Melt index prediction by support vector regression

Abstract

Melt index is considered one of the most important variables in determining chemical product quality and thus reliable prediction of melt index (MI) is essential in practical propylene polymerization processes. In this paper, a fuzzy support vector regression (FSVR) based model for propylene polymerization process is developed to predict the MI of polypropylene from other easily measured process variables. Support vector data description (SVDD) is introduced in this model as a novel fuzzy membership function and to reducing the effect of outliers and noises. A detailed comparison between the standard SVR and SVDD-FSVR models is carried out on a real plant. The research results have confirmed the effectiveness of the presented method.