Abstract: In face of the outliers in comprehensive evaluation, this paper analyses three questions: are there outliers in original data if there are outliers, how to recognize them And how to design dimensionless method for all data included in outliers For the judgment and recognition of outliers, we take “median” as a reference. Based on this, the outliers can be identified by comparing the distance of the former endpoint to the median and that of the later endpoint to the median. To the dimensionless question, based on the frequently-used linear scale transformation (max-min) method, we provide a piecewise dimensionless method by allocating dimensionless value interval to outliers and non-outliers respectively. At last, the validity of this research is illustrated by the comparison with the results of available literature. And it is found the methods in this paper can not only recognize outliers moderately, but also promote the proportionality of dimensionless results.

Key words: comprehensive evaluation, dimensionless process, outlier, linear scale transformation(max-min)method, piecewise dimensionless method

摘要： 针对综合评价中的异常值现象,讨论了原始数据中是否存在异常值、若存在异常值该如何识别异常值以及对含有异常值的评价数据如何进行无量纲化处理三个问题。关于异常值的判断与识别,给出了以“中位数”为参考点,通过比较排序后两端数据偏离中位数的距离的处理思路。对含有异常值的评价数据的无量纲化处理问题,基于常用的“极值处理法”,通过分别指定异常值和非异常值无量纲化取值区间的方式,提出了一种分段的无量纲化处理方法。最后,通过与已有文献异常值识别及无量纲化处理结果的对比分析,验证了本文方法的有效性,发现本文给出的方法能够实现对异常值的适度筛选,且能够提升无量纲化数据分布均衡性。

关键词: 综合评价, 无量纲化处理, 异常值, 极值处理法, 分段无量纲化方法