Machine learning and response surface methodology for optimization of bioenergy production from sugarcane bagasse biochar-improved anaerobic digestion

Abstract

This study aimed to optimize and model the cumulative biogas yield (CBY), cumulative methane yield (CMY), and volatile solids reduction (VSR) employing artificial neural network-genetic algorithm (ANN-GA) and response surface methodology (RSM) after adding different dosages of sugarcane bagasse biochar (SBC) with varying loading rates of rice straw (RS) and inoculum. The optimal operational conditions predicted by the RSM model were SBC addition of 2.81 w/v%, RS loading of 3.17 % TS, and inoculum loading of 3.48 % TS. Validation experiments conducted under these conditions yielded CBY, CMY, and VSR values of 533.1 ± 22.3 mL/g VS, 269.7 ± 11.3 mL/g VS, and 80.3 ± 2.9 %, representing 36.9 %, 36.4 %, and 37.9 % improvements over the control. The optimal operational conditions predicted by RSM showed higher CBY (7.8 %), CMY (6.7 %), and VSR (8 %) than the GA. The CCD-RSM exhibited higher prediction accuracy, with lower prediction errors (2.8 %, 1.0 %, and 1.0 %) for CBY, CMY, and VSR compared to the GA (4.4 %, 2.4 %, and 4.7 %). It is recommended that the optimal operational conditions identified in this study be implemented in continuous pilot-scale AD systems.