Multivariate online monitoring of a full-scale biological anaerobic filter process using kernel-based algorithms

Abstract

Multivariate statistical process control such as principal component analysis (PCA) and partial least squares (PLS) has been effectively utilized to analyze large databases accumulated in industrial plants in order to improve process performance and product quality. However, because both PCA and PLS are basically linear methods, nonlinearity in most chemical and biological processes is still a significant problem for their practical applications. Kernel-based algorithms are potentially very efficient for monitoring process disturbances and predicting key quality variables of nonlinear processes by mapping an original input space into a high-dimensional feature space. Nonlinear data structure in the original space is most likely to be linear at the high-dimensional feature space. Kernel-based methods essentially require only linear algebra, making them as simple as linear multivariate projection methods, and can handle a wide range of nonlinearities because of their ability to use different kernel functions. In this work, kernel-based algorithms such as kernel PCA (KPCA) and kernel PLS (KPLS) are applied to a full-scale biological anaerobic filter process treating high-strength organic wastewater in a petrochemical plant. KPCA is applied to detect process disturbances in real-time and KPLS to predict inferentially key process variables in the anaerobic filter process. The proposed kernel-based approaches could effectively capture the nonlinear relationship in the process variables and show far better performance in process monitoring and prediction of quality variables compared to the conventional PCA and PLS methods.