Modeling and optimization of PEM water electrolysis via an ANN–GA hybrid approach
2
Issued Date
2025-10-13
Resource Type
ISSN
03603199
Scopus ID
2-s2.0-105016317038
Journal Title
International Journal of Hydrogen Energy
Volume
177
Rights Holder(s)
SCOPUS
Bibliographic Citation
International Journal of Hydrogen Energy Vol.177 (2025)
Suggested Citation
Boekfah A., Rumnum S., Suvanjumrat C. Modeling and optimization of PEM water electrolysis via an ANN–GA hybrid approach. International Journal of Hydrogen Energy Vol.177 (2025). doi:10.1016/j.ijhydene.2025.151576 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/112259
Title
Modeling and optimization of PEM water electrolysis via an ANN–GA hybrid approach
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Advancing water electrolysis technologies is essential for scaling up green hydrogen production. Among these, proton exchange membrane water electrolyzer (PEMWE) stands out for its high efficiency and environmental compatibility. For optimal performance, PEMWE systems must deliver high hydrogen production rates (Q<inf>H<inf>2</inf></inf>) while maintaining low electrical energy consumption (E). However, accurately predicting and optimizing these outcomes remains a complex challenge due to the nonlinear interactions of multiple operational parameters. To address this, the present study introduces an integrated artificial neural network–genetic algorithm (ANN-GA) framework for simultaneously predicting and optimizing hydrogen production rates and energy consumption in PEMWE systems. The ANN is trained using the Levenberg–Marquardt backpropagation (LMBP) algorithm, known for its rapid convergence and high precision. The model takes four critical inputs—water temperature (T<inf>H<inf>2</inf>O</inf>), water flow rate (Q<inf>H<inf>2</inf>O</inf>), applied voltage (V), and current density (J)—to forecast two outputs: Q<inf>H<inf>2</inf></inf> and E. The optimized network architecture comprises four input nodes, a hidden layer with ten neurons, and two output nodes. Model performance was assessed using mean squared error (MSE), mean absolute error (MAE), and the coefficient of determination (R<sup>2</sup>). Results show that the ANN significantly outperforms conventional regression models, achieving an R<sup>2</sup> of 0.99989 and an MAE of 0.46914. Experimental validation was conducted using the GA-optimized parameters, yielding a mean prediction error of less than 13.33 %, thus confirming the robustness of the proposed approach. This study highlights the effectiveness of the integrated ANN-GA framework in enhancing prediction accuracy and optimization capabilities for PEMWE operations. By enabling fine-tuned control over system parameters, the proposed method supports the development of efficient, scalable, and sustainable hydrogen production technologies.