Applied machine learning to the determination of biochar hydrogen sulfide adsorption capacity

Biogas desulfurization using biochar is complex and highly nonlinear, affected by various variables and their interactions. Moreover, achieving maximum adsorption capacity and investigating the simultaneous effects of different variables on the efficiency of the adsorption process is challenging. In this study, machine learning algorithms were successfully applied to predict the biochar hydrogen sulfide adsorption capacity in biogas purification. Three supervised machine learning models were devised and evaluated in three-step model development to determine biochars' hydrogen sulfide adsorption capacity. In each model, a feature selection procedure was used in combination with feature important analysis to extract the most influential parameters on the hydrogen sulfide adsorption capacity and improve the total accuracy of models. The exhaustive feature selection method was used to find the best subset of features in each machine learning algorithm. The models used twenty features as input variables and were trained to learn complex relationships between these variables and the target variable. Based on features important and Shapley Additive Explanation analysis, the biochar surface's pH and the feedstock H/C molar ratio were among the most influential parameters in the adsorption process. The gradient boosting regression model was the most accurate prediction model reaching R 2 scores of 0.998, 0.91, and 0.81 in the training, testing, and fivefold cross-validation sets, respectively. Overall, the study demonstrates the significance of machine learning in predicting and optimizing the biochar Hydrogen Sulfide adsorption process, which can be an asset in selecting appropriate biochar for removing hydrogen sulfide from biogas streams.