Uncertainty-Aware Design and Operation of Hydrogen-Integrated Multi-Energy Systems for Hard-to-Abate Sectors

Green hydrogen is a promising option to decarbonise hard-to-abate industrial and remote energy systems, yet plant-level design and operation are affected by multiple uncertainties. In this thesis, uncertainty is treated explicitly in two ways: (i) long-horizon variability of renewable availability for capacity planning (two-stage stochastic design), and (ii) shorthorizon uncertainty in PV output and electricity prices revealed sequentially across market gates (multi-stage stochastic scheduling). Regulatory and certification uncertainty is treated through explicit constraints and sensitivity analysis. This thesis develops and applies a modelling toolbox for power-to-hydrogen (PtH) plants embedded in multi-energy systems, built chronologically from rule-based simulation to deterministic and, ultimately, two-stage and multi-stage stochastic mixed-integer linear programming (MILP), with the explicit aim of understanding the strengths and limitations of each modelling layer. The work begins with the development and use of the Multi Energy System Simulator (MESS), which couples detailed component models (electrolysers, fuel cells, hydrogen storage, batteries and auxiliaries) with multi-carrier mass and energy balances and rule-based controllers. Applied to industrial cogeneration and to grid-connected, PV-based PtH plants serving constant hydrogen demand across Italian bidding zones, MESS highlights how plant design and operation react to changes in the PV-to-electrolyser ratio, component degradation and green-hydrogen certification rules. The rule-based approach is transparent and flexible but offers no guarantee of optimality and requires extensive parameter tuning. These limitations motivate a shift towards optimisation-based methods. A deterministic MILP formulation is first introduced for joint design and operation of PtH plants, with particular emphasis on hydrogen storage technology selection. By enforcing techno–economic constraints over a representative year, it provides cost-optimal designs under perfect foresight and enables a consistent comparison of compressed-gas, underground and metal-hydride storage. This reveals that different technologies can deliver similar levelised cost of hydrogen (LCOH), with the preferred option that could be driven more by geology, footprint and mass requirements. At the same time, the deterministic formulation exposes its own weakness: by assuming known future conditions, it cannot capture the cost of hedging against variability in renewable availability and electricity prices. To address this, the thesis advances to a two-stage stochastic MILP applied to the joint sizing and operation of an off-grid hybrid renewable energy system for the island of Pantelleria in Italy. Here, investment decisions are taken before uncertainty is revealed, while hourly operation is optimised in the second stage under multiple renewable energy production scenarios. Results show that, at zero emissions, the deterministic design costs 8.10 Me/yr while the two-stage stochastic design increases the mean to 9.37 Me/yr (+15.7%). Comparing two-stage solutions with perfect-foresight designs quantifies the cost of robustness, defined as the scenario-weighted increase in costs of the two-stage solution relative to the deterministic perfect-foresight solution under the same CO2 cap, and shows how deterministic sizing can underestimate both total cost and emissions when variability is ignored. At this point, the focus is still primarily on design: uncertainty is treated to make better sizing choices and reduce LCOH at the planning stage, but operational decisions remain aggregated into a single recourse block. In practice, however, once a PtH plant has been built, the main lever left to reduce effective hydrogen cost and improve economic performance lies in its day-to-day interaction with electricity markets. The methodological path therefore culminates in a multi-stage stochastic MILP for market-coupled scheduling of a grid-connected PtH plant, already in place, serving a hydrogen refuelling station. Scenario trees for electricity prices and renewable generation represent day-ahead, intraday and reserve markets with non-anticipativity constraints and detailed device dynamics. This formulation captures how information is gradually revealed and optimises bidding and dispatch decisions so as to maximise operating profit and implicitly lower ex-post LCOH. The case studies show that co-optimised bidding and recourse operation can increase expected operating profit while reducing downside risk compared with strategies that treat prices as deterministic or ignore reserve-market participation, and provide a consistent way to value flexibility from batteries, fuel cells and hydrogen storage once the plant configuration is fixed. In the Barcelona case study, the electrolyser delivers the contracted 1440 kgH2/day at an average load factor of about 42% while the fuel cell remains mainly in standby to monetise reserve capacity. An appendix finally sketches how agrivoltaic layouts can be introduced as an alternative PV siting option in land-constrained PtH projects by modifying land-use and cost parameters without altering the core optimisation structure. In the Tuscany case study, agrivoltaics yields a positive net present value of 781 ke per hectare and an internal rate of return of 13% (vs. 1330 ke and 21% for ground-mounted PV), while preserving crop revenues. Overall, the thesis offers a chronologically developed and internally consistent framework for moving from rule-based simulation to market-aware stochastic optimisation of greenhydrogen-enabled multi-energy systems, clarifying when each modelling layer is appropriate and how neglecting uncertainty or market coupling can materially affect both plant sizing decisions and operating economics.