Theory-guided data-driven prediction of PEMEC degradation and durability for reliable green hydrogen production

Accurately predicting the performance degradation of Proton Exchange Membrane Electrolyzer (PEMEC) stacks is critical for optimizing the techno-economic viability of green hydrogen ( GH₂ ) production. Typical approaches for PEMEC degradation modelling choose between physics-based models that are computationally expensive and simplify the interrelated degradation mechanisms, or "black-box" machine learning models that often lack interpretability and generalizability. This work introduces a novel hybrid framework to overcome these limitations: a Theory-Guided Data-Driven (TGDD) model for predicting PEMEC performance degradation. The "theoryguided" component models known current-voltage relationships in electrolyzers, with the data-driven component predicting the (otherwise unknown) exchange current and resistance. This structure makes accurate, generalized, and physically-grounded predictions of PEMEC response curves, enabling simultaneous predictions of operating and reference points to facilitate degradation prognosis in optimal control and state-of-health monitoring. The TGDD model was developed and validated using comprehensive, long-term operational data obtained from PEMECs under various operating conditions. The model demonstrates high accuracy, with an R² of more than 94 %. Furthermore, the model exhibits enhanced robustness, interpolation, and explainability. Superior interpretability makes the TGDD framework a powerful tool for prognostics and health management in PEMEC systems, enabling future development of intelligent control strategies to mitigate degradation and accelerate the commercialization of GH₂.

Keywords: Control, Degradation, Green Hydrogen, Machine Learning, PEMEC, Prognosis, Safety.