Stochastic Optimization Framework for Reliable Hydrogen Cost Predictions: A Case Study in North-East Italy

Hydrogen produced from renewable electricity has emerged as one of the key enablers for industrial decarbonization. Accurately predicting its production costs under technological and economic uncertainty becomes essential for robust investment planning. While numerous techno-economic assessments have evaluated Hydrogen Supply Chains (HSCs), most rely on single-point estimates or narrow sensitivity ranges for electrolyzer capital expenditures (CAPEX) and efficiencies, potentially underestimating the true variability of future costs. Recent literature has addressed HSC design optimization [1], electrolyzer cost projections [2], and uncertainty quantification through Monte Carlo approaches [3,4]; however, the systematic integration of these approaches within a unified framework for local HSC design has received limited attention. This study presents a novel four-step stochastic optimization procedure for reliable Levelized Cost of Hydrogen (LCOH) and Levelized Cost of Electricity (LCOE) predictions in local HSCs. The methodology integrates: 1) systematic collection of electrolyzer CAPEX scenarios from literature, categorized by technology (Alkaline – ALK, Proton Exchange Membrane – PEM) and size range; 2) statistical analysis and Probability Density Function (PDF) fitting to identify distributions that best represent CAPEX uncertainty; 3) Linear Programming optimization of HSC design and operation using PyPSA, solved across a 19-year historical dataset (2005–2023) of solar irradiance, grid prices, and energy demands; 4) LCOH and LCOE distribution estimation through random sampling from best-fitting PDFs and Kernel Density Estimation (KDE). The framework is applied to a grid-connected HSC located in Trieste (North-East Italy), designed to meet both hydrogen and electricity demands of an industrial end-user. The system comprises a photovoltaic plant and a 5 MW electrolyzer (ALK or PEM), with annual hydrogen production of 300 t H₂/year. Two configurations are analysed: HSC1 with Electrical Energy Storage (EES) only, and HSC2 incorporating both EES and hydrogen storage (HS). Three electrolyzer efficiency scenarios (minimum, mean, maximum) reflect potential technological improvements toward 2030. Statistical analysis reveals that gamma and beta distributions best fit the CAPEX scenarios for ALK and PEM electrolyzers, respectively. Distribution fitting confirms median CAPEX values around 730 €/kW for small-scale systems, with significant variability (500–1200 €/kW for ALK, 400–1650 €/kW for PEM). Table I presents the key performance indicators for the Trieste case study across different configurations and electrolyzer technologies with mean efficiency scenarios. Results demonstrate that improving electrolyzer efficiency from minimum to maximum scenarios reduces LCOH by 33–38% (HSC1) and up to 44% (HSC2). Hydrogen storage in HSC2 decreases LCOH by approximately 2 €/kg H₂ compared to HSC1, due to enhanced electrolyzer utilization and temporal decoupling between production and demand. However, this benefit comes at the cost of increased LCOE, as the system prioritizes hydrogen production over grid exports. PEM electrolyzers achieve lower LCOH values than ALK units owing to their higher efficiency, despite comparable CAPEX ranges at medium scale. The KDE-derived distributions reveal that uncertainty in LCOH predictions is highest for small-scale PEM systems (standard deviation up to 0.6 €/kg H₂), while medium-scale ALK configurations exhibit the most stable cost projections. This work demonstrates that integrating statistical uncertainty analysis with design-operation optimization provides a robust basis for evaluating hydrogen investment strategies under cost variability. Future developments will extend the procedure to include additional HSC configurations (e.g., ammonia-based storage), alternative renewable sources, and comparative assessment across multiple European locations.