The Application of Machine Learning for Modeling the Adsorption of Heavy Metals on Biochars with Different Origins

Biochar, as the solid by-product of biomass, had been applied in multidisciplinary area such as wastewater treatment due to its microporous structure and abundant surface functional groups. The treatment efficiency of biochar for wastewater was significantly influenced by biochar characteristics and the environmental conditions. However, the diversity of biomass feedstock and uncertainty of produced biochar made the relationship complicated. Machine learning may be preferred to resolve the problem through teaching machine to find, recognize and extract relationships or rules based on statistical data. The adsorption of six heavy metals (lead, cadmium, nickel, arsenic, copper and zinc) on the fifty kinds of biochars were modeled using artificial neural network and random forest in the study, respectively, based on collected 433 datasets of adsorption experiments from the literatures. The regression models were trained and optimized to predict the adsorption efficiency according to biochar characteristics, metal sources, environmental conditions (e.g. temperature and pH in water and wastewater), and the initial concentration ratio of metals to biochars. The RF model showed better accuracy and predictive performance for adsorption efficiency with lower root mean squared error (0.038) and higher regression coefficient (R²=0.986) than ANN. The initial concentration ratio of metals to biochars was demonstrated as the most significant factor for adsorption efficiency, while the biochar properties was secondary. The CEC and pH_H2O of biochars accounted for over half of the contribution of biochar characteristics and the surface area was only accounted for 2.4% of adsorption. The accurate predicted ability of developed model could decrease experiment workload such as predicting the adsorption efficiency according to the biochar characteristics, environmental conditions, and the target metals. Moreover, the relative importance of each variable could provide a right direction for better removing heavy metals in the real water and wastewater.