Robust parameter identification for PEMFC health monitoring: A multi-condition weighted and degradation-constrained approach via improved Aquila optimizer

Accurate parameter identification is essential for health monitoring of proton exchange membrane fuel cells (PEMFCs), yet it remains challenging due to nonlinear parameter drift and strong parameter coupling under varying operating conditions. Existing calibration frameworks are largely static and therefore often fail to maintain parameter consistency in the presence of degradation evolution and cross-condition disturbances, which limits their applicability to long-term monitoring. To address this issue, this study proposes a degradation-constrained and multi-condition-consistent calibration framework based on an improved Aquila optimizer. The framework integrates chaotic initialization to enhance global exploration, adaptive weight scheduling to balance exploration and exploitation, and a degradation-mapping module to constrain parameter evolution along aging-consistent trajectories. In addition, a multi-condition weighted objective function is introduced to enforce identification consistency across variations in temperature, humidity, and load. Validation on the NedStack PS6 and Ballard Mark V datasets demonstrates compact error distributions under multidimensional perturbations, with region-wise voltage RMSE values as low as 0.012 in the activation-dominated region, 0.015 in the ohmic-dominated region, and 0.022 in the concentration-dominated region. By embedding degradation-aware constraints into the identification process, the proposed framework provides a physically interpretable parameter-evolution pathway and supports reliable online health assessment and long-term degradation modeling for PEMFC systems. • A robust physical model calibration framework is proposed using an Improved Aquila Optimizer (IAOA). • Chaotic mapping and adaptive weights are introduced to enhance global search and convergence stability. • A dynamic health degradation mapping module constrains parameter drift to physically feasible aging trends. • Multi-condition weighted identification resolves parameter coupling effects across varying operating domains. • Experimental validation on NedStack and Ballard stacks demonstrates high-precision extraction (RMSE ≤0.022).