Prediction of Plasticizer Property Based on an Improved Genetic Algorithm

Different plasticizers have obvious differences in plasticizing properties. As one of the important indicators for evaluating plasticization performance, the substitution factor (SF) has great significance for product cost accounting. In this research, a genetic algorithm with variable mutation probability was developed to screen the key molecular descriptors of plasticizers that are highly correlated with the SF, and a SF prediction model was established based on these filtered molecular descriptors. The results show that the improved genetic algorithm greatly improved the prediction accuracy in different regression models. The coefficient of determination (R-2) for the test set and the cross-validation both reached 0.92, which is at least 0.15 higher than the R-2 of the unimproved genetic algorithm. From the results of the selected descriptors, most of the descriptors focused on describing the branching of the molecule, which is consistent with the view that the branching chain plays an important role in the plasticization process. As the first study to establish the relationship between plasticizer SF and plasticizer molecular structure, this work provides a basis for subsequent plasticizer performance and evaluation system modeling.

Automated summary

In this study, a genetic algorithm with variable mutation probability was used to screen key molecular descriptors for predicting substitution factor. The improved genetic algorithm significantly enhanced the prediction accuracy. The selected descriptors mainly focused on describing the branching of the molecule, which is consistent with the importance of branching chains in the plasticization process.