Kinetic Parameter Estimation of the Polyethylene Process by Bayesian Optimization

The design and optimization of polymerization processes rely on accurate kinetic parameters of polymerization reactions since they have a significant impact on polymer properties. We develop a kinetic parameter estimation methodology for non-steady-state olefin polymerization processes. A dynamic model for olefin polymerization reaction kinetics based on the method of moments and instantaneous distribution method is established, which considers reaction temperature fluctuations and concentration variation during polymerization. Bayesian optimization (BO) algorithm is applied in the parameter estimation framework, where the kinetic model is treated as a black box function. The effectiveness of the method developed in this article was demonstrated through three cases, and the parameter estimation strategies used in other literature studies were compared. Additionally, we conduct a comparison of the fitting and extrapolation results obtained from diverse algorithms. Notably, BO emerges as a favorable alternative to deterministic algorithms, as it circumvents the challenges inherent in utilizing gradient-based optimization methods for complex polymerization dynamics models and is more suitable for scenarios with fast prediction in high-throughput experiments. Furthermore, using a dynamic kinetic model is crucial for kinetic parameter estimation in non-steady-state olefin polymerization, as it can reflect the dynamic changes with reaction temperature and concentrations.