Machine Learning Prediction and Process Optimization for Enhanced Methane Production from Straw Anaerobic Digestion with Biochar

Anaerobic digestion of straw is a crucial method for agricultural waste valorization, yet its efficiency is limited by complex factors. This study employed machine learning to predict methane yield and optimize process parameters in biochar-amended straw digestion. A comprehensive dataset integrating experimental and literature data (100 samples, 15 input variables) was constructed, incorporating operational conditions, straw characteristics, and biochar properties (e.g., dosage, particle size, specific surface area, and elemental composition). Prediction models were developed using Random Forest (RF), XGBoost, and Support Vector Machine (SVM). Results indicated that the RF model achieved the best predictive accuracy, with an R2 of 0.81 and RMSE of 36.9, significantly surpassing other models. Feature importance analysis identified feeding load, biochar dosage, and biochar carbon content (C%) as the key governing factors, collectively accounting for 65.7% of the total contribution. The model-predicted optimal ranges for practical operation were 15–30 g for feeding load and 5–20 g/L for biochar dosage. This study provides data-driven validation of biochar’s enhancement mechanisms and demonstrates the utility of RF in predicting and optimizing anaerobic digestion performance, offering critical support for sustainable agricultural waste recycling and clean energy generation.