Multivariate soft sensor for product monitoring in the debutanizer column with deep learning

Abstract

Soft sensors have been proposed extensively for predicting ill-to-measure variables in industrial processes. In this study, we developed a multivariate soft sensor for debutanizer columns. A soft sensor was proposed to replace the chromatograph-based butane content from the debutanizer column. Recently, deep learning methods have been implemented for better feature representation of complex systems. We developed an LSTM-based multivariate soft sensor that can better represent the dynamics of a debutanizer column system. Our results show that the univariate LSTM soft sensor performs better than previously proposed methods.