Investigations of the Parameter Estimation Approach Based on the Actor-Critic-Assisted Optimization Algorithm for Proton Exchange Membrane Fuel Cells

Accurate and reliable mathematical modeling, underpinned by precise parameter estimation, is crucial for the optimal control and performance analysis of proton exchange membrane fuel cell (PEMFC) systems. This study introduces the actor-critic-assisted dwarf mongoose optimization (AcDMO) algorithm to accurately extract the unknown parameters of PEMFC voltage models. Driven by an intelligent reinforcement learning engine, the proposed method integrates actor-critic dynamic perception, elite differential guidance, and the Lévy flight strategy to effectively resolve the structural flaws of traditional optimization algorithms. To rigorously validate its engineering performance, the AcDMO algorithm is evaluated against six prominent baseline optimizers within a comprehensive multi-dataset framework. This encompasses actual empirical data acquired from a custom 5 cm² PEMFC single-cell evaluated at 353 K on the YK-A10 test platform, alongside three commercial stack datasets (BCS 500 W, NedStack PS6, and Horizon 500 W). Experimental results quantitatively demonstrate that AcDMO achieves exceptional electrochemical fidelity, consistently securing an exceptionally low sum of squared errors (SSE) and capturing near-unity coefficients of determination. Furthermore, non-parametric statistical evaluations across 30 independent runs including Friedman and Wilcoxon signed-rank tests confirm the absolute structural consistency and robustness of the algorithm, decisively rejecting the null hypothesis of equal performance. Ultimately, this work provides a highly reliable, experimentally validated foundation for the digital-twin modeling of PEMFC systems under dynamic conditions.