Precise estimation of fuel cell parameters is critical for optimizing performance and developing energy systems. However, experimental data are often affected by outliers stemming from inaccurate measurements, transient operating conditions, or environmental variations. In this line, this study proposes a robust approach for estimating proton exchange membrane fuel cell (PEMFC) parameters. This study focuses on the steady-state current–voltage (I–V) characteristics and performs parameter extraction for a semi-empirical model. The proposed estimation framework employs the collaboration of the Huber loss function (HLF) in conjunction with adaptive hyperparameter and the metaheuristic Wild Horse Optimizer (WHO) to compute seven unknown PEMFC parameters. The impact of different hyperparameter (δ) values is examined on the performance of the HLF while estimating key fuel cell parameters. The sensitivity of the estimating process to the δ-value is explored using measured and estimated datasets, including accuracy, convergence rate, and resilience. The WHO-based approach is adopted to address and mitigate issues such as premature convergence and entrapment in local optima, which are common challenges in existing optimization strategies. The proposed model has been tested and verified through three test samples of standard commercial PEMFC units as benchmarks. The simulation results demonstrate that the WHO exhibits robust performance across the three benchmark PEMFC systems. Furthermore, the proposed model's generalization capability is validated under a range of operating conditions using polarization curves generated at different temperatures and cathode stoichiometries. A single globally specified parameter set reliably predicts fuel cell performance across these diverse conditions, as evidenced by its consistent ability to deliver high-quality solutions with an extraordinary level of precision under predefined experimental conditions. The proposed estimation framework outperforms three commercial PEMFC units (NedStack-PS6, Horizon-500 W, and BCS-500W), achieving Huber loss values of 1.03277845, 0.00562094, and 0.00584889, respectively. The adaptive HLF with hyperparameter (δ) ranging from 0.5 to 2.0 efficiently tackles outliers and improves convergence speed. While the hyperparameter (δ) in previous studies was kept constant, δ = 1. The proposed estimation framework closely matches the experimental data and offers significantly higher accuracy compared to existing competing methods in the literature. The results reveal that the suggested HLF enhances the robustness and immunity of the WHO optimizer, and it outperforms traditional approaches such as steady-state error.