Power System Decarbonization: Impacts of Energy Storage Duration and Interannual Renewables Variability

1. Introduction

• Long-duration storage remains an open economic question: its viability and added system value are unclear.Prior work links storage value to its size, duration, costs, and CO2 constraints.Source: §1 · pp. 1–2
• Single-year planning can produce suboptimal operation, unserved demand, and higher costs in other years, while VRE and load patterns vary annually.Prior studies show that VRE variation can significantly change system costs, but zero-carbon configurations and the effects on all generation technologies remain less explored.Source: §1 · p. 2
• The study evaluates Italy’s 2050 least-cost system under different emissions limits, storage options, and 11 years of VRE and load data.It uses integrated investment and operation optimization and compares multi-year planning with decisions based on single-year data.Source: §1 · p. 2
• The analysis also examines flexibility and low-emission technologies, storage durations up to 100 hours, and uncertainty effects on system configuration, costs, reliability, and operating costs.The authors identify these as the study’s main contributions.Source: §1 · p. 2

2. Methodology

• GenX is modified to co-optimize generation investment and operation for the 2050 system within one planning model.The optimization includes capacity expansion, hourly unit commitment and dispatch, and reserve scheduling.Source: §2 · p. 2
• The model imposes technology investment and operating limits, unit commitment and reserve constraints, annual CO2 caps, hourly energy balance, and system reserve requirements.Hourly supply must balance load unless load curtailment occurs; constraints (b)–(g) apply to each year in the stochastic formulation.Source: §2 · p. 3
• Multi-year planning minimizes expected demand-meeting costs across weather and load years, while single-year investments are also tested against other realized years.The objective incorporates multi-year fixed and variable costs, with weights normalizing costs to an annual value.Source: §2 · p. 3
• The mixed-integer formulation is relaxed to a linear program to speed computation, an approximation earlier studies found to have limited impact on expansion results.The planning model requires assumptions about generation, load, renewables, storage, technology costs, and fuel prices.Source: §2 · p. 3

3. Input Data and Assumptions

• The study builds on prior Italy analysis that projected renewables could exceed 90% penetration by 2050 but did not account for interannual VRE variability.Current RES generation is reported as 35–40%, with variation related to water availability and hydro generation.Source: §3 · p. 3
• For 2050, coal and oil expansion are excluded, while CCGT, OCGT, and CCGT with post-combustion CCS are available as thermal options.CCS is assumed to capture 88% of CO2; biofuel generation is treated as carbon-emissions-free, and thermal-unit outages and maintenance downtime are not modeled.Source: 3-1. Generation Units · p. 4
• Renewable inputs include solar, wind, hydro, geothermal, and bioenergy, with hydro, geothermal, and co-generation bioenergy expansion constrained or fixed.Hydro is limited to about 22 GW and 45 TWh annually; run-of-river hydro has constant hourly availability, and bio electricity-only units are dispatchable and provide reserves.Source: 3-1. Generation Units · pp. 3–4
• Hourly solar and wind availability uses Renewables Ninja data for 2006–2016, scaled to projected 2050 averages, while technology costs use European Commission projections.A 10% discount rate is used for annuities.Source: 3-1. Generation Units · p. 4
• Projected 2050 generation requirements are 404 TWh, with 7% transmission and distribution losses and no electricity imports or exports assumed.Transmission network limits are omitted; plants are aggregated, and operating reserves are set as percentages of hourly load and VRE generation.Source: 3-3. Load and System Representation · p. 5

4.1. CO2 Emission Constraints and Battery Installation

• 80% of installed capacity and 73% of demand are renewable without batteries; with batteries, renewable capacity reaches 91% and annual generation share reaches 85%.Source: §4.1.1 · p. 5, Figure 1 · p. 6, Figure 2 · p. 6
• Battery cases show greater weather/load variation in generation shares, particularly for wind and bio, than cases without batteries.Without batteries, max–min variation is 13% for gas, 19% for wind, 9% for solar, and 20% for bio; with batteries, it is 20%, 35%, 11%, and 40%, respectively.Source: §4.1.1 · p. 5, Figure 1 · p. 6
• Tighter CO2 limits shift flexible generation from gas to CCS and then bio; batteries reduce gas and bio generation while favoring solar over wind.Under zero emissions, CCS is unsuitable because its capture efficiency is below 100%.Source: §4.1.1 · pp. 5–6, Figure 1 · p. 6, Figure 2 · p. 6
• Average curtailment is 4% for solar versus 23% for wind, and tighter CO2 limits increase renewable curtailment.Solar curtailment is higher with batteries and wind curtailment lower; 8-hour battery capacity is 2–5 times that of 3-hour batteries across CO2 constraints.Source: Figure 3 · p. 6, §4.1.1 · p. 6

