PemNet: A Transfer Learning-Based Modeling Approach of High-Temperature Polymer Electrolyte Membrane Electrochemical Systems

Widespread adoption of high-temperature polymer electrolyte membrane fuel cells (HT-PEMFCs) and HT-PEM electrochemical hydrogen pumps (HT-PEM ECHPs) requires models computational tools that provide accurate scale-up optimization. Knowledge-based modeling has limitations as it is time consuming information about the system not always available (e.g., material properties interfacial behavior between different materials). Data-driven on other hand, easier to implement, but often necessitates large datasets could be difficult obtain. In this contribution, knowledge-based data-driven are uniquely combined by implementing a Few-Shot Learning (FSL) approach. A model originally developed for HT-PEMFC was used generate simulated data (887,735 points) pretrain neural network source model. Furthermore, HT-PEMFCs successfully applied ECHPs - utilizes similar materials cell. Experimental from both with operating conditions (~50 points each) were train 8 target via FSL. Models unseen reached high accuracies in all cases (rRMSE 1.04 3.73% HT-PEMCs 6.38 8.46% ECHPs).