Trickle bed reactors (TBRs) represent a promising strategy for ex-situ biomethanation and biomethane production with high methane content. The influence of operating parameters has been extensively studied across both laboratory and pilot scale TBRs to optimize critical outputs/key performance indicators such as methane purity, methane production rate (MPR), and hydrogen (H 2 ) transfer. In this study, machine learning (ML) techniques have evolved into an effective approach for studying the interactions among operating parameters and their effect on TBR performance, enabling more accurate prediction and optimization under both atmospheric and pressurized conditions. Ten ML regression models were employed to predict key parameters in the two identical TBRs operating at atmospheric and pressurized conditions. The results indicate that RF, XGBoost, and kNN outperform the other models, achieving R 2 > 0 . 95. Particle Swarm Optimization (PSO) was then applied to identify optimal operating conditions for the TBRs with the predicted models. Data from external pilot-plant TBRs were used to validate the ML regression models with the unseen data from the external pilot scale TBRs. Interestingly, while RF and XGBoost exhibit the best overall performance, kNN and ANN provide more accurate predictions for individual data points from the external pilot-scale TBRs. • Machine learning modelling workflow was created for biological methanation in Trickle Bed Reactors. • Certain models were able to confidently predict the methane purity at the exit of the reactor, the methane production rate, and the hydrogen transfer rate. • Explainability of the models was ascertained through SHapley Additive exPlanations (SHAP) values. • Optimized operating conditions were estimated through Particle Swarm Optimization of the ML models. • The model was experimentally validated with data coming from other similar biogas upgrade systems, solidifying its universal applicability.
Changizi et al. (Wed,) studied this question.