论文部分内容阅读
针对化工过程数据的多尺度性和非线性特性,提出了一种多尺度核主元分析方法(MSKPCA)监控过程的运行状态。使用小波变换在不同尺度下分解测量信号．然后借助于核函数对分解后的数据进行非线性变换,在变换后的线性空间中用主元分析(PCA)提取过程数据的主要特征,构造监控统计量T2和Q来检测故障。在此基础上,提出了一种贡献图方法．计算过程变量对故障的贡献量,用于故障变量的分离。在TE过程上的监控结果表明,MSKPCA可以比PCA和动态PCA更迅速地检测到过程故障,贡献图方法能够正确地分离故障变量。
Aiming at the multi-scale and non-linear characteristics of chemical process data, a multi-scale kernel principal component analysis (MSKPCA) monitoring process is proposed. Using wavelet transform to decompose the measurement signal at different scales. Then, the decomposed data is transformed nonlinearly by means of kernel function. Principal component analysis (PCA) is used to extract the main features of the process data in the transformed linear space, and the monitoring statistics T2 and Q are constructed to detect the fault. On this basis, a contribution graph method is proposed. Calculate the contribution of process variables to the fault for the separation of fault variables. The monitoring results on the TE process show that MSKPCA can detect process faults more quickly than PCA and dynamic PCA, and the contribution graph method can correctly separate the fault variables.