This Section outlines the methodology used to optimise the investment decisions and storage trajectories outlined in
Section 3. The objective was to demonstrate alternative pathways for large-scale Hydrogen investments, including long-term storage. An investment model is solved for two study years (2030 and 2040) considering annualised capital costs with an assumed discount rate of 6%. A high level of operational detail facilitates a detailed representation of operational costs, including O&M costs, start-up costs, fuel costs, taxes and penalties. First, a description of the scenarios explored is provided which focus on future pathways where Hydrogen plays a significant role in the drive towards a net-zero energy system. A description of the test system along with the implementation of the investment model in the SpineOpt tool [
14] is provided, along with some more specific modelling details of the Hydrogen conversion processes and long-term storage.
2.1. Scenarios
A total of six scenarios are explored for two future years (2030 and 2040). In addition, a further set of simulations are completed for the six scenarios (2030 only) with a lower level of operational detail, in order to explore the impact and highlight the importance of considering a high level of operational detail in investment models exploring the future role of Hydrogen.
The ENTSO-E TYNDP 2020 [
15] Global Ambition scenario (GA) was chosen as a base scenario for this work. Wind, solar and load time series, as well as Hydrogen demand and fuel and carbon prices are all based on the TYNDP Global Ambition scenario. Generation capacities for technologies which are not included as investment options in this work (e.g., hydro, waste, biomass) are also taken from this scenario. The remaining five scenarios facilitate the exploration of alternative pathways for Hydrogen adoption and the integration of large-scale Hydrogen storage within future energy systems. The six modelled scenarios are described below:
Global Ambition (GA): The Global Ambition scenario is one of three scenarios considered in ENTSO-E TYNDP 2020 and is in line with COP21 targets. In this scenario, there is a focus on a centralised approach to the energy transition. This scenario has been chosen as the base case for this work.
High Fuel Price (HFP): The objective of this scenario is to explore the impact of fossil fuel prices on investment decisions. In the High Fuel Price scenario, the assumed fuel prices, including for “blue Hydrogen” (i.e., Hydrogen from methane reforming combined with carbon capture and storage) are increased by 30% from those assumed in the GA scenario. Carbon prices are increased to 50/100 EUR /tonne in 2030/2040, respectively, compared to 35/80 €/tonne in the GA scenario.
Hydrogen Network (HN): This scenario assesses the impact of increased Hydrogen demand on technology investments. Here, Hydrogen is assumed to meet a portion of the country’s heating demand, in addition to the predominant demand from transport and industrial Hydrogen assumed in the GA scenario. Reassigning natural gas pipelines is one of many options being considered for the bulk transport of Hydrogen [
16]. A portion of the natural gas network is assumed to be reassigned to carry Hydrogen, facilitating a relatively low cost (for the end-user) conversion to Hydrogen space heating. The HN Hydrogen demand time-series includes space and water heating demand for 100,000 dwellings in 2030, and 500,000 by 2040. It is also assumed that heat pump uptake is reduced when an alternative low carbon heating solution is available to householders, with the electricity demand in the HN scenario updated to reflect this.
Technology Breakthrough (TB): In the Technology Breakthrough scenario, the impact of uncertain electrolyser prices and efficiencies are explored. Large cost reductions for electolysers are anticipated in the coming decades, in addition to improved efficiencies driven by the ongoing research and development and anticipated economies of scale. The TB scenario assumes investment costs at the lower end of projections [
17,
18] for both 2030 and 2040 (700/300 EUR /kW, respectively, compared to 1000/600 EUR /kW, respectively, assumed for the GA scenario). Modest efficiency improvements are also assumed.
Variable Renewable Energy (VRE): For “green Hydrogen” to compete with other low carbon solutions and to be adopted at scale, wide-scale infrastructure will be required, as well as low cost and efficient electrolysers. Cost effective “green Hydrogen” will also be reliant on low cost renewable energy. In the Variable Renewable Energy scenario, lower investment costs for both wind and solar generation are assumed, exploring the synergistic relationship between low cost renewable generation and Hydrogen generation and storage.
Restricted CAES (RC): Hydrogen fuelled compressed air energy storage offers a flexible potential investment for future high renewables energy systems, providing valuable peaking capacity and energy storage across different time scales, using both the compressed and stored air, and the stored Hydrogen as a fuel source. However, the locating of large-scale CAES plants is geographically restricted, relying on suitable underground storage (e.g., underground salt formations). In the Restricted CAES scenario, limits are imposed for CAES investments, exploring both the impact on alternative generation and storage solutions and the impact on overall Hydrogen investments when this investment option is limited.
