Optimising the power regeneration and chemical oxygen demand removal in microbial fuel cell systems using integrated soft computing methods and multiple-objective optimisation

Microbial fuel cells (MFCs) have recently emerged as a sustainable technology for simultaneously treating wastewater and generating electricity. However, optimising their operational parameters to enhance performance remains a complex challenge. This study proposes an integrated framework that combines advanced machine learning models—long short-term memory (LSTM) and gated recurrent unit (GRU)—with a multi-objective genetic algorithm (MOGA) to optimise chemical oxygen demand (COD) removal and power output. Experimental data were obtained by varying glucose concentrations (1–9 g/L), yeast extract concentrations (1–5 g/L), and aeration rates (0–110 mL/min). Among the models evaluated, the LSTM model performed best in predicting COD removal. In contrast, the GRU model outperformed the others in power prediction. These surrogate models were incorporated into the MOGA to identify nine Pareto-optimal solutions. Experimental validation confirmed the high accuracy of the proposed approach, with average errors of 5.47 % for COD and 3.29 % for power. This work offers a cost-effective and scalable optimisation strategy, significantly reducing the need for exhaustive experimental trials while improving the efficiency and applicability of MFCs in real-world scenarios.