空安全管理。