Nonlinear PLS with Neural Component Analysis Structure

Abstract

To handle the nonlinear feature in the industry process, this paper combines partial least squares (PLS) and neural component analysis (NCA), named as NCA-PLS. Different from NCA, the principal components are selected based on the correlation coefficient with KPI variables rather than the variance. As such, by redesigning the PCs extraction mechanism, NCA-PLS can successfully extract the KPI-related components from the process data and use them for process monitoring.