A voltage balancing strategy based on AESO‑MF‑PC for a two‑stage ISOP cascaded converter in off‑grid DC hydrogen production systems

Modular DC-DC converters with a two-stage input-series output-parallel (ISOP) configuration play a pivotal role in off-grid hydrogen production systems. The input-series connection enables direct integration with medium-voltage DC buses, while the output-parallel structure supports the high-current requirements of electrolyzers. However, under DC bus fluctuations caused by renewable intermittency, load variations, and measurement noise, conventional controllers and fixed-gain observers struggle to simultaneously maintain fast dynamic response and effectively suppress capacitor voltage and output current ripples. These limitations can shorten electrolyzer lifespan. To address this issue, a novel voltage balancing strategy, AESO-MF-PC, is proposed by integrating Model-Free Predictive Control with an Adaptive Extended State Observer. Specifically, the AESO uses a Kalman filter-based gain adaptation mechanism and a regularized cost function to enhance estimation robustness under stochastic disturbances. The implementation details of the proposed strategy are presented, and closed-loop stability is analyzed. Operating at a fixed switching frequency, the proposed strategy analytically derives the control law and avoids the exhaustive online optimization typical of conventional predictive control. Additionally, it enables online gain adaptation without increasing the observer order, maintaining high real-time efficiency while achieving a favorable trade-off between dynamic response and ripple suppression. Simulation and experimental results demonstrate the feasibility and effectiveness of the proposed strategy compared with both conventional PI and fixed-gain LESO methods.