Optimization of methane production in anaerobic digestion: A hybrid approach using experimental analysis, process simulation and machine learning

Anaerobic digestion offers a sustainable route for converting food waste into renewable biogas, but its efficiency requires careful optimization of operational parameters. This study presents an integrated experimental, process simulation, and machine learning framework to enhance methane production from food waste and cow dung co-digestion. A batch-scale 5 L digester with a 4:2:1 substrate-to-inoculum ratio was pre-fermented for three months to establish microbial activity and initiate digestion, after which methane, temperature, and pressure were monitored in real time for 15 consecutive days. Methane concentration peaked on Day 3 at 2982 ppm, followed by a gradual decline to 378 ppm by Day 15, consistent with substrate depletion and reduced microbial activity. Aspen Plus simulations, structured on the ADM1 framework, identified 320 K (47 °C) and a hydraulic retention time of 15 days as optimal conditions for methane yield, though limitations were observed in the absence of pretreatment. Complementary ML analysis enhanced predictive accuracy: Random Forest effectively modeled methane dynamics, while SVR achieved the lowest error metrics (RMSE = 339.74; MAPE = 1.57 %), establishing it as the most reliable forecasting tool. K-Means clustering and ANFIS further provided emission phase categorization and interpretability of methane concentration patterns. The findings highlight critical factors- temperature, retention time, feedstock ratio, and buffering in improving AD performance. By integrating experimental validation, process simulation, and ML based forecasting, this study advances the intelligent design and scalable optimization of AD systems for renewable energy production.