High-temperature process heat is among the most challenging industrial end-uses to decarbonize because it combines high thermal power density, tight temperature–time specifications, and product-quality sensitivities that depend on furnace atmosphere and heat-transfer factors. This is particularly true for steel reheating furnaces in hot rolling mills (e.g., walking-beam and pusher furnaces), which consume 2.0~2.4 GJ/t of finished product [3], and where slabs are typically heated to > 1100 °C (often ~1250 °C) with strict uniformity requirements to ensure stable rolling, avoid defects, and minimize yield losses through oxidation-related scaling [4]. Data compiled by the European Commission highlights that this downstream step remains a nontrivial fraction of total CO₂ emissions at around 0.13 tCO₂ per ton of product (cf. 1.29 tCO₂/t for blast furnaces), and it motivates the need to evaluate viable decarbonization alternatives at the process level, not only at the level of upstream steelmaking routes [5].

These high temperatures make it infeasible to decarbonize reheating furnaces through technologies like high-temperature heat pumps (HTHPs), which are commonly proposed in other energy-intensive, but lower-temperature processes [4]. Recent literature converges on two primary pathways for reheating-furnace decarbonization: (i) direct electrification and (ii) fuel substitution with green hydrogen (i.e. hydrogen produced from renewable electricity) [6]. A third pathway, (iii) a closed carbon loop with carbon capture and synthetic natural gas (SNG) (produced from captured CO₂ and green H₂ via methanation) offers process compatibility with existing furnace infrastructure but adds conversion steps and auxiliary units.

Direct electrification. Direct electrification eliminates combustion emissions at the point of use and can be highly efficient, but its application to large, continuous reheating furnaces is subject to important practical constraints. Replacing fuel burners with electric heating removes the flue-gas stream that supports heat recovery and affects convective mixing, meaning that furnace flow and heat-integration concepts may need to be substantially redesigned to maintain productivity and temperature uniformity [4]. Major technologies for direct electrification include resistive heaters, induction coils and plasma torches [7]. In the present study, resistive heating was examined but found unsuitable for the larger furnaces because its heat density is insufficient. Induction heating was also excluded, as it is not compatible with the site-specific products due to spatial constraints and their inhomogeneous shapes. Plasma torches, by contrast, offer a burner-like electrification concept with high power density and a hot gaseous heat-transfer medium, making them a promising option for future applications. However, because they currently remain at a relatively low technology readiness level and still face challenges related to efficiency, heat transfer, NOx formation, product quality, and maintenance, they were not included in the present study either.

Fuel substitution with green hydrogen. Hydrogen (H₂) combustion is often positioned as a relatively straightforward retrofit option because it can retain established burner–furnace architectures. High-temperature process-heat reviews note that steel reheating furnaces can technically be converted to hydrogen and that pilot/industrial deployments are approaching high readiness levels (often framed as TRL ~8–9), while also emphasizing that feasibility hinges on site constraints and available infrastructure [4]. Industrial burner development for reheating furnaces has demonstrated stable operation up to 100% H₂ with low-NO_x concepts at typical furnace temperatures, supporting the practical plausibility of hydrogen retrofits [8].

However, hydrogen shifts the decarbonization challenge upstream: the overall emissions benefit depends on the electricity mix and the availability of renewable power for electrolysis, purchase of externally produced hydrogen (via truck or pipeline), or both. In a detailed hot-rolling reheating-furnace case study comparing natural gas, electrification, hydrogen/air, and hydrogen/oxygen concepts under different European grid mixes, the CO₂ reduction potential was shown to be highly sensitive to the carbon intensity of electricity supply and the pace of grid greening [6]. As a result, hydrogen-based decarbonization cannot be assessed independently of power-system boundary conditions - particularly in contexts where industrial electrification competes for limited renewable electricity [4], [9].

