Accurately predicting the performance of MOF-based mixed matrix membranes for CO2 removal using a novel optimized extreme learning machine by BAT algorithm

Abstract

The increasing number of mixed matrix membranes (MMMs) based on metal organic frameworks (MOFs) for carbon capture has created a demand for accurate and swift evaluation of their separation performance. Machine learning (ML) has emerged as a valuable tool for this purpose, providing an efficient approach for screening these materials and accelerating their practical application. In this study, we developed and optimized a novel and reliable hybrid machine learning paradigm based on the extreme learning machines (ELM) method using the BAT algorithm optimization. In order to predict the performance of MMMs, including gas permeability and selectivity parameters, nine machine learning models were developed by incorporating descriptors and fingerprints for polymer featurization, physical and structural features of MOFs as well as operating conditions. The impact of input features on MMMs performance was also explored using RReliefF analysis. The study found that the performance of the hybrid ELM-based algorithms was significantly improved by using the BAT algorithm. Furthermore, the RReliefF analysis revealed that the cage size of the MOF and the type of polymer matrix used are the most significant parameters in forecasting the permeability of MOF-based MMMs, while the loading amount and pressure were identified as critical determinants of selectivity. Overall, these findings contribute to the development of more efficient and accurate methods for evaluating MMMs, which are crucial for carbon capture applications.