Physics-Informed Hybrid Modeling for End-to-End Prediction of Hydrogen Refueling Process Using Operational Data

Abstract

As Hydrogen Fuel Cell Electric Vehicles (HFCEVs) expand, accurate prediction of the entire refueling process is essential for improving refueling protocols and station operation. This study proposes a Physics-Informed Hybrid Modeling Framework that combines thermodynamic interpretability with data-driven calibration. The framework consists of five sub-models for supply pressure, flow rate, supply temperature, tank temperature, and tank pressure, preserving governing-equation structures while automatically calibrating 24 empirical parameters using Bayesian optimization. Approximately 32 months of operational data from a commercial hydrogen refueling station were used, with 4,943 training scenarios for parameter calibration and 4,943 validation scenarios for Initial Value Problem (IVP)-based prediction. The framework consistently reproduced refueling behavior across all sub-models, showing particularly high accuracy for supply pressure, tank temperature, and tank pressure. These results demonstrate its potential as a practical simulation tool for optimizing station operation and designing high-speed refueling strategies.