Prediction of Reactivity Ratios in Free Radical Copolymerization from Monomer Resonance–Polarity (Q–e) Parameters: Genetic Programming-Based Models

Abstract

Abstract The principal deficiency of the widely utilized Alfrey–Price (AP) scheme for computing reactivity ratios in the widely used free radical copolymerization is that it ignores important factors, such as the steric effects. This often leads to inaccurate reactivity ratio predictions by AP model. Accordingly, in this study, exclusively data-driven, Q–e parameter-based new models have been developed for the reactivity ratio prediction in free radical copolymerization. In the model development, a novel artificial intelligence formalism known as “genetic programming (GP)” that performs symbolic regression has been employed. The GP-based models possess a different functional form than AP model. Further, parameters of GP-based models were fine-tuned using Levenberg–Marquardt (LM) nonlinear regression method. A comparison of AP, GP and GP-LM as well as artificial neural network (ANN)-based models indicates that GP and GP-LM models exhibit superior reactivity ratio prediction accuracy and generalization performance (with correlation coefficient magnitudes close to or greater than 0.9) when compared with AP and ANN models. The GP-based reactivity ratio prediction models developed here due to their higher accuracy and generalization capability have the potential of replacing the widely used AP models.