Artificial intelligence-driven optimization of biohydrogen production: ANN-GA, RSM, and python synergy for novel Alcaligenes ammonioxydans utilizing sugarcane bagasse

Biohydrogen (bioH₂) production through dark fermentation presents a promising and sustainable alternative to fossil fuels, especially when utilizing lignocellulosic agricultural residues. In this study, sugarcane bagasse (SB) was selected as the feedstock due to its high carbohydrate content, abundant availability, and low cost, making it an ideal substrate for microbial bioH₂ production. A newly isolated and efficient bioH₂-producing bacterium, Alcaligenes ammonioxydans SRAM was employed to ferment the pretreated bagasse under anaerobic conditions.

To optimize bioH₂ yield, four critical process parameters substrate concentration, inoculum ratio, acid pretreatment concentration, and pH were systematically investigated using a Central Composite Design (CCD). Two advanced modelling approaches, Response Surface Methodology (RSM) and Artificial Neural Networks (ANN), were used to develop predictive frameworks based on the experimental data. ANN models were developed in MATLAB and Python, demonstrating superior performance over RSM by accurately capturing complex nonlinear interactions with significantly lower prediction errors.

To enhance the optimization process, the ANN model was further integrated with a Genetic Algorithm (GA), resulting in a hybrid ANN-GA model implemented in Python. This hybrid approach effectively determined the optimal conditions for maximum bioH₂ production, achieving a minimal prediction error of 0.02. The optimized parameter set included a substrate concentration of 48.98 g/L, an inoculum ratio of 8.21 % v/v, an acid concentration of 3.56 % v/v, and a pH of 7.02.

This study clearly highlights the potential of A. ammonioxydans SRAM for high-efficiency bioH₂ production and presents a robust ANN-GA-based optimization framework for enhancing bioH₂ yields from SB, advancing the transition to renewable energy sources.