• In the no-limit case, batteries reduce average emissions from 97 g to 54 g per kWh consumed, while emissions still vary 40% across weather and load data.The energy-cost figure compares cost per MWh against CO2 emissions with and without batteries across weather and load data.Source: §4.1.2 · pp. 6–7, Figure 4 · p. 7
• Almost 20% energy-cost reduction with batteries in the zero-emission system, versus 7% without a CO2 constraint.Without batteries, energy not served can reach 20 GWh and unmet reserves 40–60 GWh; batteries reduce both substantially.Source: Figure 5 · p. 7, §4.1.2 · p. 7
• At 100-hour duration, ESS breakeven cost reaches 85 USD/kWh; long-duration storage is not critical for this case study.The analysis holds energy capacity constant while changing power to represent duration, and excludes ESS investment costs from the optimized system costs.Source: §4.2 · pp. 7–8, Figure 6 · p. 8
• Breakeven cost is higher for the first, smaller ESS installations, while uncertainty affects short-duration systems more; at 100 hours, variation stays below 5%.For 3-hour ESS, breakeven-cost deviations are 30%, 20%, and 12% for 1, 10, and 100 GWh, respectively.Source: §4.2 · p. 8, Figure 6 · p. 8
• Lower-duration storage boosts solar generation, while longer durations shift generation shares back toward the no-ESS mix; at 100 GWh, bio remains lower because ESS supplies flexibility.One- and three-hour ESS particularly increase solar’s contribution by shifting midday generation to evening peak-load hours.Source: §4.2 · p. 8, Figure 7 · p. 8
• The 11-year optimization minimizes expected costs across all weather and load years, and its zero-emission portfolio is compared with single-year plans.The analysis uses historical hourly patterns from 2006–2016 while scaling annual VRE availability factors to the same 2050 value.Source: §4.3 · pp. 8–9
• Single-year solutions vary substantially by technology: offshore wind capacity varies up to ±100%, while the 11-year co-optimized solution differs from the single-year average.Annual VRE availability factors are scaled to the same 2050 value; historical years contribute their hourly variation patterns.Source: §4.3 · p. 9, Figure 8 · p. 9
• Planning from 2009 or 2010 performs worst in other years, whereas 2008, 2015, and the 11-year plan show narrower operating-cost variation.The authors attribute the poor outcomes to mismatch between VRE output and load, causing capacity inadequacy and costly load curtailment.Source: §4.3 · p. 9, Figure 9 · p. 9, Figure 10 · p. 10

5. Discussion

• Without ESS, the optimal plan tends toward costlier flexibility options such as bioenergy and natural gas with CCS; ESS enables more RES and lowers investment and operating costs.The study does not model social or political limits on resource expansion, or network, transmission, and distribution constraints.Source: §5 · p. 10
• Full decarbonization without energy storage increases energy generation cost by 40%, versus 20% with batteries, largely because bio-generation provides costly flexibility.The increase in system cost is greater when emissions are cut from 10% to zero than when moving from no limit to 10% emissions.Source: §5 · p. 10
• ESS benefits decline at higher installed capacity; durations under 24 hours have the greatest configuration and cost impacts, with shorter storage favoring solar and longer storage favoring wind.Durations beyond 100 hours, including seasonal storage, were not analyzed.Source: §5 · p. 10
• Across 11 years, hourly VRE and load variation changed optimal expansion despite constant average annual capacity factors, and single-year planning could raise costs in other years.The study recommends optimizing across multiple years or using a single-year worst-case scenario for a more robust plan.Source: §5 · p. 10
• The study’s assumptions and inputs are highly uncertain, particularly given the long horizon to 2050.Source: §5 · p. 10

6. Conclusion

• High RES penetration occurs even without a CO2 constraint or economic RES incentives, while gas power plants with CCS play a minor decarbonization role because capture efficiency is below 100%.Source: §6 · p. 10
• Under the given technology-cost assumptions, ESS lowers system costs and reduces ENS, while storage under 10 hours adds significant value and 10–100-hour storage has modest marginal value per kWh.ESS increases solar curtailment in the optimal mix; first installed capacities deliver greater system-cost savings and higher breakeven cost.Source: §6 · p. 11
• In the Italy case study, energy costs vary by up to 10% depending on the year optimized, and single-year plans can produce costly, high-curtailment operation in other years.The study concludes that capacity expansion should be optimized across multiple years of weather and load data.Source: §6 · p. 11
• Future work includes optimizing ESS duration within capacity expansion, modeling degradation, testing broader technology-cost and fuel-price assumptions, and including transmission networks and grid stability.Source: §6 · p. 11