Fuel substitution of hydrogen in place of natural gas causes changes in the reheating furnace atmosphere, with a notable reduction in CO₂ and increase in water vapour (H₂O) [10]. Although industrial-scale tests of hydrogen-combustion atmospheres are still scarce, several laboratory-scale studies have been carried out to investigate the potential effects that the different chemical composition and heat transfer characteristics may have on product quality. Scaling is a major product quality concern in hot rolled products as it causes surface defects, reduces the process yield and complicates downstream surface treatments [11]. A recent study found a moderate increase in scaling in the hydrogen-air atmosphere [12]. An earlier study showed that the differences in scaling are dependent on the specific grade of steel being reheated, however [13]. Although experiments are ongoing, a CO₂-poor atmosphere could also induce decarburization issues, with implications for surface hardness and long-term fatigue [14]. The combination of changes in scaling and decarburization effects implies that a pure hydrogen fuel might not be a one-size-fits-all solution and that other alternatives should be investigated for at least some steel grades.

Fuel substitution with carbon capture and Synthetic Natural Gas. An alternative that requires minimal process modifications is to continue using methane gas (CH₄) as fuel while making the system near-carbon-neutral via CO₂ capture and utilization: CO₂ is captured from the flue gas and methanated with green hydrogen to form SNG, which is recycled as furnace fuel. Pilot demonstrations show that high capture and conversion performance are technically achievable, particularly when heat and material integration are leveraged to reduce auxiliary energy demand [15]. Process-integration studies of power-to-SNG chains further quantify how overall efficiency depends strongly on electrolysis type (steam electrolysis vs co-electrolysis), operating pressure, and heat integration between (exothermic) methanation and heat demands [16].

Nevertheless, these loops add substantial CAPEX and OPEX (capture, methanation and additional hydrogen production), and they incur an efficiency penalty through additional conversion steps. In a dedicated MILP optimization of a steel hot rolling mill, SNG produced via CO₂ capture and methanation was consistently more expensive than direct hydrogen across electricity-price scenarios, even though SNG offers storage and infrastructure compatibility advantages [17]. Still, its high compatibility with existing production processes makes it a worthwhile option to consider in feasibility studies.

Because technology choice depends on discrete design decisions and time-dependent operation, recent work increasingly uses optimization-based approaches to compare decarbonization pathways under realistic constraints.

Mixed-integer linear programming (MILP) enables joint optimization of (i) discrete technology selection, (ii) capacity sizing, and (iii) time-resolved dispatch under variable production demands, energy prices and infrastructure constraints. In the hot rolling context, a recent MILP study comparing on-site hydrogen and on-site SNG under dynamic electricity pricing demonstrates that exploiting price variability via flexible operation and storage can substantially reduce levelized costs of energy (LCoE) relative to flat-price assumptions; it also shows that optimal electrolyser selection shifts with electricity-price level due to efficiency–CAPEX trade-offs [17].

Studies from other sectors that also require high-temperature process heat show that deep decarbonization is often achieved via hybrid solutions, and that the last step toward near-zero emissions can be disproportionately expensive. In glass-furnace techno-economic optimization, partial electrification combined with on-site renewables and storage can achieve moderate abatement at modest cost increases, whereas pushing toward very high abatement levels requires substantially larger investments in storage or hydrogen infrastructure [18]. Multi-stage MILP optimization in cement similarly indicates that combining multiple measures (e.g., alternative energy supply, capture, and integration) can deliver major emissions reductions while highlighting the importance of phased investments and policy incentives for economic viability [19]. Cluster-level MILP studies in high-temperature chemical sectors show that the availability of CO₂ transport/storage and power infrastructure can dominate outcomes and that integration across units can reduce costs and emissions compared to isolated decarbonization measures [20]. Collectively, these findings support the expectation that hot rolling mills will benefit from evaluating portfolios of decarbonization options and from explicitly representing time-varying operation and infrastructure constraints, in order to best quantify the business impact of deep decarbonization targets.

