基于聚类HMM模型的QAR数据趋势分析

发布时间：2018-01-16 21:30

本文关键词：基于聚类HMM模型的QAR数据趋势分析　出处：《中国民航大学》2017年硕士论文　论文类型：学位论文

【摘要】：在航空飞行中,飞行安全始终是一个至关重要的问题。目前国内各大航空公司使用快速存取记录器(Quick Access Recorder,QAR)作为机载飞行数据记录设备来获取飞机日常运营的飞行数据。所记录的参数包括飞机的飞行参数、飞机发动机及主要部件的性能参数、机舱内外的环境参数。QAR监控作为一项科学而有效的技术手段,有利于保障飞行安全,提高运营效率,所得的监控结果为飞行技术检查、安全评估、安全事件调查和机务维修提供重要依据。为了更好地利用QAR数据为飞机的故障检测和预警决策提供有效地支持,面对大量的QAR数据,运用数据挖掘的方法对QAR数据进行分析,得到QAR数据运行趋势,通过趋势分析建立QAR数据的状态模型,从而识别机载设备所处状态进行故障诊断和预测,为安全飞行提供决策支持和保障。本文针对传统的QAR数据分析方法只关注异常点而忽略了QAR的状态趋势特征,提出了一种聚类隐马尔可夫模型(Hidden Markov Model,HMM),并将其应用在QAR数据上,通过对QAR数据中空中颠簸数据进行趋势分析并建立隐马尔可夫模型,有效预测了空中颠簸故障的程度,验证了该方法的有效性。本文的研究主要完成了以下几个方面的工作:1、研究了基于聚类的HMM模型并以空中颠簸故障为例,进行了QAR数据的状态趋势分析。针对HMM模型在QAR数据中的应用进行研究,鉴于飞机飞行过程中如果发生故障,则会表现为QAR数据的异常,会经历不同的状态,可建立HMM模型来描述飞机发生故障过程中的状态变化。2、针对QAR是时间序列这一特性,研究时间序列的挖掘方法,并着重研究了时间序列的分段表示方法,分析了多种线性分段表示方法应用在QAR数据的使用情况。另外,针对传统的用线段均值描述分段的不足,提出一种新的斜率和端点差的绝对值的描述方法,更有利于QAR数据聚类。3、研究QAR数据的聚类算法,分析并针对一种基于广度优先邻居搜索聚类算法在QAR数据上的局限性,对其进行改进,主要的改进方法即对数据进行分区,并对每个区内的数据进行线性分段表示,进行分区聚类,得到各分区聚类的结果,再进行聚类,即二次广度优先邻居搜索聚类。
[Abstract]:In aviation flight, flight safety is always a crucial issue. At present, domestic airlines use Quick Access Recorder. QARs are used as airborne flight data recording equipment to obtain flight data of aircraft daily operation. The recorded parameters include the flight parameters of the aircraft, the performance parameters of the aircraft engine and the main components. As a scientific and effective technical means, the environmental parameter. QAR monitoring inside and outside the engine room is helpful to ensure flight safety and improve operation efficiency. The result of monitoring is flight technical inspection and safety evaluation. Safety incident investigation and aircraft maintenance provide important basis. In order to better use QAR data to provide effective support for aircraft fault detection and early warning decision, a large number of QAR data are faced. Using the method of data mining to analyze the QAR data, the running trend of QAR data is obtained, and the state model of QAR data is established through the trend analysis. In order to identify the status of airborne equipment for fault diagnosis and prediction. In order to provide decision support and guarantee for safe flight, this paper focuses on the outliers of traditional QAR data analysis methods and neglects the state trend characteristics of QAR. A Hidden Markov Model (HMMM) is proposed and applied to QAR data. By analyzing the trend of air turbulence data in QAR data and establishing hidden Markov model, the degree of air turbulence fault is predicted effectively. The effectiveness of the proposed method is verified. In this paper, the following work has been done: 1, and the HMM model based on clustering has been studied. The state trend analysis of QAR data is carried out, and the application of HMM model in QAR data is studied. In view of the failure of aircraft during flight, it will appear as the exception of QAR data. The HMM model can be established to describe the state change in the process of aircraft failure. According to the characteristic that QAR is a time series, the mining method of time series is studied. In addition, the paper studies the segmented representation of time series, and analyzes the use of many linear piecewise representation methods in QAR data. In addition, aiming at the shortcomings of the traditional line segment mean method to describe segmentation. A new method of describing the absolute value of slope and endpoint difference is proposed, which is more favorable to clustering of QAR data. 3. The clustering algorithm of QAR data is studied. Aiming at the limitation of a breadth-first neighbor search clustering algorithm in QAR data, the paper improves it. The main improvement method is to partition the data. The data in each region are represented by linear segmentation, and the results of the clustering are obtained. Then the clustering is carried out, that is, the secondary breadth-first neighbor search clustering.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V241;TP311.13

【参考文献】