Parameter estimation of proton exchange membrane fuel cells using enhanced Fourier transform optimizer with adaptive learning mechanisms

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Abstract The seven unknown parameters in the Amphlett semi-empirical proton exchange membrane fuel cell (PEMFC) model are critical to accurate performance prediction, system design, and degradation diagnosis. Gradient-based methods are often unreliable in this context because the sum-of-squared-errors (SSE) objective is highly nonlinear and multimodal, while existing population-based approaches tend to exhibit substantial run-to-run variability. To address these challenges, this paper proposes the Enhanced Fourier Transform Optimizer (EFTO), a hybrid adaptive algorithm that combines a success-history differential evolution branch with elite-guided mutation and a covariance-adaptive sampling branch, dynamically balanced through an online branch-selection mechanism. On the IEEE CEC 2017 benchmark suite (28 functions, D = 50, 30 independent runs), EFTO achieved a Friedman rank of one with an average rank of 1.607 and a Wilcoxon p ≤ 1.17 × 10⁻ 3 against six competing algorithms. A supplementary comparison with three recognized state-of-the-art solvers found no statistically significant performance differences relative to two of the three solvers ( p > 0.05), while EFTO recorded the lowest average runtime of 8.653 s. When applied to three commercially validated PEMFC stacks, EFTO achieved machine-precision reproducibility, with a standard deviation below 10⁻ 14 V 2 across 30 independent runs and best SSE values of 2.0656 V 2 (NedStack PS6), 1.0564 V 2 (AVISTA SR-12), and 0.8139 V 2 (Ballard Mark V). These results were statistically confirmed by Wilcoxon signed-rank tests (R⁺ = 465, p = 1.73 × 10⁻⁶). Component wise ablation showed that the covariance-adaptive sampling mechanism was the most influential component, with removal of the covariance-adaptive sampling mechanism causing a 98.7% degradation in SSE ( p = 1.09 × 10⁻ 24 ). Stack-specific sensitivity analysis further demonstrated that the identified parameters remain physically consistent with the electrochemical operating regime of each stack, establishing EFTO as a competitive and consistent tool for PEMFC parameters identification on the evaluated benchmark configurations.

Links and sources

• PDF from original source

Comments