Predicting adsorption and separation performance indicators of Xe/Kr in metal-organic frameworks via a precursor-based neural network model

Abstract

Compared with traditional product-based prediction model, predicting adsorption and separation performance (nSP and nAPI) of Xe/Kr mixture in metal–organic frameworks (MOFs) using precursor-based prediction model can obviously enhance efficiency and save cost. In this study, nSP and nAPI of MOFs in M−IRMOF analogues database (MIAD) were calculated using Grand Canonical Monte Carlo (GCMC), Ideal Adsorbed Solution Theory (IAST). Combined with density functional theory and high-throughput screening, two novel artificial neural network models (BPNN-nSP and BPNN-nAPI) with low errors and high regression coefficients were built to predict nSP and nAPI of Xe/Kr in MOFs just using physical parameters of MOFs precursors (organic linker and metal center). In addition, tested by data of experimental MOFs, RMSE values of BPNN-nSP and BPNN-nAPI models are only 0.013 and 0.08. Therefore, these two models are reliable and generalizable for predicting nSP and nAPI, which contribute to development of nuclear power technology and satisfy industrial requirements.