基于Copula函数和LU分解法的汽轮机数据特征选择

摘要

为了能够有效利用高维空间的数据并且解决冗余维数对最终结果的影响,本文提出了一种采用Copula函数和前向替换(LU)分解法的维数降低特征选择方法,即由Copula函数提供一个适合的相关性模型来比较多变量分布的数据,LU分解可以快速获得维数的线性组合相关系数,然后通过相关系数分析和消除不相关或冗余数据以及其他线性组合的变量,保持数据信息的完整性,这样可对多维数据进行特征选择.将本文方法与主成分分析(PCA)、简单主成分分析(SPCA)、奇异值分解(SVD)特征选择方法用于某厂2个月内汽轮机3种工况下大量数据的处理分析.结果表明,本文方法对识别的故障数据维数的降低数和识别效率较其他方法效果更好.%To make efficient use of data in high-dimensional space and eliminate the influence of redundant dimension on the final result,this paper presents a method of dimension reduction using Copula function and LU (forward substitution) decomposition method,in which the Copula function provides a suitable correlation model to compare the multivariate distribution data and the LU decomposition method can quickly obtain the linear combination of correlation coefficient.Then,through the correlation coefficient analysis and elimination of irrelevant or redundant data and other linear combinations of variables,the integrity of the data and information can be maintained,so feature selection can be carried out for the multi-dimensional data.Moreover,the above method and the principal component analysis (PCA),the simple principal component analysis (SPCA) as well as the singular value decomposition (SVD) feature selection method were applied to the processing and analysis of a large amount of data of three kinds of turbine operating conditions in a factory within 2 months.The results show that the method is better than the other methods in reducing the number of identified faulty data and enhancing the identifying efficiency.