Much of the peer-reviewed literature for decarbonization in the iron and steel sector is dedicated to the production of raw steel, while the studies on downstream processing of steel products remain comparatively sparse. The most relevant techno-economic assessment study is the doctoral dissertation by Löffler [21], recently expanded upon in a paper by Zabik, Birkelbach et al. [17]. They apply a MILP methodology to optimize the technology choice, capacity sizing and operation of a decarbonized energy supply system for a hot rolling mill in Austria, and carry out a detailed comparison of the LCoE of a hydrogen- vs SNG-based system under a variety of different energy price scenarios. There are however three additional elements not explored in their case study, and which our present work includes to provide a fuller understanding. These are namely different electrical grid capacity scenarios, which influence the planning of major investments; inclusion of onsite PV and wind power generation in the mix; explicit modelling of carbon capture subsystems; and hybrid solutions combining the advantages of more than one technology.

On the one hand, the strong increase in electricity demand expected from onsite production of green hydrogen calls for an evaluation of different grid connection regimes, especially given the structural bottlenecks that the Austrian power grid faces in the near future [22]. If a substantial part of the additional demand is satisfied through onsite renewables, grid feed-in dynamic of excess power in times of high generation and/or low process demand also need to be considered. The development of plans for investments in the grid is in turn influenced by the demands expected from electricity consumers in general and energy-intensive industries in particular. Consequently, the present work can provide insights into the interrelation of industrial planning and grid conditions and thus be helpful to decisionmakers on both sides.

On the other hand, the previous study only considers electricity supply from the general grid. The present work also introduces the possibility of building solar PV and wind power generation connected directly to the plant, which potentially decreases dependency on the grid but also requires consideration of seasonal variations in generation potential which do not necessarily correlate to variations in process heat demand. The optimal electrical mix for each scenario is calculated trough a representative period-based approach.

In addition, their study simplifies the carbon looping subsystem as a per-unit carbon capture cost; an explicit modelling of the available technologies for carbon capture would enable an optimization of the design of the carbon capture subsystem as well as of the heat integration between its energy heat demands and the waste heat from the furnaces and the methanation reaction on the other.

Finally, their model presents a conventional binary choice between electrolysis technologies, while this work allows for more general (albeit potentially less feasible in practice) hybrid solutions in the electrolysis, carbon capture and methanation subsystems that implement more than one technology in order to take advantage of each one’s relative strengths.

By examining the optimal decarbonized system under different grid connection regimes, explicitly modelling and optimizing the closed carbon loop subsystem and allowing for multi-technology solutions while simplifying the energy price conditions, the present work aims to complement previous case studies and provide a fuller picture of the challenges and opportunities the steel processing industry faces on its road to the net-zero economy.

In the reheating furnace context, post-combustion carbon capture consists generally of collecting the flue gas of furnaces and passing it through a device that separates the CO₂ from the rest of the flue gas, so that it can be either stored away or utilized in some other process. In this case, the purpose of the CC is to produce SNG that is burned again in the furnaces, in a quasi-closed loop. The carbon capture process requires electricity and/or heat to separate the CO₂ from the rest of the species in the flue gas, with the specific values varying by technology. Assuming constant efficiency over the operating range, the linearised input-output relations for a CC unit with recovery rate RCC, power demand dpCC and heat demand dqCC are:

where the subindices "in", "cap" and "slip" refer to the mass flows of CO₂ that comes in with the flue gas, that is captured, and that slips from the CC unit without being captured, respectively.

A multitude of different CC technologies exist, divided into several families and at different technology readiness levels (TRL) [41]. In this work, two possibilities from the absorption family and one from the adsorption family are considered.

Amine Scrubber (AS). This method works by passing the flue gas through an amine bath in a continuous process. The amine is then regenerated and the CO₂ stored. The process has a low power demand but higher heat requirements, and is currently at a very high TRL [41]. The investment costs are taken to be cinvAS= 225 kEUR/(t/h), based on quotes from suppliers.

