ABSTRACT
The integration of optimization techniques and deep learning models, which offer a promising avenue for improving the efficiency and sustainability of biodiesel production processes from baobab seed oil (BSO), is rare. This study utilized a multi-input-multioutput (MIMO) deep learning technique and the most recent central composite design (CCD) optimization tool to model and optimize the yield and properties of biodiesel produced from BSO. First, the baobab seed oil was extracted using a solvent extraction method. BSO was subsequently analyzed and converted to biodiesel by reacting CH3OH catalyzed by waste banana bunch stalk biochar activated by KOH. Multiobjective optimization and prediction of the biodiesel yield (Y) and several key fuel properties, including the cetane number (CN), kinematic viscosity (VS), and purity (P), were achieved. With better correlation coefficients of 0.9709, 0.9464, and 0.9714 for response training, response testing, and response validation, respectively, and a root-mean-square error of 0.00755, the MIMO model on the logsig transfer function accurately predicted the biodiesel yield and properties more than did the MISO and response surface methodology models. The optimum Y (96 wt %), CN (48), VS (3.3 mm2/s), and P (98.3%) were concurrently accomplished at a reaction temperature of 56 °C, a reaction time of 115 min, a CH3OH/BSO molar ratio of 15:1, a catalyst dosage of 6 wt %, and a stirring speed of 400 rpm with 98% optimal validation accuracy. CCD sensitivity analysis revealed that the CH3OH/BSO ratio was the most sensitive (50.9%) input predictor among the other input variables studied.
