.png){kind=icon}
.png){kind=icon}
Return
Self-loop physics-informed neural network for model predictive control of PEM electrolyzers
.png){kind=icon}
PRE
AI
Z
L
H
DOI：10.1016/j.jprocont.2026.103707.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Effective thermal management of proton exchange membrane (PEM) electrolyzers is critical for ensuring both high efficiency and long-term durability. In this work, a physics-informed neural network (PINN) surrogate model is developed to accurately and efficiently predict the thermal behavior of a PEM electrolyzer, meeting real-time control requirements. The proposed surrogate employs a self-loop architecture that directly incorporates control inputs-electrical current and coolant flow rates-enabling recursive one-step-ahead predictions of temperature evolution under dynamic operating conditions. Physical knowledge derived from energy-balance equations is embedded into the training loss, enforcing consistency with conservation laws while reducing the need for extensive datasets and enhancing generalization. The trained PINN closely reproduces the results of a high-fidelity first-principles thermal model, maintaining negligible error across diverse transient scenarios. Furthermore, when integrated within a nonlinear model predictive control (NMPC) framework, the PINN surrogate successfully regulates the electrolyzer's temperature at desired setpoints, even under rapidly fluctuating power inputs. Overall, the proposed approach demonstrates a computationally efficient and physics-consistent pathway for real-time modeling and control of complex energy systems.
Keywords:
PINN
PEM electrolyzer
Thermal management
Self-loop PINN
Real-time control
Journal
IF:
3.9
Papers:
3.4K
Citations:
7.3K
Organization
Cited Papers