Data-driven robust optimization and capacity configuration of integrated hydrogen production system considering hierarchical synergistic operation of e-SMR and hybrid electrolyzers

Producing hydrogen from renewable energy sources (RES) is a promising technical route to mitigate energy resource depletion and environmental issues caused by the utilization of fossil fuels. However, single-type hydrogen production approach is difficult to adapt to the inherent volatility of RES. To address the complex source-load uncertainties, this study proposes a novel integrated hydrogen production architecture coupling an electrified steam methane reforming (e-SMR) unit with hybrid electrolyzers. First, a high-fidelity mathematical model is constructed via precise parameter identification. Second, to address source-side uncertainties, an improved CNN-BiLSTM-Attention network optimized by the slime mold algorithm (SMA) is developed to generate high-precision robust renewable energy prediction intervals. Subsequently, a three-tier hierarchical synergistic operation strategy based on empirical mode decomposition (EMD) is established: low-frequency power components are allocated to e-SMR and alkaline (ALK) units, while high-frequency fluctuations are absorbed by proton exchange membrane (PEM) electrolyzers, ensuring precise frequency-to-equipment matching. Driven by these data-driven boundaries and physical constraints, a bi-level multi-objective data-driven robust optimization framework is formulated and solved via an improved NSGA-II algorithm to determine the optimal capacity configuration. Results indicate that the prediction model achieves an average accuracy of 94.31%. Validated under both steady and fluctuating demand scenarios, the proposed configuration strictly limits the levelized cost of hydrogen (LCOH) fluctuation to 5.90% and energy loss to 0.41% under extreme uncertainties. Comparative analysis reveals that the proposed system reduces LCOH by 30.70% and 17.65% compared to pure water electrolysis and e-SMR baselines, respectively. This work provides a highly robust and economically viable techno-economic solution for stable hydrogen production.