Prediction of the melt flow index using partial least squares and support vector regression in high-density polyethylene (HDPE) process

Abstract

In polyolefin processes the melt flow index (MFI) is the most important control variable indicating product quality. Because of the difficulty in the on-line measurement of MFI, a large number of MFI estimation and correlation methods have been proposed. In this work, mechanical predicting methods such as partial least squares (PLS) method and support vector regression (SVR) method are employed in contrast to conventional dynamic prediction schemes. Results of predictions are compared with other prediction results obtained from various dynamic prediction schemes to evaluate predicting performance. Hourly MFIs are predicted and compared with operation data for the high density polyethylene process involving frequent grade changes. We can see that PLS and SVR exhibit excellent predicting performance even for severe operating situations accompanying frequent grade changes.