Principal component analysis (PCA) is widely used in industrial process monitoring.PCA has drawbacks when dealing with non-Gaussian process data and the covariance matrix it uses can only evaluate the linear relationship among variables and ignore the nonlinear dependence information.To solve this shortcoming,a novel process monitoring method based on logarithmic transformation and maximal information coefficient-PCA is proposed.Firstly,logarithmic transformation is used to transform the process data to improve the data distribution in a certain degree.Then,the covariance matrix can be replaced by the MIC matrix which can measure the non-linear correlation between the variables,so as to improve the monitoring effect on nonlinear and non-Gaussian process.Finally,the feasibility and effectiveness of the proposed method are verified by the Tennessee Eastman (TE) process simulation.%主元分析(principal component analysis,PCA)被广泛应用于工业生产过程监测.PCA假设数据服从高斯分布且协方差矩阵仅能评估变量间的线性关系,无法衡量变量间非线性依赖程度.基于此,提出了一种基于对数变换和最大信息系数(maximal information coefficient,MIC)PCA的过程监测方法.首先,应用对数变换对过程数据进行变换,在一定程度上改善数据分布.然后,采用可以度量变量间的非线性相关性的MIC矩阵替换协方差矩阵,从而改善对非线性非高斯过程的监测效果.最后通过在田纳西-伊斯曼过程(tennessee eastman process,TE)仿真研究验证了该方法的可行性和有效性.
展开▼