Anaerobic digestion (AD) is a complex biochemical process influenced by nonlinear interactions among feedstock characteristics, operational parameters, and reactor dynamics, making experimental optimization expensive, time-consuming, and difficult to scale. This paper presents a novel AI-driven surrogate simulation framework designed to rapidly approximate AD behavior and predict methane yield with high computational efficiency. The framework integrates Deep Neural Networks (DNN), Gaussian Process Regression (GPR), Random Forest Surrogate (RF) based surrogate modeling, and Support Vector Regression (SVR) to learn process-response relationships from a structured AD dataset consisting of physicochemical features, operating conditions, and experimentally validated methane performance indicators. Surrogate models were trained to emulate reactor behaviour, quantify prediction uncertainty, and generate response surfaces for virtual experimentation. Results demonstrate that DNN and GPR achieve superior surrogate fidelity, with GPR additionally providing robust uncertainty bands, while RF and SVR offer efficient approximations with faster computational speeds. The proposed surrogate framework enables rapid what-if analysis, parameter sensitivity exploration, and real-time simulation of methane performance without requiring laboratory-scale digestion runs. This work establishes a scalable foundation for intelligent AD optimization, virtual biogas plant prototyping, and AI-enabled decision support systems for sustainable biomass-to-energy conversion.
Chatterjee et al. (Wed,) studied this question.