华南理工大学学报(自然科学版) ›› 2007, Vol. 35 ›› Issue (1): 123-128.

• 计算机科学与技术 • 上一篇    下一篇

多变量统计分析中独立变量数目的判定方法

姚志湘1 蹇华丽2 刘焕彬1   

  1. 1.华南理工大学 制浆造纸工程国家重点实验室,广东 广州 510640; 2. 华南农业大学 食品学院,广东 广州 510642
  • 收稿日期:2005-12-20 出版日期:2007-01-25 发布日期:2007-01-25
  • 通信作者: 姚志湘(1968-),男,博士后,广西工学院副教授,主要从事化学计量学方面的研究。 E-mail:zxyao@21cn.com
  • 作者简介:姚志湘(1968-),男,博士后,广西工学院副教授,主要从事化学计量学方面的研究。
  • 基金资助:

    中国博士后科学基金资助项目(2005037583) ;广西自然科学基金资助项目(桂科基0448010)

Determination of Independent Variable Number in Multi -Variable Statistical Analysis

Yao Zhi-xiαngJian Hua liLiu Huan-bin1   

  1. 1. State Key Laboratory of Pulp and Paper Engineering , South China Univ. of Tech. , Guangzhou 510640 , Guangdong , China;2. College of Food Science , South China Agricultural Univ. , Guangzhou 510642 , Guangdong , China)
  • Received:2005-12-20 Online:2007-01-25 Published:2007-01-25
  • Contact: 姚志湘(1968-),男,博士后,广西工学院副教授,主要从事化学计量学方面的研究。 E-mail:zxyao@21cn.com
  • About author:姚志湘(1968-),男,博士后,广西工学院副教授,主要从事化学计量学方面的研究。
  • Supported by:

    中国博士后科学基金资助项目(2005037583) ;广西自然科学基金资助项目(桂科基0448010)

摘要: 数据空间的不均匀性是导致基于样本协方差矩阵特征值的多变量系统独立变量数目判定误差的原因.为此,文中提出了一种多变量统计分析中独立变量数目的判定方法.该方法首先对样本数据马氏距离化然后添加均匀白噪声以掩蔽数据空间的不均匀性,最后降序排列协方差矩阵的特征值,计算对数值,并求二阶差分,所得序列值分为与零值有微小差异的噪声贡献和与零值有明显差异的独立变量贡献,从而实现了拙立变量数目的准确判定.算例和实验结果均表明,该方法判定结果准确、清晰、稳定。

关键词: 多变量系统, 独立变量, 数目, 白噪声, 独立分量分析

Abstract:

In the multi-variable system , there generally exist errors in the determination of independent variable number based on covariance matrix eigenvalues because of the nonuniformity in data space. To solve this problem ,a method is proposed to determine the independent variable number in multi-variable statistical analysis. In this method , sample data are first changed in Mahalanobis distance , and are then added with uniform white noise to shield the nonuniformity in data space. Mter the processing with uniform white noise , by arraying the covariance matrix eigenvalues in descending order and calculating the eigenvalues logarithm , the second difference is obtained.So , a successive value is deduced , which is divided distinctly into two parts , one is related to the noise that is al-most equal to zero , and the other is related to the independent variables that are obviously different from zero , ma-king the independent variable number accurately determined. The simulation example and the experiments all re-veal that the proposed method is exact , distinct and steady.

Key words: multi-variable system, independent variable, number, white noise, independent component analysis