Data-driven optimization of proton exchange membrane water electrolyzers using an integrated artificial neural network–genetic algorithm framework
2
Issued Date
2026-03-24
Resource Type
ISSN
03603199
Scopus ID
2-s2.0-105034477508
Journal Title
International Journal of Hydrogen Energy
Volume
220
Rights Holder(s)
SCOPUS
Bibliographic Citation
International Journal of Hydrogen Energy Vol.220 (2026)
Suggested Citation
Boekfah A., Seanglumlert C., Rumnum S., Rattanaphan S., Punurai W., Suvanjumrat C. Data-driven optimization of proton exchange membrane water electrolyzers using an integrated artificial neural network–genetic algorithm framework. International Journal of Hydrogen Energy Vol.220 (2026). doi:10.1016/j.ijhydene.2026.154194 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/116107
Title
Data-driven optimization of proton exchange membrane water electrolyzers using an integrated artificial neural network–genetic algorithm framework
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Proton exchange membrane water electrolyzers (PEMWEs) represent a pivotal technology for clean hydrogen generation, directly converting electrical energy into chemical energy through water splitting. Despite their intrinsic advantages—such as low specific energy demand, high-purity hydrogen output, and seamless integration with renewable power sources—their commercialization remains constrained by performance–cost trade-offs. Addressing this limitation requires predictive and optimization tools capable of capturing nonlinear electrochemical behavior under diverse operating conditions. This study presents a hybrid artificial neural network–genetic algorithm (ANN–GA) framework for the concurrent prediction and optimization of key PEMWE performance indicators: cell voltage (V), hydrogen generation rate ((Formula presented) ), and specific electrical energy consumption (E). The ANN were trained using four key inputs—water temperature ((Formula presented) ), flow rate ((Formula presented) ), current density (J), and active area (A)—based on a dataset comprising 598 experimental operating points, divided into training (70%), validation (15%), and testing (15%) subsets. The optimized network architecture, comprising two hidden layers with fifteen and ten neurons, demonstrated excellent predictive accuracy, achieving R2 = 0.99844, MAE = 4.5781, RMSE = 39.799, MAPE = 5.9704 × 1011%, and MARE = 5.9704 × 109. Validation against independent experimental data confirmed the robustness of the model, with all error metrics remaining within acceptable limits and a mean prediction deviation 2.27%. Coupled with the GA, the model successfully identified operating and design parameters that maximize efficiency while minimizing energy demand. The proposed ANN–GA framework establishes a computationally efficient and experimentally validated methodology for PEMWE optimization, advancing the development of economically viable and high-performance hydrogen production systems.