Endogenous learning for green hydrogen in a sector-coupled energy model for Europe

Abstract: Many studies have shown that hydrogen could play a large role in the energy transition for hard-to-electrify sectors, but previous modelling has not included the necessary features to assess its role. They have either left out important sectors of hydrogen demand, ignored the temporal variability in the system or neglected the dynamics of learning effects. We address these limitations and consider learning-by-doing for the full green hydrogen production chain with different climate targets in a detailed European sector-coupled model. Here, we show that in the next 10 years a faster scale-up of electrolysis and renewable capacities than envisaged by the EU in the REPowerEU Plan is cost-optimal in order to reach the +1.5{\deg}C target. This reduces the costs for hydrogen production to 1.26 Eur/kg by 2050. Hydrogen production switches from grey to green hydrogen, omitting the option of blue hydrogen. If electrolysis costs are modelled without dynamic learning-by-doing, then the electrolysis scale-up is significantly delayed, while total system costs are overestimated by up to 13% and the levelised cost of hydrogen is overestimated by 67%.

• The paper demonstrates that endogenous learning in a sector-coupled model can reduce green hydrogen costs significantly, reaching as low as 1.26 €/kg by 2050.
• The paper employs a high-resolution MILP framework with piecewise linear approximations to optimize investments in renewable and electrolyzer technologies across sectors.
• The paper highlights that dynamic, learning-by-doing methods lead to earlier investments and lower system costs compared to exogenous cost projection methods.

This paper, "Endogenous learning for green hydrogen in a sector-coupled energy model for Europe" (2205.11901), investigates the role of green hydrogen in the European energy transition by using a detailed sector-coupled energy system model that incorporates endogenous learning-by-doing for key technologies like hydrogen electrolysis and renewable energy generators. It addresses limitations of previous models which often neglect sector coupling, temporal variability, or the dynamic cost reduction effects of technological learning.

The authors utilize the open-source PyPSA-Eur-Sec model, a techno-economic energy system model that minimizes total system costs over a defined time horizon (2020-2050 in this study). The model optimizes investment in and dispatch of generation, storage, and transmission capacity across the electricity, heating, transport, and industry sectors. A key feature is the high temporal resolution, using 10 typical days per year, which is crucial for capturing the variability of renewable generation and optimizing the operation of flexible technologies like electrolyzers.

A core practical contribution of the paper is the implementation of endogenous learning-by-doing for solar PV, onshore wind, offshore wind (global learning proportional to European capacity), and hydrogen electrolysis (local European learning). This is modeled using experience curves, where the specific investment cost $c$ of a technology decreases as its cumulative installed capacity $E$ increases, following the relationship:

$c(E) = \overline{c_0} \cdot \Big(\frac{E}{\overline{E_0}\Big)^{-\alpha}$

where $\overline{c_0}$ and $\overline{E_0}$ are initial cost and experience, and $\alpha$ is the learning index derived from the learning rate $LR$ ($\alpha = \log_2(1/(1-LR))$).

Implementing endogenous learning in an optimization model results in a non-linear, non-convex problem. To make it solvable by commercial solvers, the authors use a piecewise linear approximation of the cumulative cost curve. This is achieved by defining a set of interpolation points and using Special Ordered Sets of Type 2 (SOS2) variables. The model becomes a Mixed Integer Linear Problem (\gls{milp}), which can be solved efficiently using solvers like Gurobi, albeit with higher computational requirements than purely linear models. The implementation includes a temporal delay, where cost reductions in an investment period depend on cumulative capacity from the previous period, reflecting a more realistic lag in learning effects.

The model considers three competing hydrogen production pathways: grey (\gls{SMR}), blue (\gls{SMR} with CCS), and green (electrolysis). Hydrogen can be used in various sectors, with demand being partly exogenous (e.g., fixed shares in industry, transport) and partly endogenous, where hydrogen competes with other options (e.g., heat pumps vs. hydrogen boilers in heating). The optimization determines the cost-optimal mix of production technologies and hydrogen usage across sectors under different total CO₂ emission budgets for Europe (corresponding to +1.5°C, +1.7°C, and +2°C global warming targets, with carbon neutrality by 2050).