Amine Scrubber with Heat Pump (ASHP). The AS process can be made more efficient by integrating a heat pump for internal heat recovery. This eliminates the need for external heat input and replaces it with a smaller electricity input, at the cost of higher installation costs [42]. Those costs cinvASHP= 1.697 kEUR/(t/h) are based on the costs for the AS itself plus the cost premium for the heat pump provided in the aforementioned paper.

Pressure Swing Adsorption (PSA). This method separates CO₂ by feeding flue gas into a pressurized tank and then greatly reducing the pressure of the tank, which causes the CO₂ to be adsorbed into a purposefully designed solid coating (typically zeolite-based) while the other components of the gas remain free. This batch process requires no heat input but instead is powered by electricity, and although its TRL for CO₂ capture is lower, it is considered in the literature to promise low operation and investment costs [41], [43]. As exact costs figures for this type of system are hard to find in the literature, an estimate of cinvPSA = 1,111 kEUR/(t/h) has been calculated from two studies that provide partial cost figures [44], [45].

As the CC units should ideally be located as close as possible to the source of the flue gas, the system is envisioned as two separate parts: a choice of AS, ASHP and/or PSA for the group of furnaces A/B and another for the pair of furnaces C, with access to recovered waste heat shared between all. This is illustrated on the right side of Figure 1. If the nominal capacity of each unit in terms of captured mass per unit of time is ṁCO2,peakCC, then the associated investment costs are:

Hydrogen (H₂) and carbon dioxide (CO₂) can react exothermically to produce methane (CH₄). In the ideal reaction, the so-called Sabatier process can be summarised [47] in the following stoichiometric formula:

In a methanation reactor, this reaction can be carried out in a controlled manner to achieve specific compositions and conversion rates, depending on the desired properties of the resulting gas mixture. The fraction of residual carbon dioxide and hydrogen left in the resulting gas depend on the actual stochiometric ratios in the initial mixture and on the respective conversion rates χCO2meth and χH2meth. In the linearized model it is assumed that the reactor is operated at a fixed stochiometric ratio of feedstocks and thus at fixed conversion ratios and waste heat generation Then the linearized input-output relations for the reactor model, relating the CO₂ and H₂ inputs to the CH₄, waste heat, CO₂ and H₂ outputs are, respectively:

One-Stage Methanation (OSM). In a conventional methanation reactor, not all the CO₂ is converted into methane; in addition to a large concentration of CH₄, also some amount of residual H₂ and CO₂ are present. This changes the LHV and flame properties of the fuel with respect to the near-pure methane of natural gas; consequently in this work said mixture is referred to as low-grade synthetic natural gas (lgSNG). All the parameters for this type of reactor were collected from an industry supplier source.

Two-Stage Methanation (TSM). By cooling the lgSNG and passing it through another methanation reactor, a much higher purity of methane can be achieved, making it effectively identical to fossil-sourced natural gas, i.e. [48]. All the parameters for this type of reactor were collected from an industry supplier source as well.

The investment costs associated with the methanation reactor(s) are:

In case of load and or/cost variations in the energy supply (due to the intermittency of renewable sources) or in the energy demand (due to production requirements), energy storage can provide flexibility by smoothing out those fluctuations. In this work two possibilities have been considered: electro-chemical storage in batteries and chemical storage in H₂ tanks. Both are modelled using a linearized approach with a charging/discharging efficiency parameter η and a standing state-of-charge loss λ.

