Development of a New Regression Method Considering Predictability and Molecular Design of Amine Compounds for CO2 Separation and Recovery

Abstract

In carbon dioxide capture and storage (CCS), the chemical absorption method with amine compounds has been widely investigated as a method for capturing CO2. In this way, amine compounds with high performances of CO2 absorption and desorption are required for cost reduction of CO2 separation and recovery. One of the approaches to find amine compounds with high performances is molecular design with quantitative structure-property relationships (QSPR) models and structure generators. In this study, ensemble learning and genetic algorithm-based partial least squares (GAPLS), which is a variable selection method, were combined to construct predictive regression models. This method is named ensemble GAPLS (EGAPLS). In ensemble learning, prediction results from multi-models are integrated to give a better result than those of each single model. Moreover, considering the variance of the predicted values, it is possible to evaluate the reliability of the final prediction result. We constructed the QSPR models and evaluated the predictive accuracy of these models by cross-model validation (CMV) with the data of absorption rate and desorption capacity with tertiary amine compounds. The modeling results showed that the EGAPLS models had the highest predictive accuracy. The constructed EGAPLS models were applied to molecular design, and accordingly, promising chemical structures were obtained for CO2 separation and recovery.