Comparative evaluation of mechanistic models for biogas production in DIET-enhanced anaerobic digestion

Statistical modelling serves as a valuable tool for predicting biogas production during the anaerobic digestion (AD) of organic substrates. This study offers a comparative analysis of three widely used models, specifically, the first-order kinetic model, the modified Gompertz model, and the Chen and Hashimoto (CH) model. A novel dataset from co-AD of sewage sludge with wheat husk and enhanced by direct interspecies electron transfer (DIET) was utilized for modelling. The fundamentals of machine learning (ML), such as model calibration, selection, sensitivity analysis, testing, and evaluation, were carried out. Model calibration employed optimization algorithms (L-BFGS-B, BFGS) to minimize the sum of squared errors using training data (days 1-30). Model selection based on root mean squared error from validation data (days 31-40) identified the CH model as optimal, with 40-67 % lower root mean square error compared to alternatives. Sensitivity analysis using Morris (local) and Sobol' (global) methods revealed that CH constant (K_CH) exhibited the lowest primary influence on model output (μ∗ = 52.6, first-order Sobol index = 0.02), while ultimate biogas potential (G₀) showed highest sensitivity (μ∗ = 191.5, first-order Sobol index = 0.84). Model evaluation on test data (days 41-51) demonstrated satisfactory predictive accuracy with R² and slopes approaching unity across all samples. The CH model shows specific applicability to the DIET-enhanced co-AD system. Broader conclusions regarding the applicability of the model across diverse AD processes would benefit from further ML-based investigations.