Virtual Database Construction and Machine-Learning-Assisted High-Throughput Evaluation of Amorphous Porous Carbon Materials as Iodine Sorbents

We present a comprehensive approach to enable the high-throughput screening and analysis of amorphous porous carbon (APC) materials as effective I 2 sorbents for the nuclear industry. A diverse virtual database of 19,599 APC models was established from scratch through liquid quenching molecular dynamics simulations. Large-scale grand canonical Monte Carlo simulation at a series of I 2 concentrations was carried out for sampled APCs to generate an array of I 2 adsorption capacities. Machine learning and SHapley Additive exPlanations (SHAP) analysis were employed to investigate the impact of various extracted (structural and chemical) features of the APC materials on their respective I 2 adsorption behavior, revealing influential factors (surface area, pore size ranges, etc.) for APC development that varied with I 2 concentrations. This work attempts to provide both fundamental databases and research frameworks to accelerate the development and enhance the understanding of APC materials.