Abstract

Proton Exchange Membrane Fuel Cells (PEMFCs) are a promising solution for decarbonized energy systems, yet their adoption is hindered by their complex behavior under dynamic loads and long-term degradation. This paper presents a hybrid modeling framework combining an Equivalent Electrical Circuit Model (EECM) with a data-driven degradation compensation approach using a Gated Recurrent Unit (GRU) neural network. The model is trained and validated using over 1000 hours of experimental data from a single PEMFC cell, capturing both transient voltage dynamics and aging-induced voltage loss. A novel two-stage compensation method is introduced: one based on operating current to correct steady-state offsets, and one based on operating time to model degradation effects. The final model is scaled to represent a 1kW PEMFC stack and integrated into a simulation with a DC-DC converter. Results show that as the cell degrades, higher current is required to maintain constant power output, leading to increased hydrogen consumption and reduced system efficiency. Furthermore, voltage degradation across the operating range leads to a continuous decrease in the converter's maximum deliverable power over time. These findings highlight the importance of degradation-aware modeling for predictive simulation and reliable control of fuel-cell-based power systems.