Supervised machine learning for prediction of zirconocene-catalyzed α-olefin polymerization

A new approach is demonstrated in which an Artificial Neural Network (ANN) was trained with first-principles data to predict the chain length, polydispersity (Đ) and adiabatic temperature for a zirconocene-catalyzed polymerization reaction. The ANN-generated data shows good agreement with the theoretical results, with an overall R² of 0.9987. Using its significantly enhanced computational speed, the ANN was used to analyze the reaction space, providing insights into trends seen in molecular weight and Đ with various combinations of kinetic parameters, particularly pointing out regions of desirable and undesirable operation. The network was trained in reverse and used to generate reaction rate constants from chain length and Đ, enabling a new form of kinetic deduction for polymerization reactions. This training was used to derive potential rate constants for different catalysts reported in the literature. Overall, this data indicates that ANNs are a plausible tool for analyzing data from complex metallocene-catalyzed olefin polymerizations.