Estimation of ethylene/1-butene copolymerization conditions using the autoencoder model

Abstract

Abstract Chain microstructures of ethylene/1-butene copolymers produced at specific polymerization conditions can be calculated using the copolymerization kinetic model. However, this kinetic model cannot be solved inversely to estimate polymerization conditions from desired microstructures. In this work, the autoencoder (AE) model, the machine learning techniques based on artificial neural network (ANN) concept, was developed to help estimate polymerization conditions to produce polymers with desired microstructures (e.g., molecular weight distribution (MWD), chemical composition distribution (CCD), and number and weight average molecular weight). Two models were developed in this work: one with only microstructural distributions (MWD and CCD) and the other with additional information on average microstructures and polymer yield. The results showed that the proposed AE models can adequately estimate polymerization conditions from desired microstructures with acceptable mean square error (MSE). More specific microstructures lead to better estimation of polymerization conditions with lower MSE.