If each storage has a maximum state-of-charge (SoC) capacity SoCpeak , the investment cost for storages is:

Charging/discharging losses for the battery storage were (somewhat optimistically) considered negligible for the purposes of this study, but auxiliary losses were accounted for at 1%/h as a typical value for multi-MWh-scale, fast-cycling storages [49].The investment cost was set at cinvbat =100 kEUR/MWh, in line with European Commission expectations of battery packs falling below 100 USD/kWh in 2027 [50]. For the hydrogen tank storage the round-trip efficiency was set at 80% as an intermediate point between the 60~70% range of liquefied storage and the (pressure-dependent) 80~95% range of compressed storage [51], the auxiliary demand equal to the battery, and the investment cost assumed to be a higher cinvH2 tank = 400 kEUR/MWh, with these less optimistic parametrization likely favoring batteries.

Bringing together eqs. (6), (7), (8) and (9) yields the total annual OPEX:

while the total CAPEX are obtained from eqs. (13), (17), (20), (24) and (25):

bringing the total annualized costs to:

with an interest rate i =3% and amortization period n =20 years yielding a CAPEX annualization factor of 6.7%.

It is expected that the grid will be strained in the future due to increased demand both in the plant and in the broader region as both the steel industry and other sectors electrify their processes. Although quantitative forecasts of future grid availability are outside the scope of this work, it was deemed relevant to establish three exploratory scenarios regarding the grid connection regime. The three scenarios respectively represent optimistic, baseline and pessimistic possibilities for the medium-term development of the power grid in the region and in Austria at large.

Unconstrained grid regime. With a peak capacity of around 7.2 MW/MW_q, it represents a scenario where all the investments in grid capacity are realized and the mill can draw, in practice, as much power as it needs from the grid. This boundary condition potentially tilts the optimal solution towards a more interconnected system.

Partially constrained grid regime. With a 0.72 MW/MW_q peak capacity, this roughly corresponds to the current situation before decarbonization and electrification measures are undertaken.

Strongly constrained grid regime. With only 0.24 MW/MW_q of peak capacity available, which equals around one-third of today’s maximum theoretical capacity, it represents a pessimistic scenario in which grid expansion investments to alleviate structural bottlenecks lag behind industry decarbonization commitments and thus place limits on the mill’s ability to draw electrical power from the grid, potentially pushing the optimal solution towards more electricity-efficient technologies and/or a more self-contained system with expanded onsite renewable generation.

Figure 2 shows a breakdown of the levelized costs of heat (LCoH), in terms of EUR/MWh of product heat. The most apparent result is that in all scenarios, switching to a decarbonized energy system incurs a stark increase in the LCoH. If hydrogen can be burned directly in the furnaces, the LCoH is shown to increase by between 124% (if the electrical grid is unconstrained) and 166% (if it is strongly constrained). If the CC+SNG subsystem is required, the costs increase further: by 187% in the unconstrained grid (28% more expensive than only hydrogen) and by 231% in the strongly constrained grid (24% more expensive), a more than three-fold increase. There are two main factors for this difference: on the one hand, the additional CAPEX and OPEX associated with the CC+SNG subsystem, which are not present in the pure-H₂ path; on the other, the additional CAPEX and OPEX induced in the electricity and hydrogen supplies, due to the conversion losses in the methanation step and to the lower combustion efficiency of SNG relative to hydrogen. The results are consistent with the lower end of the 25~43% bracket that Zabik, Birkelbach et al. found for the SNG cost premium [17].

Figure 2.

Waterfall plot for the cost breakdown in both the H2 and the SNG path, including expenses (green), incomes (red) and intermediate totals (blue). For each trio of bars, the top series is for the unconstrained grid case; the middle one, for the partially constrained grid; and the one in the bottom, for the strongly constrained grid. The dashed line marks the BAU case cost of 109.80 EUR/MWh, consisting of 73.80 EUR/MWh of natural gas costs and 36.00 EUR/MWh of emissions certificates.

Upon a closer look at the individual grid scenarios, some interesting differences arise, which can be explained by the Sankey diagrams of energy flows in Figs. Figure 3 and Figure 4:

Figure 3.

