Modeling and optimization of biogas production and COD removal from olive pomace: A hybrid approach using response surface methodology and artificial neural networks

This study explores a hybrid modeling and optimization for anaerobic digestion, combining both Response Surface Methodology (RSM) and Artificial Neural Networks (ANN) in a complementary approach. Olive pomace served as a model substrate to examine the influence of four key factors: dilution (ranging from 1/9 to 1/3), inoculum-to-substrate (I/S) ratio (from 1/6 to 1/2), pH (from 6 to 8), and substrate grinding (ground or unground). Using a Box-Behnken design, the optimal conditions identified through RSM were a dilution of 2/9, an I/S ratio of 1/2, pH of 8, and unground substrate. Under optimal conditions, biogas production reached 412.52 mL/gVS and Chemical Oxygen Demand (COD) removal was 64.89 %, with experimental validation showing prediction errors below 5 %. To improve prediction accuracy, two data augmentation strategies were tested based on the RSM dataset. Among them, a noise-based perturbation method implemented using a machine learning framework (ChatGPT) enabled highly accurate ANN training, reducing prediction errors to just 0.72 % for biogas and 0.16 % for COD removal. These findings highlight the potential of combining classical statistical tools with modern AI techniques for modeling complex biological processes with greater precision.