Accurate and efficient parameter identification of fuel cells using tree seed algorithm based on adaptive differential evolution and Dempster-Shafer evidence theory

Accurate parameter identification is vital for reliable modeling and control of proton exchange membrane fuel cells (PEMFCs). Conventional single parameter identification methods primarily rely on single metaheuristic algorithms, which have limitations in accuracy, robustness, and convergence. To overcome these limitations, this study develops a hybrid algorithm that integrates the tree seed algorithm (TSA) with adaptive differential evolution (ADE) and Dempster–Shafer evidence theory (DST) for PEMFC model parameter identification. TSA provides the overall optimization framework, maintaining a balance between global and local exploration. ADE improves the diversity and search capability of the seed population, while the adaptive feedback mechanism dynamically adjusts the number of seeds and search trends according to iterative results, maintaining a better balance between exploration and exploitation. DST further fuses the positional and objective information from each iteration, refining the search range and accelerating convergence. Validation on a self-designed 50 W PEMFC system shows that the mean absolute percentage error (MAPE) of the proposed model remains within 1.5 %. For three different commercial fuel cell stacks, the proposed method achieves at least a 22.9 % improvement in computational efficiency compared with several advanced algorithms. These results confirm the high accuracy, robustness, and efficiency of the proposed TSA-ADE-DST algorithm, providing a practical tool for PEMFC modeling and system optimization.