2.2. Test System and Investment Model
The test system used for this case study is based on the All-Island power system of Ireland. The input data [
19], along with the model [
20], are both openly available. Existing power plants which are expected to be still operational are included in the base model. Additional capacities are also included for technologies which are not included as investment options, e.g., waste and biomass plant, which have capacities fixed at the levels assumed for the GA scenario. For variable renewable generation, installed capacities in the base model are based on GA levels. However, additional investments are also possible, allowing total installed capacities to increase, depending on the modelled scenario.
Table 1 shows the capacities included in the base model (before investments are considered).
Table 2 shows all considered investment options and the capacity considered for each investment decision. Note that for energy storage investments, decisions in increments of 1 MWh are considered, and for renewable generation and batteries, investment decisions are made in increments of 1 MW. For plants with more complex efficiency curves, investments are less granular and standard sizing is assumed, with the conventional plant aligning with the ENTSO-E data. With a focus on very large-scale Hydrogen generation, a plant size of 100 MW has been selected for the electrolyser. For the OCGT and CCGT plant, efficiencies and costs are all based on those assumed for the ENTSO-E TYNDP 2020 Global Ambition scenario. Both power and energy capacity investments can be made independently for the batteries, with costs based on [
21]. CCS is modelled as a post combustion carbon capture and storage unit. The plant is represented in Spine as two separate units with independent investment variables and associated annualised costs. The plant performance, in terms of fuel use, electricity output and emissions is captured for all operating points at an hourly resolution using the user constraint (see
Section 2.3). Costs and performance are modelled as per [
22] and plant operation is co-optimised as part of the overall problem, in order to minimise costs, with bypassing of the CCS unit possible. More details are provided for the electrolysers and CAES plant in
Section 2.3.
The investment model is run for 2030 and 2040 (with the 2040 base portfolio updated based on 2030 results and anticipated retirements) using the SpineOpt co-optimised operations and investments model. SpineOpt is an energy system modelling framework, implemented in Julia [
20] and developed specifically for detailed operational and planning studies for future energy systems with high shares of variable renewable generation and complex cross-sectoral interactions. SpineOpt’s generic structure consisting of nodes, units and connections allows SpineOpt models to be extended easily to include any number of sectors, commodities and energy conversion units. The flexible spatial, temporal and stochastic structures allow the model detail to be carefully tailored for each sector and region of interest, ensuring meaningful results while managing the computational burden. SpineOpt is open source and the complete code and documentation is available online [
23].
The SpineOpt modelling framework implements enhanced representative days with ordering and weighting using the SpinePeriods companion model [
24] which allows for the reduction of the model size while capturing arbitrage across the full model optimisation horizon. Each period of the model horizon (which can be flexibly defined by the user, e.g., day or week) is mapped to a corresponding representative period. Most problem variables such as unit flows and unit online statuses exist only for the representative periods, thus reducing the size of the overall optimisation problem. However, the state variables of long term storage nodes exist for every real (non-representative) interval over the full model horizon. For each real interval, the storage state variables interacts with the other problem variables from the corresponding mapped representative intervals. This allows the state of charge of long term storage to be optimised across the full optimisation horizon and co-optimised with short term operations.
The objective function is shown in Equation (
1), which considers investment costs, O&M costs, start-up costs, fuel costs, taxes and penalties which are associated with the slack variables of the demand balance and reserve constraints. The Mixed Integer Programming (MIP) optimisation is solved using CPLEX 12.9 [
25] and an optimality gap of 1%. It should be noted here, that SpineOpt is a flexible modelling framework that allows specification of a wide variety of energy systems in a very flexible way using only nodes, units and connections. The problem formulation in its most general form is presented in the Mathematical Formulation section of the online documentation [
23]. Here we present the formulation of the specific case of the model implemented for this work.
Unit and storage investment decision variables (
,
) are included for all units with a defined
and for all nodes with a defined
, which are included in this model as annualised investment costs with an assumed discount rate of 6%. The total unit and storage investment costs are shown below in Equation (
2) and Equation (
3).
