Book

Accurate identification of electrochemical parameters is essential for fuel-cell performance prediction, control, diagnostics, and digital-twin operation; however, conventional metaheuristic approaches generally require a new iterative optimization for each stack or operating condition and do not explicitly transfer information between physically related states. This study proposes a unified Physics-Informed Heterogeneous Graph Neural Network for Fuel Cells (PI-HGNN-FC) for multicondition parameter identification and performance modeling of Proton Exchange Membrane Fuel Cells (PEMFCs) and Solid Oxide Fuel Cells (SOFCs). Polarization measurements, operating conditions, stack descriptors, and electrochemical mechanisms are represented through a heterogeneous graph and processed using relation aware multi-head attention.The evaluation considers BCS 500 W and NedStack PS6 PEMFC benchmarks together with a 96-cell tubular SOFC over 1073–1273 K and 1–9 atm, using source-equivalent calibration, cross-stack prediction, leave-one-condition-out validation, measurement-noise and outlier perturbations, sparse-data identification, uncertainty calibration, ablation analysis, repeated-run statistics, and computational assessment. The physics-equivalent PI-HGNN-PE proposed framework the established seven-parameter calibration level with BCS 500 W and NedStack PS6 SSE.The framework therefore provides a foundation for adaptive parameter tracking, fuel-cell digital twins, online diagnostics, and real-time optimal operation of PEMFC and SOFC systems.