壽命分布中的異常值檢驗(yàn)方法研究.pdf

1、華中科技大學(xué)碩士學(xué)位論文壽命分布中的異常值檢驗(yàn)方法研究姓名：肖浩申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：概率論與數(shù)理統(tǒng)計(jì)指導(dǎo)教師：劉次華20041026AbstractIn reliability analyse, the life of many types of products, such as electronic de-vices and components, are usually described by exponential dis

2、tribution, Weibulldistribution, extreme value distribution, lognormal distribution, etc. But becauseof various kinds of reasons, a few unusual data may enter into the sample of lifeexperiment, as certainly will impose ne

3、gative influence on our study, even will leadto mistaken conclusion. So it becomes a very important link in data processingto identify and reject outliers in sample. This paper goes on deep research on theoutlier tests f

4、or several common life distributions. Based on different ideas, severaleffective methods are proposed from different angles.Exponential distribution’s no-memory property has convenient applications inMathematics. Therefo

5、re, People are glad to assume that sample has exponentialdistribution. Since Epstein and Sobel finished their famous work of statistical anal-ysis for life data following exponential distribution, in the beginning of 195

6、0s, expo-nential distribution began to be put into practice extensively. First of all, by meansof Monte Carlo and Bayesian Methods, this paper carry on search on the problemof outliers in an Exponential Sample, and prese

7、nt a new Bayesian test for outliersbased on the noninformative prior distribution. The method adopts a new idea: testonly uses middle data, data located in both ends are discarded, then test goes on se-quentially from mi

8、ddle to bottom. Simulation result indicates it is relatively robust,can overcome the masking effect effectively. It is Weibull distribution that usedextensively most in life distributions. The extreme value distribution

9、has a simplelogarithm relation with Weibull distribution. This paper is based on some existingtheoretical foundation, uses many mathematics skills, moves the same method ofinspecting outliers to these two kinds of distri

10、butions, and derives the approximatedistribution functions of test statistics.Secondly, various kinds of test methods of normal distribution with fine efficiencyhave already been proposed. How to exploit this sum of exis