两参数Weibull分布多个异常数据的检验

Detection Of Many Outliers in the Two-Parameter Weibull Distribution

针对两参数Weibull分布的异常大数据给出了一个新的检验方法.该方法利用样本分位数与两参数Weibull分布有密切联系的I型极值分布尺度参数的同变估计来构造新的检验统计量——Z~*型统计量.讨论了该方法的检验效率,表明利用Z~*型统计量作检验可以避免Masking效应,如果样本中存在多个异常大数据,反复利用该方法也可以检验出这些异常数据.

A method of detection for the upper outliers in the two-parameter Weibull distribution is discussed. The sample fractile and equivariant estimators of scale parameter of the extreme are used to construct the new test statistic——Z~* type statistic.The discussion of testing efficiency indicates the test statistic can avoid the masking effect.If there are many upper outliers in the sample,we can detect them by repeated use of the method.