Quadratic Mutual Information based Regression for Prediction of Quality Variable in Batch Process

Abstract

Abstract Quality prediction is of great importance for batch processes. Predicting quality variable is a challenging task because of various factors such as strong nonlinearity and non-Gaussian exist in batch data. A quadratic mutual information based regression method is proposed to handle the problem. The proposed method takes into account higher order statistics that reveal the non-linear dependencies between the process variables and important quality variables. Furthermore, the proposed method is implemented without the hypothesis of Gaussian distribution of the dataset as in MPLS. The effectiveness of the QMIR method is illustrated by a dataset of industrial Escherichia coli fermentation process, compared with MPLS.