This work presents a hybrid mechanistic/data-driven modeling approach for predicting the performance of PEM electrolyzer cells and optimizing their dynamic operations. The system of interest involves a cyber-physical prototype for lab-scale water electrolysis with online monitoring and data collection [4]. A mechanistic model is first developed based on open literature [5], which comprises detailed thermal modeling, mass transfer modeling, electrochemical modeling, etc. Key process parameters are regressed based on experimental data (e.g., for polarization curve) [6]. However, notable mismatches are still observed between experimental data and mechanistic predictions in voltage variations and hydrogen flowrates, showing inefficacy to accurately capture the physical phenomena during continuous operations. Hybrid modeling [7,8] is thus employed to further enhance the model prediction capabilities by integrating low-fidelity data from mechanistic models with high-fidelity lab-scale experimental data. Specifically, this work leverages the Physics-Informed Neural Network (PINN) framework [9]. Physical loss is integrated into the PINN loss function, accounting for the correction of temperature and hydrogen flowrate predictions. The information of both absolute values and time-derivatives are considered to enhance both the accuracy and interpretability of the model. The developed PINN demonstrates a more accurate representation of the relationships between current density, voltage, and associated thermal and mass transfer phenomena, thereby improving the predictive model's accuracy and reliability. The performance of the first-principles model (white-box), hybrid model (grey-box), and machine learning-based model (black-box) are systematically evaluated and compared. This study provides a robust hybrid model that describes the mass and energy dynamics within the electrolyzer, laying the foundation for exploring control strategies, while also offering insights for the design and operations of PEM electrolyzers.
References
[1] IEA (2024), Share of renewable electricity generation by technology, 2000-2030, IEA, Paris https://www.iea.org/data-and-statistics/charts/share-of-renewable-elect…, Licence: CC BY 4.0
[2] Hassan, Q., Algburi, S., Sameen, A. Z., Salman, H. M. & Jaszczur, M. A review of hybrid renewable energy systems: Solar and wind-powered solutions: Challenges, opportunities, and policy implications. Results in Engineering 20, 101621 (2023).
[3] Shiva Kumar, S. & Himabindu, V. Hydrogen production by PEM water electrolysis – A review. Materials Science for Energy Technologies 2, 442–454 (2019).
[4] Liu, Y., Akundi, S. S., Braniff, A., Dantas, B., Tian, Y., Niknezhad, S. S., Khan, F. I., & Pistikopoulos, E. N. (2024). Cyber-physical systems in chemical and energy processes. In Methods in Chemical Process Safety (Vol. 8, pp. 215-241). Elsevier.
[5] Majumdar, A., Haas, M., Elliot, I., & Nazari, S. (2023). Control and control-oriented modeling of PEM water electrolyzers: A review. International Journal of Hydrogen Energy, 48(79), 30621-30641.
[6] Dantas, B., Akundi, S. S., Liu, Y., Braniff, A., Niknezhad, S. S., Khan, F., Pistikopoulos, E. N., Lima, F.V., Tian, Y. Model-based Operability and Safety Optimization for PEM Water Electrolysis. Systems & Control Transactions. Accepted.
[7] Bradley, W., Kim, J., Kilwein, Z., Blakely, L., Eydenberg, M., Jalvin, J., Laird, C., & Boukouvala, F. (2022). Perspectives on the integration between first-principles and data-driven modeling. Computers & Chemical Engineering, 166, 107898.
[8] Karniadakis, G. E., Kevrekidis, I. G., Lu, L., Perdikaris, P., Wang, S., & Yang, L. (2021). Physics-informed machine learning. Nature Reviews Physics, 3(6), 422-440.
[9] Moayedi, F., Chandrasekar, A., Rasmussen, S., Sarna, S., Corbett, B., & Mhaskar, P. (2024). Physics-informed neural networks for process systems: handling plant-model mismatch. Industrial & Engineering Chemistry Research, 63(31), 13650-13659.