Industrial & Engineering Chemistry Research, Vol.59, No.13, 6021-6032, 2020
Nonlinear Process Quality Prediction Using Wavelet Denoising OSC-SVM-PLS
Modern industrial processes are becoming more and more complex and contain linear and nonlinear parts as well as some unknown disturbances, which pose difficulties for product quality prediction. The existing methods for industrial process quality prediction such as partial least-squares (PLS) and kernel partial least-squares (KPLS) sometimes fail to address these issues. In this Article, a new hybrid denoising-linear-nonlinear quality prediction method is proposed. After collecting the industrial process data, the soft threshold wavelet denoising method is first used to remove the noise contained in the process data. Then, the orthogonal signal correction (OSC) is used. The OSC method replaces the traditional data preprocessing method to remove information related to the quality data in the process data, which generally reduces the number of principal components used in the modeling. A support vector machine (SVM) approach is then introduced to give the model the ability to process nonlinear data. A numerical simulation example and an example of the penicillin fermentation process based on actual chemical processes were used to verify the prediction performance of the proposed method. The results show that the proposed method outperforms traditional linear PLS and nonlinear KPLS methods in terms of prediction ability.