基于改进动态核主元分析算法的化工过程故障检测

Fault Detection of Chemical Process Based on Improved Dynamic Kernel Principal Component Analysis Algorithm

为克服工业过程数据的非线性、动态性等限制,提出基于希尔伯特黄变换的动态核主元分析算法(HHT-DKPCA)。首先用标准化正常数据计算均值和标准差,之后对原始数据进行标准化。然后对协方差矩阵进行特征分解,将数据映射到低维空间。HHT变换去噪前对每个主元变量进行降维,然后将去噪后的数据重新映射到高维空间,并重新计算T^(2)和SPE。将HHT-DKPCA与PCA、KPCA、DKPCA、HHT-PCA在TE过程故障数据上的处理结果进行比较,结果表明,HHT-DKPCA具有更高的故障检测率。

For purpose of overcoming the nonlinear dynamics limitations of industrial process data,a Hilbert-Huang transform-based dynamic kernel principal component analysis algorithm(HHT-DKPCA)was proposed,in which,having the standardized normal data based to calculate both mean value and standard deviation and then the original data standardized,including having the covariance matrix decomposed and the data mapped to the low-dimensional space.For each principal component variable,having the dimension reduction per-formed before Hilbert-Huang transform de-noising,and then,having the data de-noised remapped to high-dimensional space and both T^(2) and SPE recalculated.Comparing the results of HHT-DKPCA with PCA,KPCA,DKPCA and HHT-PCA in TE process data set shows that,the HHT-DKPCA has higher fault detection rate.