In SpineOpt, the temporal resolution of energy flows, unit online decisions and investment decisions can all be defined independently and can change by look-ahead time using temporal block objects. The investment temporal block has a resolution of 1 year while the remaining decision variables have a resolution of 1 hour, using 12 weighted representative days generated using SpinePeriods [
24]. A further temporal block is used to define the mapping of the non-representative days to a representative day, which allows the trajectory of long-term storages to be considered and the storage investments to be optimised. This will be described in more detail in
Section 2.3.
Time series for demand, wind and solar generation are all taken from ENTSO-E TYNDP 2020 Global Ambition scenario, using the 1984 climate year. Total annual Hydrogen demand is also taken from the Global Ambition scenario, which comes predominantly from the industrial and transport sectors. A weekly profile for the transport related Hydrogen demand based on [
26] is applied to the annual estimate for Hydrogen transport demand for Ireland from the Global Ambition scenario. Unit constraints include minimum generation levels and minimum up and down times and start-up costs are included. System constraints include an inertia floor and primary and tertiary operating reserve requirements. In addition to being met by generating units, demand can also be met by demand side response (DSM) with an assumed variable operation and maintenance (VOM) cost of €500/MWh. 10% of the DSM capacity can also provide system operating reserves. In addition, 2150 MW of DC interconnectors are included −1450 MW to GB and 700 MW to France. GB and France are each represented as a single generating unit with a time varying VOM cost, representing the marginal price, with an average value matching the TYNDP 2020 prices. The VOM varies with net load as per the country specific matched times series for TYNDP 2020 Global Ambition. This allows the flows on the interconnectors to be approximated and reserve provision is also facilitated. Future work will use a full European model to estimate the country specific marginal prices.
2.3. Hydrogen Conversion and Storage
SpineOpt has been designed as a generic energy system modelling framework and it does not assume specific types of energy carriers or sectors. A wide variety of energy systems, technologies and transport physics can be implemented using the fundamental elements of nodes (representing balance, storage and demand), connections (representing transport) and units (representing conversions). Any number of sectors can be included and co-optimised within the model and arbitrary energy conversion units can be added. SpineOpt is a powerful tool when considering a high degree of sector coupling and when modelling emerging technologies, such as electrolysers and Hydrogen-fueled CAES. This Section provides more details of the Hydrogen technologies included in the model. These include electrolysers for the conversion of electricity to Hydrogen, CAES and Hydrogen turbines for electricity generation and both the underground Hydrogen and compressed air storage.
Figure 1 shows a simplified version of the Hydrogen node (labelled “H2”)as implemented in the SpineOpt model. In the diagram the SpineOpt objects of units are in red and nodes are in purple, while the black lines represent the relationships between the various objects on which the various model parameters are defined. The red arrows indicate the direction of flow. The Hydrogen node has an associated time varying demand, with the time series depending on the year and the scenario. Storage can be added to the node in SpineOpt by giving the node a state, and for existing storage, defining a node state capacity
. For nodes with storage investments enabled,
represents the storage capacity per storage investment, which is set at 1 MWh in this model. An importer unit (labelled “Importer_H2”) allows “blue Hydrogen” to be imported to the Hydrogen node based on cost estimates for the relevant years. “Blue Hydrogen” is assumed to be an important transitional fuel while “green Hydrogen” scales up. As such, the importer capacity is sized to meet the Hydrogen demand in the 2030 GA scenario. Additional Hydrogen demand, including increases assumed for 2040 scenarios, must be met by the generation of “green Hydrogen”. The Hydrogen node, and any invested storage capacity at the Hydrogen node, i.e., underground Hydrogen storage, is connected to the electricity node (labelled “ELEC_IE”) via three different unit types: electrolysers, Hydrogen CAES and as an alternative, a Hydrogen gas turbine.