Sankey diagrams of the optimal systems for the H2 route (a: unconstrained grid; b: partially constrained grid; c: strongly constrained grid). All quantities are peak capacities per MW of product heat; the quoted percentages are each unit's average load factor. The diagrams are not to scale with each other; the heat demand is the same in all cases.

Figure 4.

Sankey diagrams of the optimal systems for the CC+SNG route (top: unconstrained grid; middle: partially constrained grid; bottom: strongly constrained grid). All quantities are peak capacities per MW of product heat; the quoted percentages are each unit 's average load factor. CO₂ slip and emissions compensation from biomethane are not shown. In the CC units, the upper row is for A/B and the lower row for C.

Unconstrained grid. In the unconstrained grid scenario (top bar of each trio in Figure 2), most of the electricity is supplied directly from the grid, with only a minor contribution from wind power, and lesser amounts from hydro and PV, with no battery storage required. Hydrogen is produced largely onsite with a smaller contribution from grid purchases.

Partially constrained grid. In the partially constrained grid scenario (middle bar), around half of the electricity demand is satisfied by the grid and the other half by renewables (mostly wind) and some battery storage. However, the total electricity consumption is low, because the optimal solution involves purchasing most hydrogen from the grid.

Strongly constrained grid. In the scenario where the grid is most strongly constrained (bottom bar), it can only supply part of the necessary electrical power, so a large capacity of wind power is installed and there is also substantial buying of hydrogen from the grid. Flexibility to offset the intermittency of wind generation is provided through three mechanisms: exporting excess electricity to the grid, charging or discharging the battery storage depending on the supply and demand at each time step, and producing excess hydrogen and exporting it.

When different options are available whose OPEX and CAPEX form a Pareto front, optimization can show which of them, or which combination of them, offers the lowest total cost under a set of boundary conditions. The optimal operating patterns optimized simultaneously with the capacity decisions shown in Figure 3 and Figure 4 are highlighted in the following.

Electrolysis subsystem. Figure 5 shows the operation of the electrolysis subsystem during one of the 24-hour representative periods (specifically, October 10th). It is apparent from the plots that, regardless of the decarbonization route and grid capacity the three methods of supplying green H₂ fulfil specific roles. First, the high-CAPEX, low-OPEX SOEC acts as a base load supply, operating at a near-constant, very high load factor which takes advantage of its high efficiency to offset the larger investment cost. Second, the low-CAPEX, high-OPEX AWE/PEM follows the demand profile more closely whenever abundant wind power is available (typically at night), taking advantage of its more flexible operation possibilities. Finally, the option to directly purchase green hydrogen from the grid is used for peak-of-the-peak supply, thus avoiding extra electrolyser capacity that would otherwise go largely unused; it is also used to fulfil variable loads when the availability of wind power is low, particularly in the partially constrained grid scenario.

Figure 5.

Optimal operation of the electrolysis subsystem for both routes and under the three grid scenarios, during one of the representative days.

Looking more closely at the three grid scenarios, some extra insights can be gained. When the grid is unconstrained, the electrolyser capacity is dominated by AWE/PEM in the H₂ route and by SOEC in the CC+SNG route. This is due to the highly exothermal nature of the methanation process increasing the availability of waste heat, which can be reused in the SOEC unit to further lower its operational costs. Meanwhile, in the strongly constrained scenario, the intermittency of renewable generation causes the AWE/PEM to dominate the optimal solution - a higher capacity of SOEC would require a higher base load supply of electricity, in turn necessitating additional investment in battery storage. The role of storage is in any case quite limited, in line with the static price scenarios from Zabik, Birkelbach et al., and would presumably be more relevant if fluctuating prices were introduced into the model [17].

Figure 6.

Optimal operation of the carbon capture subsystem under the three grid scenarios, during the same representative day.

