Integrated multi-objective optimization of sodium bicarbonate pretreatment for the outer anatomical portion of corncob using central composite design, artificial neural networks, and metaheuristic algorithms

Abstract

Effective pretreatment of raw biomass is crucial in establishing a successful biorefinery process, and response surface methodology (RSM), a statistical optimization method, is routinely employed for optimizing pretreatment methodologies. However, relying solely on statistical methods fails to capture the nonlinear relationships between pretreatment methodology and lignocellulose recalcitrance. Machine learning (ML) approaches excel at understanding these relationships. Therefore, combining statistical and ML optimization can establish better pretreatment parameters. This study presents a multi-stage approach to identify and optimize a chemical pretreatment for the outer anatomical portion of corncobs (CO). Various chemicals were screened using fixed factor screening and regular two-level factorial design (2FI), and among them NaHCO3 was chosen for further optimization, using central composite design (CCD). Moderate delignification (33.96%), minimum sugar loss (5.26 mg/g of CO), and a high enzymatic saccharification yield of up to 82% were achieved with the design. Three distinct metaheuristic algorithms, namely Teaching-Learning-Based Optimization (TLBO), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA) were employed to optimize three specific hyperparameters of the artificial neural network (ANN), to create hybrid metaheuristic-optimized ANN models, which were subsequently used to validate the results obtained from the CCD. Among the different models, The TLBO-optimized ANN model provided the most accurate predictions than the CCD.