Data-driven prediction of ejector performance for PEMFC hydrogen fuel recirculation system

Ejector-based hydrogen supply and circulation subsystem emerged as a cost-effective solution for operating a Proton Exchange Membrane Fuel Cell (PEMFC) system. However, design optimization of the ejector-based subsystem requires accurate hydrogen entrainment performance modelling using computationally intensive CFD simulations. This work introduces a machine-learning and data-driven approach to predict the ejector’s entrainment performance with much-reduced computations accurately. Comprehensive ejector performance data from twenty PEMFC stacks were generated under different maximum stack power, ranging from 6.5 to 200 kW. An optimal linear regression model using eight selected features showed less than 10% mean absolute percentage error (MAPE) for both training and testing datasets. Furthermore, the model was validated on small, medium, and large PEMFC stacks with a MAPE below 8% on ejector performance compared to the CFD simulation results.