Machine learning approaches to rediscovery and optimization of hydrogen storage on porous bio-derived carbon

Abstract

Three Machine learning (ML) models were constructed and tested to model the hydrogen adsorption of activated carbons (ACs). Although many efforts have been accomplished, the details of micro structure-efficiency of ACs still remains uncovered. Data-based modeling contains the crucial properties of ACs’ structural characteristics such as micropore surface area, pore volume, and pore-size distribution are utilized as inputs for hydrogen adsorption. The proposed models demonstrated their high feasibility to predict the hydrogen uptake with the RMSEs ranged from 0.06 to 0.19. Among them, the Support vector machine (SVM) exhibited the highest accuracy for prediction of hydrogen adsorption on ACs. The sensitivity analysis also showed the importance of micropore surface area and pore widening on hydrogen uptake. Subsequently, the effects of microstructure properties on hydrogen storage were optimized, employing the SVM-based genetic algorithm (GA) technique. By using the optimized structural properties obtained from GA, the hydrogen uptake increased by 2.5 wt%. The precise tuning of textural features significantly revealed the importance of microstructural properties in bio-derived ACs for H2 uptake. This can be a helpful guide for fabrication procedures of porous carbons for H2 adsorption applications.