Key Findings and Practical Implications:

1. Green Hydrogen Cost Reduction: The model predicts significant cost reductions for green hydrogen production, reaching as low as 1.26 €/kg by 2050 in scenarios aligning with ambitious climate targets (+1.5°C). This is lower than many conventional exogenous projections and underscores the potential impact of large-scale deployment and learning.
2. Required Scale-Up: Achieving ambitious climate targets necessitates extremely rapid scale-up of both electrolysis and renewable generation capacity. For a +1.5°C scenario, the model shows cost-optimal annual build-out rates for solar and wind that are several times higher than historical maximums, highlighting the significant practical challenge in manufacturing, infrastructure, and permitting. The EU's REPowerEU targets for wind capacity by 2030 are found to be lower than what the model suggests is cost-optimal for achieving even the +2°C target, while PV targets align better with the +1.7°C to +2°C range.
3. Hydrogen Production Shift: The cost-optimal pathway involves a strong shift from grey hydrogen (SMR) to green hydrogen (electrolysis). Under the base assumptions, blue hydrogen (SMR+CCS) is not deployed at scale. However, sensitivity analyses show blue hydrogen could become competitive under optimistic conditions like significantly lower investment costs for CCS or larger CO₂ storage potential. This is a crucial finding for investors and policymakers debating the future role of blue hydrogen.
4. Hydrogen Demand and Use: Hydrogen demand increases significantly in ambitious scenarios, primarily driven by the production of synthetic fuels and synthetic methane for hard-to-abate sectors like transport and industry. Other uses include fuel cells and shipping. Technologies like retrofitting natural gas boilers for hydrogen are found to be less cost-optimal compared to alternatives like heat pumps.
5. Importance of Endogenous Learning: Comparing the endogenous learning method with simpler exogenous or sequential methods reveals significant differences. The endogenous method, which has foresight of future cost reductions through learning, leads to earlier investments in emerging technologies, resulting in lower total system costs (up to 13% lower) and lower levelized cost of hydrogen (up to 67% lower by 2030) compared to the exogenous method. This demonstrates that ignoring learning effects can lead to delayed investments and an overestimation of transition costs.
6. Modeling Method Trade-offs: The paper discusses the trade-offs between modeling methods. The endogenous MILP offers superior representation of investment timing and cost dynamics of learning technologies but is computationally expensive. Exogenous methods are computationally cheap but rely heavily on accurate (and uncertain) external cost projections. The sequential method provides a middle ground, updating costs based on realized capacity iteratively, offering lower computational cost than endogenous while still linking capacity to cost. The choice of method depends on the research question, available computational resources, and the expected nature of learning (local vs. global).

Implementation Considerations and Limitations:

• Computational Resources: The endogenous MILP is computationally intensive, requiring significant memory and processing time (e.g., 21 hours and 30 GB RAM for a single scenario run with 12 threads) compared to linear methods (minutes/hours and 3 GB RAM). Spatial and temporal aggregation are necessary for tractability.
• Spatial Resolution: The use of a single European node for the energy network simplifies transmission effects. A sensitivity analysis with higher spatial resolution (for the exogenous method) shows increased total system costs and higher electrolysis capacities due to grid constraints, indicating that the main results might underestimate grid infrastructure costs and overestimate the ease of deploying electrolysis anywhere.
• Assumptions: Key assumptions like no hydrogen imports, exogenous transport sector pathways (especially the share of ICE vehicles influencing early synthetic fuel demand), and limiting learning only to investment costs (not O&M, efficiency, lifetime) influence the results. The assumption of local learning for electrolysis might underestimate cost reductions if global deployment is faster.
• Feasibility of Build Rates: The extremely high build rates identified as cost-optimal for ambitious targets might be practically infeasible due to supply chain, labor, or permitting bottlenecks, which are not explicitly modeled as hard constraints. This could potentially shift the cost-optimal pathway towards other options or slower transitions.

In summary, the paper provides valuable insights for the practical implementation of green hydrogen strategies by demonstrating the significant potential for cost reduction through learning and highlighting the rapid scale-up required for ambitious climate targets. It emphasizes the importance of dynamic modeling of technological learning to avoid underestimating early investments and overestimating transition costs, offering guidance on selecting appropriate modeling methods based on available resources and study objectives. The findings suggest that current EU targets for renewables deployment might need to be more ambitious to support the required green hydrogen production volumes.