Hybrid Deep Learning Optimization of Catalytic Parameters in Anaerobic Digestion for Enhanced Biogas Production

Optimizing catalytic parameters in anaerobic digestion (AD) for biogas production is inherently complex due to the interdependency of biological, chemical, and physical variables, and conventional optimization approaches often fail to capture the nonlinear system dynamics involved. This study develops a hybrid modeling framework combining a Deep Belief Network (DBN) with the Osprey Optimization Algorithm (OOA) selected for its superior convergence behavior and resistance to local optima entrapment compared to Genetic Algorithms and Particle Swarm Optimization in high-dimensional process spaces to predict and optimize biogas production rates in catalytically assisted AD systems. Key input variables encompass catalyst type, catalyst concentration, pH, oxidation-reduction potential (ORP), volatile solids (VS) removal rate, organic loading rate (OLR), and hydraulic retention time (HRT). The dataset is assembled from peer-reviewed experimental studies retrieved from Scopus and Web of Science, with preprocessing steps including protocol normalization, outlier removal, and feature scaling applied to mitigate variability arising from heterogeneous experimental conditions. The DBN-OOA model is trained with k-fold cross-validation to guard against overfitting and benchmarked against standalone ANN and ANFIS models using R², RMSE, and MAE metrics, achieving predictive accuracy of R² > 0.96 on the held-out test set. Optimization results indicate that coordinated adjustment of catalyst loading and OLR relative to a baseline condition of no catalyst addition at a fixed OLR of 2.0 g VS/L·day yields up to 22% improvement in volumetric biogas production. These findings demonstrate that the proposed framework offers an effective, computationally accessible tool for multi-parameter optimization of catalytic AD processes, with potential for integration into digital monitoring systems for biogas plant management.