CC subsystem. In the carbon capture subsystem, meanwhile, the optimal operation is strongly related with the availability of electricity. In the unconstrained grid scenario, the PSA (using abundant electricity) is used to fulfil the baseline demand, while the AS (using waste heat for which it competes with the SOEC) is activated as a backup when the load of the furnaces is high. In the two constrained grid scenarios, on the other hand, the PSA provides a small, uniform capacity while the AS handles most of the carbon capture demand. In these two latter scenarios, due to the lower capture rate, a larger carbon slip is present which must be offset by a larger purchase of biomethane. These results again highlight the crucial role that the grid conditions have on the optimal decarbonization strategy.

In scenarios with unconstrained grid access, the model favours technologies with higher investment cost but lower operating cost and/or higher efficiency, because sufficient electricity is available to utilize them at high load factors. This leads to steadier operation of units such as SOEC and PSA. By contrast, under constrained grid conditions the model shifts toward more flexible technologies and supply options, including AWE/PEM, hydrogen imports, wind power and storage, because electricity scarcity increases the value of operational flexibility. Thus, the optimal design and the optimal operation should be interpreted jointly: the installed capacities reflect the expected operating regime, while the operating profiles explain why those capacities are economically preferred.

Parameter Variations

It follows from Figure 2 that the CAPEX of wind power, the cost of purchasing green hydrogen from the gas grid and the CAPEX of the electrolysers are some of the largest contributors to the total energy costs. However, there is considerable uncertainty in the values of those parameters, which may cause significant deviations in the actual energy costs. Therefore, a sensitivity analysis was conducted to study if and to what extent the possible changes in those parameters affect the LCoH.

Wind CAPEX. The 60 EUR/MWh LCoE assumed for the onshore wind production is realistic but somewhat on the optimistic side, for two reasons. On the one hand, one could expect a "low-hanging fruit" effect by which the best locations for wind parks are built out first, and future expansions will have a lower average load factor and thus higher LCoE. On the other hand, transmission lines required to connect new wind parks directly to the steel mill could have an extra cost. Therefore, the cost variations studied are 1×, 1.25× and 1.5× the baseline value of cinvwind.

Hydrogen purchase price. The 150 EUR/MWh assumed as baseline, corresponding to about 4.5 EUR/kg, is within the lower range of current prices for renewable H₂ currently, but it is expected that prices will go down in the future thanks to economies of scale and government subsidies (the "hydrogen economy"). Thus, the studied variations are 1×, 0.8× and 0.6× the baseline cost cbuyH2.

Electrolysers CAPEX. The investment costs for electrolysis systems have uncertainties in both directions. On the one hand, it is likely that technological innovations and economies of scale will bring down the production cost of electrolysis cells. However, it is also possible that ancillary infrastructure and adaption of gas supply systems in the furnaces (for the H₂) will induce additional investment costs. To account for all the possibilities, the studied variations are 0.5×, 0.75×, 1×, 1.5× and 2× the baseline cinvAWE/PEM and cinvSOEC.

Considering all the possible combinations of parameter values, decarbonization routes and grid capacities, a total of 270 separate optimization runs were evaluated. It should be noted that, although the parameter variations were treated independently in this study, some of these varied parameters may be correlated in practice. In particular, the cost of grid-purchased hydrogen can partly be shaped by upstream electricity prices, electrolyser investment costs and wind farm CAPEX. In the present study, these parameters were varied independently in order to isolate their directional influence on the optimal solution. The main results, again using LCoH as the main metric, are shown in Figure 7.

Figure 7.

Distribution of LCoH generated by the 45 variations, for both routes (H₂ in blue, CC+SNG in orange). Each dots represents an individual variation. The dotted lines mark the respective baseline cost.

The most obvious result is that, for the price of purchasing hydrogen from the grid, there is a threshold under which the optimal solution consists of forgoing any on-site electrolysis capacity simply supplying all the H₂ through the grid, regardless of the values of the other cost parameters. The exception is the strongly constrained grid scenario, where some electrolysis capacity is still required to process excess renewable energy.

