Modeling and optimization of biomethanation of rice straw with biochar supplementation using response surface methodology and machine learning

Abstract

Anaerobic digestion (AD) of lignocellulosic wastes offers sustainable waste management with the production of renewable energy and nutrient-rich bio-slurry. However, the chemical recalcitrant structure of lignocellulosic waste hinders its hydrolysis and biomethanation under AD. Biochar addition has been reported to alleviate toxicity inhibition and improve the degradability of lignocellulosic wastes, biogas and methane yield, and stability of the AD process. Therefore, in this study, substrate loading (% total solids (TS)), inoculum loading (% TS), and biochar dosage (w/v%) were optimized to maximize the methane yield by using central composite design (CCD) based response surface methodology (RSM) and genetic algorithm (GA). The second-order quadratic model was established by CCD-RSM, which revealed the notable interaction between substrate loading and biochar dosage (p-value < 0.0001) and between inoculum loading and biochar dosage (p-value < 0.05). Based on the root mean square error (RMSE) and coefficient of determination (R²) values, the cumulative methane yield (CMY) prediction performance of the artificial neural network (ANN) (RMSE = 0.876, R² = 0.9894) was more reliable and accurate than CCD-RSM (RMSE = 3.34, R² = 0.9956). The GA optimal conditions showed 8.6% higher methane yield (293.7 ± 7.26 mL/g VS) than the CCD-RSM (270.2 ± 10.69 mL/g VS). The methane yield obtained at optimal conditions of GA was 54.9% higher than the control. The CCD-RSM and ANN-GA can also be used for process modeling and optimization in other contexts. The optimal outcomes obtained in this study could pave the way for the prediction and operation of continuous AD of rice straw supplemented with additives such as biochar for large-scale bioenergy production.