Units in SpineOpt may have any number of input flows from nodes and any number of output flows to nodes. Arbitrary affine constraints can be defined involving any or all of these flows to represent conversion processes. Electricity can be converted to Hydrogen via the electrolyser units which are included as an investment option. PEM electrolysers are assumed and a detailed operational model is included, with minimum and maximum load levels and efficiency which varies with input electrical energy as outlined in [
27]. Here, the electrolyser efficiency curve is approximated using one of SpineOpt’s generic conversion constraints. The
constraint in its simplest form allows a linear relationship to be defined between the outgoing flow and the incoming flow from and to a unit (for the elctrolyser the flow of electricity from the electricity node, and the flow of Hydrogen to the Hydrogen node), using the parameter
. By including the
parameter, the varying efficiency of the electrolyser is captured (see Equation (
4)). In SpineOpt, more complex efficiency curves can also be represented by defining an array of operating points for the unit, facilitating the decomposition of the flow variable in to multiple segments (i.e., incremental heat rates). Full details are provided in the documentation [
23].
where
and
represent the flow and units online variables and the coefficients applied to the variables are the parameters
and
. The indices represent the units (
u), nodes (
n), direction of flow (
d) and time-slice (
t). Equation (
4) is applied to all
unit,
node,
node tuples which have a
defined—i.e., it can also be applied to conventional generating units, such as the Hydrogen gas turbine, defining the relationship between the flow of Hydrogen to the gas turbine and the flow of electricity to the electricity node.
The CAES plant uses similar generic constraints to describe its operation. However, the CAES plant is slightly more complex, and is modelled as 3 separate units connected to 3 different nodes.
Figure 2 shows a simplified implementation of the CAES plant in SpineOpt (for clarity, temporal blocks and stochastic structures are omitted). As per
Figure 1, the units are shown in red, and the nodes are in purple. In addition, the yellow symbol represents user defined constraints. The black and grey lines represent various relationships between the model objects, with the black lines also representing flows, with the arrows indicating the direction of flow. Equation (
4) is applied to the air compressor unit (
in
Figure 2) which defines the relationship between the flow of electricity from the electricity node and the flow of air to the compressed air storage node. On the generation side, the CAES plant is modelled as two additional units,
and
in
Figure 2, and once again, Equation (
4) determines the relationship between the flow of compressed air and Hydrogen from their respective nodes and the flow of electricity from the two units, the combined flow being the power output of the plant. As the two generating units do not operate independently, a further user constraint is applied linking the output of the two units. SpineOpt’s user constraint allows the user to define arbitrary linear constraints involving most of the problem variables. Equation (
5) shows an instance of the generic user constraint with all the relevant parameters, which allows the relationships between the various flows of the CAES generating units to be captured, with the
index representing the user constraints. In summary, the CAES plant is represented in Spine as three different units, with independent investment variables and associated annualised costs, and a compressed air node, which also has an associated investment variable and annualised investment cost. Equation (
4) is used to quantify the efficiency of each unit component of the plant and the user constraint, Equation (
5), links the operation of the CAES generating units (air expander and H2 generator). This methodology allows for a detailed representation of the CAES plant at an hourly resolution in terms of fuel use (both Hydrogen and compressed air) and electricity generation. The state of the compressed air node (i.e., energy content of the compressed air cavern) is also modelled at an hourly resolution and the plant operation is co-optimised as part of the overall investment problem.
As described previously, storage can be added to a node in SpineOpt by giving the node a state, and, for existing storages, defining a node state capacity
. When investments in storage capacity at a given node are enabled,
represents the storage capacity per storage investment, which is set at 1 MWh in this Model. However, optimising investments in long-term storage is challenging in typical investment models, which rely on reduced temporal representations to maintain tractability for very large problems. Long time horizons can be considered at a low resolution (e.g., one year at a daily resolution) which allows requirements for seasonal storage to be captured [
28]. However, such a low resolution does not capture the flexibility needs of systems with high shares of variable generation, for which a high level of temporal detail is essential [
29]. While it is possible to capture the systems short term flexibility needs with suitable selected representative periods, in order to simultaneously capture seasonal storage requirements, more advanced methodologies are required [
30]. In this work, SpinePeriods [
24] generates and orders representative days using an optimisation approach which approximates the annual duration curves [
31]. The remaining, non-representative days are each mapped to a representative day, which is used to model the state of charge of a storage node over the full horizon, allowing the consideration of the arbitrage which takes place between the represented days. Thus an energy trajectory of the storage node can be generated and the energy capacity of the storage node optimised. A cyclic condition for the node state is also enforced, which ensures the node state at the end of the optimisation is at least as high as the initial value at the beginning of the optimisation. In SpineOpt, a map containing a representative day for each day in the horizon is included in a third temporal block (along with the investment temporal block, used for the investment decisions, and the representative days which are used for the operational decisions, see
Section 2.2) which applies appropriate constraints to the energy level of the long term storage node for each day of the year. The full formulation is described in [
30].