The variations in wind power CAPEX do not have a large influence on the LCoH, as they induce a reduction in wind power capacity and an increased electricity purchase from the grid, again with the exception of the strongly constrained grid. In that case, higher investment costs translate into a higher LCoH.

Finally, the variations in the specific investment cost of electrolysers seem not to have a large impact on average, but cause a larger spread in the distribution of the LCoH generated by the variations of the other parameters. However, higher investment costs do translate to an increase in LCoH while they are still below the grid purchase costs. This can be observed especially in the case of an unconstrained electricity grid connection, where increasing electrolyser costs are directly linked to an increase in LCoH.

The sensitivity analysis reveals that the response of the system is not purely gradual, but exhibits threshold-like behaviour. Most importantly, the attractiveness of on-site hydrogen production depends on the interaction between hydrogen purchase price, grid constraint and costs for renewable electricity. When hydrogen from the grid is sufficiently cheap, the model tends to forgo on-site electrolysis in favour of external supply. However, under strongly constrained grid conditions, on-site electrolysis remains valuable even at low hydrogen purchase prices because it provides a sink for excess renewable electricity. Hence, grid constraints determine whether the system shifts towards a purchase-based or production-based hydrogen supply strategy.

In this work, a model of a steel hot rolling mill in Austria was developed, with (i) direct combustion of green hydrogen and (ii) post-combustion carbon capture with methanation as the possible routes for decarbonizing its energy supply, and several technologies available for each subsystem. The design and operation of the energy system were simultaneously optimized using a mixed integer linear problem (MILP) formulation. Even after optimizing for minimal annualized costs, a major increase in the levelized cost of heat (LCoH) was predicted.

However, the increase in LCoH was found to be noticeably dependent on the constraints of the system. If direct combustion of green hydrogen is allowed, the cost is predicted to increase to 2.5 - 2.9 times the BAU LCoH; if synthetic natural gas (SNG) is required, then the increase is 3.2 - 3.6 times the BAU LCoH, mainly due to the extra investment costs and conversion losses of the carbon capture and methanation steps. The strictness of the limitations on the electrical grid to provide flexibility not only has a large impact on the headline LCoH, but also on the underlying configuration of the energy system.

Finally, the parameter variation analysis showed that the capital costs of wind power generation and the purchase costs of hydrogen from the gas grid also have a large influence on the LCoH and the configuration of the system, with the investment costs of electrolysers having a comparatively minor influence. Together, the results highlight the importance of accurate modelling of the boundary conditions when predicting the costs of the energy transition, but show that optimization methods can help in revealing the cost-optimal strategies for minimizing said costs.

From a business perspective, the absolute LCoH levels suggest that, under the assumptions of this study, neither route is likely to be adopted on the basis of direct cost competitiveness alone. Rather, deployment would depend on favourable infrastructure conditions together with policy support and/or market mechanisms that reward low-carbon steel production.

The insights obtained from the present work suggest a range of refinements of the model that could prove to be interesting to investigate in future research. On the one hand, (i) the outsize impact of grid boundary conditions on the design and operation of the system calls for a more detailed modelling of the grid dynamics and the electricity and hydrogen markets, including time-dependent prices and capacities, with price fluctuations in particular having a potentially major effect on the sizing and operation of energy storages. On the other hand, (ii) more detailed modelling of operational conditions of the technologies involved in the electrolysis, carbon capture and methanation subsystems could prove useful in the context of broader feasibility studies, with more emphasis on the efficiency changes over the operating range of each technology, as well as more realistic models of energy storage technologies. Finally, the model could be expanded with (iii) additional alternative fuel possibilities (ammonia, methanol, enrichment of combustion air with oxygen produced as by-product of electrolysis) as well as with (iv) the search for optimal decarbonization pathways, i.e. intermediate solutions with partial decarbonization steps and the associated optimization of the timing of investment decisions.