EventsThe 10th International Electronic Conference on Water Sciences
Published
This submission belongs to the session S8. Wastewater Treatment and Reuse of the event The 10th International Electronic Conference on Water Sciences
Published date
19 Nov, 2026
Academic Editor
author-avatarChristos S. Akratos
Citation
pouria paydarnik, Mohammad Salimi Delshad, Hossein Bonakdari, Hybrid ADM1-Machine Learning Framework for Optimizing In-Situ Hydrogen Biomethanation and Predicting Methane Production in Anaerobic Digestion, in Proceedings of The 10th International Electronic Conference on Water Sciences, 24 November–27 November 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

Hybrid ADM1-Machine Learning Framework for Optimizing In-Situ Hydrogen Biomethanation and Predicting Methane Production in Anaerobic Digestion

Mohammad Salimi Delshad 1
image
1. Department of Civil Engineering, University of Ottawa, Ottawa, Canada
Abstract

Anaerobic digestion (AD) is a key technology for sustainable wastewater treatment and resource recovery, yet raw biogas typically contains 50-65% methane (CH4), limiting its utility as a fuel and as a grid-injectable renewable gas. In-situ hydrogen biomethanation (HBM) the direct injection of hydrogen (H2) into operating digesters to drive hydrogenotrophic conversion of CO2 into additional CH4 offers a route for biogas upgrading.

The standard ADM1 model was extended with: An external H2 injection term in the gas-phase balance; mass transfer coefficients (kLa) for H2, CO2, and CH4 via diffusivity-based based on (Pauss et al., 1990); and an inorganic carbon limitation factor derived from (Bensmann et al., 2014). Dynamic simulations were conducted across a matrix of kLa and H2 injection scenarios. The resulting simulation dataset was used to train, test, and compare four machine learning (ML) algorithms: Support Vector Regression (SVR), Random Forest (RF), Extreme Learning Machine (ELM), and Evolutionary Polynomial Regression (EPR).

Simulations revealed that H2 conversion efficiency ranged from 57-61% under low mass-transfer conditions to above 96% under high-transfer conditions, with outlet CH4 content reaching up to 84%. Among the ML models, SVR achieved the highest predictive accuracy (R2 = 0.997), followed by RF (R2 = 0.988). EPR with an R2 = 0.930, an equation to calculate methane production by inputs, which is suitable for industrial estimation without model infrastructure. The ELM model failed to predict CH4 production with an R2 = 72.4.

The findings establish that improving gas-liquid mass transfer should precede any increase in H2 supply, as additional H2 delivered under weak transfer conditions reduces conversion efficiency and gas quality rather than improving them. The combined ADM1-ML framework provides both mechanistic process insight and practical prediction tools for AD optimization. The EPR-derived formula further offers a computational tool for CH4 estimation at industrial-scale plants, while SVR remains the strongest choice where predictive precision is the primary objective.

Keywords
Anaerobic digestion
In-situ biomethanation
ADM1
Machine learning
Biogas upgrading
Mass transfer
Methane prediction
Evaluation of Bacteriophage-Immobilized Biopolymeric Microparticles for Metal Bioremediation in Water Samples from Different Regions of Brazil