EVALUATION OF FOOD WASTE AND SEWAGE SLUDGE ANAEROBIC CO-DIGESTION: KINETIC MODELING, META-ANALYSIS, AND LONG-TERM OPERATION WITH MICROAERATION

Abstract

Machine learning modeling has recently gained attention in bioprocess engineering research for its precision prediction, optimization, and failure detection. However, due to its “black box” nature, interpretability approaches are needed to integrate to improve their understandability. In our study, the conventional approach of bioprocess assessment (i.e., kinetic modeling, a systematic review with meta-analysis, and long-term continuous operation) assisted with statistical analysis, and machine learning modeling was employed. In the batch experiments, higher food waste content resulted in higher specific methane yield (SMY), indicating higher biodegradability during co-digestion. The superimposed model with the first-order kinetic and modified Gompertz structure exhibited better accuracy among others in co-digestion. Meta-analysis reveals synergistic interactions of lignocellulosic biomass with animal manures and food waste with animal manures and lignocellulosic biomass (relative synergistic index, RSI > 1.20). Based on correlation analysis, multilinear regression, and tree-based regression, temperature was identified as a key parameter to improve methane yield in co-digestion of lignocellulosic biomass and fats, oils, and grease. However, food waste content is more important in food waste co-digestion. Long-term anaerobic co-digestion reaffirms that higher food waste content resulted in higher methane yield due to its rapid biodegradability and reveals the interactions of microaeration on hydrogenotrophic methanogenesis. The time-series model, specifically the trained nonlinear autoregressive network with exogenous inputs (NARX), also showed promising application on continuous systems with R2 = 0.8–0.9.