Hydrogen Supply Chain
MILP · Multi-echelon
Electrolyzers · Pipelines · Storage · Demand
Where should green-hydrogen electrolyzers be built? How should hydrogen move from producer to consumer — pipeline, truck, liquefied ship, or ammonia conversion? How much storage is needed to buffer hourly variability? The hydrogen supply chain (HSC) problem is a multi-echelon MILP with production-location binaries, flow decisions on multiple transportation modes, storage sizing, and demand-satisfaction constraints. Almansoori & Shah (2006) gave the seminal formulation; the 2020+ literature exploded alongside the global green-hydrogen push.
The problem
Green hydrogen from wind to steel plant
Hydrogen is having a moment. The IEA's Net Zero roadmap calls for 200 Mt/year of clean hydrogen by 2050, up from ~1 Mt today (almost all fossil-based). Steel, ammonia, heavy trucking, shipping, aviation fuel, seasonal electricity storage — hydrogen is the chemical vector that decarbonizes sectors electricity alone cannot reach. But realizing this requires building out a physical supply chain: tens of gigawatts of electrolyzers, thousands of kilometers of pipelines or shipping routes, storage at salt caverns or as ammonia.
The operations-research problem: given expected demand centers (refineries, steel plants, mobility hubs), candidate production sites (near cheap renewables), transportation-mode costs (pipeline: high capex, low opex; truck: high opex, flexible; LH2 ship: only viable for transoceanic), and storage technologies (salt cavern, pressure tanks, liquefied, ammonia, LOHCs), decide what to build and how much to ship. Multi-period (plans over a decade), multi-product (gaseous H2, liquid H2, ammonia, methanol), multi-mode.
Mathematical formulation
Multi-echelon MILP with siting and flow binaries
Notation
| Symbol | Meaning | Units |
|---|---|---|
| $\mathcal{P}$ | Candidate production sites | — |
| $\mathcal{D}$ | Demand nodes | — |
| $\mathcal{M}$ | Transport modes (pipe, truck, LH2) | — |
| $I_p^{\mathrm{elec}}$ | Electrolyzer capex at site $p$ | $/kW |
| $c_m^{\mathrm{trans}}$ | Transport cost, mode $m$ | $/kg-km |
| $D_d$ | H2 demand at node $d$ | kg/day |
| $x_p$ | Build electrolyzer at $p$ | {0,1} |
| $K_p$ | Electrolyzer capacity | kW |
| $f_{p,d,m}$ | H2 flow, producer $p$ to demand $d$ via $m$ | kg/day |
ObjectiveNPV total cost
Electrolyzer capex + transport infrastructure capex + variable transport + electricity cost.
Constraints
Demand satisfaction:
Production capacity:
Build logic:
Infrastructure: pipeline builds are binary $y_m$ with capacity bounds; storage at each site; multi-period accumulation for investment phasing.
Complexity
Static single-period HSC with ~10 sites, ~20 demand nodes is a few thousand binaries — Gurobi solves in minutes. Multi-period (10-year horizon, 8760 operational hours via representative periods) scales to tens of thousands of binaries; Benders decomposition standard.
Real-world data
IRENA Global Hydrogen Trade
IRENA 2022 provides cost curves for green hydrogen production by country, pipeline and shipping tariffs, and scenario-based trade flows through 2050.
Hydrogen Council + McKinsey Hydrogen Insights
Hydrogen Council publishes annual investment and project pipeline data. Useful for demand forecasts by sector (steel, mobility, ammonia, refining).
Illustrative 5-site network (this page)
5 candidate production sites (good wind/solar), 4 demand nodes (steel plant, refinery, port, mobility hub), 2 transport modes (pipeline, truck). Single-period MILP with NPV cost.
Interactive solver
Production siting + transport mode choice
Scenario parameters
Producer siting + flows
Solution interpretation
The network diagram shows the producer-demand topology: which electrolyzer sites the solver builds (gold nodes), which demand centers each serves, and by which transport mode. Cheap-electricity producers dominate at low hydrogen demand; as demand scales, more sites come online.
Pipeline vs truck: pipelines have huge upfront cost but collapse transport cost per kg; trucks are flexible but expensive per kg-km. Rule of thumb: pipelines pay off above ~50 kt/year flow on a given corridor. The solver's mode choice is highly sensitive to the pipeline capex slider.
The levelized cost of hydrogen (LCOH) at each demand node is the shadow price of constraint (2). Typical range: $3–$8/kg for green H2 in 2024, targeting $1–$2/kg by 2030–2035 per IEA Net Zero.
Extensions & variants
Coupled electricity-H2 operations
Electrolyzers as flexible electricity load, providing grid services while producing H2. Joint optimization of UC/ED and H2 supply.
Multi-period stochastic HSC
Uncertain demand growth and electricity price trajectories. Scenario-based stochastic MILP with scenario reduction.
Ammonia as H2 carrier
Convert H2 → NH3 for easier liquid transport, then optionally reconvert at destination. Opens long-distance shipping routes.
Salt-cavern seasonal storage
Large-volume low-cost H2 storage for seasonal balancing (summer wind surplus → winter heat demand). Integer siting + continuous sizing.
Multi-energy system optimization
HSC as one layer in a broader electricity + heat + gas + H2 multi-energy system.
Carbon-constrained HSC
Net-zero constraint on well-to-wheel emissions. Forces blue H2 off the table above a carbon-price threshold.