Predictive Control and Artificial Neural Networks For Energy Optimization in Microbial Fuel Cells: A Critical Review

Abstract

Microbial fuel cells (MFCs) combine wastewater treatment with direct bioelectrochemical energy recovery, but their practical deployment remains constrained by low and variable power density, nonlinear dynamics, biological uncertainty, and sensitivity to substrate and load disturbances. This critical narrative review examines how predictive control and artificial intelligence can improve MFC energy performance. A structured, non-exhaustive search of DOI-indexed records and publisher platforms was updated through August 2026 using combinations of terms related to MFCs, predictive control, neural networks, machine learning, voltage regulation, power prediction, and adaptive control. Eligible studies were critically compared by model structure, manipulated and controlled variables, constraint handling, data requirements, validation maturity, interpretability, and deployment readiness. Mechanistic models remain essential for interpreting mass transfer, microbial kinetics, charge balance, and electrode overpotentials; however, parameter uncertainty and scale dependence can limit real-time use. Proportional-integral-derivative controllers provide transparent regulation near nominal conditions but lose effectiveness under strong coupling and regime shifts. Model predictive control can explicitly handle constraints and multivariable interactions, although its performance depends on model fidelity and online optimization. Artificial neural networks offer flexible nonlinear approximation, but they require representative data and rigorous external validation. The principal contribution is a cross-framework assessment showing that the most defensible architecture is hybrid: physical balances define feasible behavior, neural models compensate for residual nonlinearities, and predictive control coordinates feed, load, and operating constraints. Five evidence gaps persist: weak cross-reactor transferability, inadequate uncertainty treatment, limited closed-loop experimental validation, non-standardized comparisons, and incomplete scale-up economics. Progress therefore requires reproducible, constraint-aware systems tested under common disturbances and long-duration operation.