An improved gorilla troops optimizer for enhanced parameter estimation of proton exchange membrane fuel cells

Purpose This paper aims to deliver accurate and robust parameter identification for proton exchange membrane fuel cells (PEMFCs) by enhancing the gorilla troops optimizer (GTO) with physics-consistent search mechanisms and evaluation-aware control. Design/methodology/approach We propose an improved GTO (IGTO) tailored to PEMFC polarization modeling. The identification problem is formulated as the minimization of the discrepancy between measured and model-predicted voltage–current characteristics, using a weighted sum of squared errors (SSE) objective with an optional Huber loss and light ridge regularization. IGTO integrates (i) vector (parameter-wise) bounds aligned with PEMFC physics, (ii) adaptive schedules for the main control parameters to coordinate exploration and exploitation, (iii) explicit elitism to preserve the best solution, (iv) opposition-based restart with jitter to mitigate stagnation and (v) early stopping under objective stabilization. The method is validated on three commercial PEMFC stacks (Horizon 500 W, BCS 500 W and NedStack PS6) and benchmarked against baseline GTO and representative metaheuristics under matched settings. Findings Across all stacks, IGTO achieves lower identification errors and improved stability over R = 30 independent runs. The best obtained SSE values are 2.06435 (NedStack PS6), 1.114\times 10-2 (BCS 500 W) and 1.102\times 10-2 (Horizon 500 W), with consistently faster convergence and reduced run-to-run dispersion compared with the baseline. Practical implications The method enables routine, reliable PEMFC parameter tuning for diagnostics, control and digital-twin applications, shortening calibration time and reducing computational cost. Social implications More accurate fuel-cell models support higher efficiency, reliability and lifetime, contributing to cleaner energy systems and reduced emissions. Originality/value IGTO provides a coherent, PEMFC-driven enhancement of GTO that couples physics-aligned vector bounds, robust identification objectives and evaluation-aware search control (adaptive schedules, elitism, restart and early stopping), enabling reliable parameter extraction in a practical MATLAB workflow.