A novel machine learning framework for predicting biogas desulfurization breakthrough curves in a fixed bed adsorption column

Abstract

The corrosive and toxic hydrogen sulfide (H2S) in biogas is a pressing concern for researchers, spurring the development of predictive models for breakthrough curves in fixed-bed adsorption columns. This study utilizes Machine Learning (ML) to predict biochar H2S adsorption model parameters and establishes novel connections between contributing features and these parameters. The ML models could predict the process BC with an R2 score higher than 0.8 in 75 % of cases in the testing sets. A comprehensive features importance analysis reveals that Biochar's pH is the most influential factor, positively correlating with BC parameters and enabling more efficient H2S removal followed by H2S concentration, H/C molar ratio, and particle size. Understanding these relationships offers practical guidance for improving biogas purification, making this study valuable in addressing the H2S challenges and showcasing the potential of ML in predicting BC in fixed bed adsorption columns, providing a foundation for enhancing the adsorption process.