Techno-economic optimization of power allocation in large-scale hydrogen production with degradation-aware control strategies

This paper evaluates six real-time power-allocation strategies (Round Robin, Equal, DV-order, DV-weighted, Vdeg-MP-reward, and Adaptive Degradation Balancing, ADB) for a 250 MW PEM electrolyzer fleet (556 stacks) utilizing 2024 Danish offshore wind and spot price data. Stack heterogeneity is modeled via a quadratic, current-dependent degradation law. For each method, we report annualized outputs including hydrogen production, operational costs, production-gate Levelized Cost of Hydrogen (LCOH), and diagnostic metrics such as degradation spread and estimated stack lifetime. Across heterogeneity scenarios, the methods yield divergent production-gate LCOH values ranging from 4.01 to 5.05 €/kg, highlighting significant economic trade-offs. To validate the resiliency of these strategies, we performed extensive robustness checks, including bounded sensitivity analysis on physical degradation parameters, Tornado variance decomposition, and multi-year bootstrap uncertainty analysis. These analyses demonstrate that while the baseline Equal method offers high robustness under extreme degradation stress, the Vdeg-MP-reward method achieves the global minimum production-gate LCOH (4.01 €/kg) and highest production volume by dynamically balancing marginal revenue against future degradation penalties. Conversely, ADB yields the optimal strategy for asset longevity, achieving the lowest replacement annuity (5.05 M€/year—approximately 28 % lower than the Equal baseline). These results indicate that accounting for degradation kinetics materially improves long-term asset utilization and operational stability in large-scale Power-to-X systems. • Degradation-aware control optimized for 250 MW electrolyzer fleet. • Vdeg-MP strategy achieves minimum LCOH of 4.006 EUR/kg. • ADB method reduces stack replacement annuity by 28 %. • Rigid sorting logic causes 24 M€/year in startup penalties. • Robustness validated via Monte Carlo and parameter drift tests.