Engineering biochar from biomass pyrolysis for effective adsorption of heavy metal: An innovative machine learning approach

The application of biochar in heavy metal removal has attracted significant attention due to its characteristics. However, engineering biochar materials through trial-and-error methods is traditionally applied, but it is time-consuming and labor-intensive. Applying machine learning (ML) promises to substantially enhance the efficiency of engineering biochar with desired properties. Here, ML was employed for the first time to predict multiple properties of biochar that are critical to heavy metal adsorption, such as cation exchange capacity (CEC). Previous studies have focused on correlations between the CEC and application performance of biochar, but no study has reported the direct prediction of heavy metal adsorption capacity (qe) from biomass. Therefore, a biomass-production-application hybrid ML model (test R-2 0.996) was constructed by integrating the biomass-production-biochar-properties ML model (test R-2 0.941) and the biochar-properties-application (heavy metal adsorption) ML model (test R-2 0.960). This innovative hybrid model facilitated the screening of biomass feedstock and the optimization of pyrolysis conditions, using only elemental composition data of biomass. With nine biomass feedstocks in the lab, the hybrid ML model effectively provided the optimum solution for producing biochar with the highest q(e) (e.g., predicted q(e) of similar to 0.60 mmol/g for Cd2+). Finally, the experimental verification of the optimum solution showed that the adsorption capacities of the as-produced optimum biochar were comparable with the hybrid model-predicted ones (validation R-2 0.859), and the adsorption mechanism study echoed the dominant role of cation exchange, showing the great potential of such hybrid ML models to promote the production of designer biochar.