Current-Step-Based Fast Electrochemical Parameter Identification for PEMWE Using a Physics-Informed Neural Network

Abstract

Electrochemical parameter identification is crucial for evaluating the electrochemical processes in proton exchange membrane water electrolysis (PEMWE). Conventional characterization techniques-including polarization-curve fitting, electrochemical impedance spectroscopy (EIS), cyclic voltammetry (CV), and current interruption (CI)-face significant limitations for rapid diagnostics under high-current dynamic operation, arising from constraints in instrument current rating, measurement time, zero-current control, and noise amplification in numerical differentiation. In this study, we present a simple current step (CS) method to accurately identify key electrochemical parameters and perform overpotential breakdown by using a simplified equivalent circuit model with a current source. To address the numerical instability in derivative calculation caused by sampling noise during voltage transient analysis, a physics-informed neural network (PINN) is introduced to enhance signal smoothness while guaranteeing physical consist ency. Compared with standard characterization, the proposed CS-PINN method demonstrates high accuracy, with an error of less than 2% in overpotential breakdown, less than 5.3% in ohmic resistance, and 2.8% in the Tafel slope (at 5 A/cm2). These results confirm that the CS-PINN method provides a fast, accurate, and equipment-friendly route for rapid electrochemical parameter identification in PEMWE.