Soft Sensor Model Based on Kernel Slow Feature Analysis and Dynamic inner Principal Component Analysis

Abstract

It is important to establish relations between process variables and online estimation of quality variables for nonlinear and dynamic process. In the paper, a soft sensor model based on the kernel slow feature analysis and dynamic inner principal component analysis is proposed. First, KSFA method is used to reduce the dimension of the process data and extract nonlinear slow speed features, so as to solve the nonlinear relationship between driving force and input data. Second, DiPCA method was adopted to explicitly capture the potential variables with the most dynamic changes in the data, which improved the effectiveness of dynamic feature extraction and solved the feature extraction of linear data. At last, the PLS regression method is adopted to build the model between the principal and quality variable, and finally achieve a good prediction. The effectiveness of the proposed method was demonstrated by applying it to an industrial penicillin simulation.