Bayesian optimization of HDPE copolymerization process based on polymer product-process integration

Abstract

A framework for finding an optimal operating point of HDPE with the target properties is proposed. A bimodal high-density polyethylene (HDPE) copolymerization process is established to calculate molecular weights, dispersity, and predict properties. The process incorporates a probability-based description method with reactivity ratio to describe the content of monomer A/B and sequence distribution. A properties prediction model is established to predict glass transition temperature, melt temperature, melt index, and density. Bayesian optimization algorithm is utilized to optimize the process. This algorithm combines Gaussian process regression and an expectation improvement model to determine the optimal mass flow of feed and temperature of reactors. The objective function is modified using a penalty function to incorporate property constraints. The optimization goal is to maximize profits by identifying the optimal composition of feed and reactor temperature. Optimized bimodal HDPE copolymerization process leads to a significant increase in profit, with a 21.8 % improvement compared to the original process. The effect of melt index, is considered in determining the optimal operation range and maximizing profit. This provides valuable guidance for producing new grades of HDPE.