Data-driven prediction and multi-objective optimization of pemfc performance using an ANN–GA hybrid model

Proton exchange membrane fuel cells (PEMFCs) are regarded as a key clean energy technology for transportation, portable power devices, and stationary power generation due to their high efficiency, low operating temperature, and zero-emission characteristics. Improving PEMFC performance while reducing system cost remains a critical challenge, requiring accurate prediction tools and robust optimization strategies. This study proposes a novel, unified artificial neural network–genetic algorithm (ANN–GA) framework for simultaneous performance prediction and optimization of PEMFC systems. A multilayer perceptron ANN, with its architecture and hyperparameters optimized using a genetic algorithm, was trained using 239 experimentally obtained datasets to predict cell voltage ( V ) and power density ( I ). The model accounts for key operating and design parameters, including hydrogen flow rate ( Q H 2 ), anode relative humidity ( RH a ), anode back pressure ( P a ), cell operating temperature ( T PEMFC ), anode stoichiometric ratio (λ a ), oxygen flow rate ( Q O 2 ), cathode relative humidity ( RH c ), cathode back pressure ( P c ), stack number ( n ), active area ( A ), and current density ( J ). Sensitivity analysis revealed that operating temperature is the most influential factor affecting PEMFC performance, followed by stack number. The optimized ANN exhibited excellent predictive accuracy, achieving a coefficient of determination of R² = 0.99868 and a mean squared error of 0.0007655, with a mean absolute prediction error of 6.27% across the independent ANN test dataset, corresponding to a coefficient of determination of R² = 0.99868. For the optimization stage, the trained ANN was coupled with a genetic algorithm to perform multi-objective optimization, in which PEMFC performance indicators and cost-related outputs were simultaneously predicted and subsequently aggregated using a weighted-sum strategy to identify an optimal trade-off operating condition. The proposed framework represents a distinct advancement over existing data-driven PEMFC models, offering a computationally efficient, experimentally validated, and practically deployable tool for the design and optimization of high-performance, cost-effective PEMFC systems for next-generation hydrogen energy applications.