Establishing best practices for modeling long duration energy storage in deeply decarbonized energy systems

TL;DR

This paper tackles the challenge of accurately modeling long duration energy storage (LDES) in macro-energy system planning by applying the Kotzur period-linking method within a continent-scale capacity expansion model. It systematically analyzes how spatial and temporal abstractions affect LDES value, using a shadow-price approach on a forced LDES case to quantify marginal benefits under various uncertainties. The findings show that temporal resolution and period-linking are the dominant factors in capturing LDES value, while spatial resolution has a comparatively modest effect; incorrect modeling can overestimate decarbonization costs by up to ~20% for cheaper LDES. The study delivers practical best-practice guidance for modelers, including linking representative periods and using a virtual-discharge CRM mechanism to fairly credit storage, thereby improving reliability and reducing biases in planning for deep decarbonization.

Abstract

Long duration energy storage (LDES) may become a critical technology for the decarbonization of the power sector, as current commercially available Li-ion battery storage technologies cannot cost-effectively shift energy to address multi-day or seasonal variability in demand and renewable energy availability. LDES is difficult to model in existing energy system planning models (such as electricity system capacity expansion models), as it is much more dependent on an accurate representation of chronology than other resources. Techniques exist for modeling LDES in these planning models; however, it is not known how spatial and temporal resolution affect the performance of these techniques, creating a research gap. In this study we examine what spatial and temporal resolution is necessarily to accurately capture the full value of LDES, in the context of a continent-scale capacity expansion model. We use the results to draw conclusions and present best practices for modelers seeking to accurately model LDES in a macro-energy systems planning context. Our key findings are: 1) modeling LDES with linked representative periods is crucial to capturing its full value, 2) LDES value is highly sensitive to the cost and availability of other resources, and 3) temporal resolution is more important than spatial resolution for capturing the full value of LDES, although how much temporal resolution is needed will depend on the specific model context.

Figures (18)

• Figure 1: Illustration of LDES period-linking methodology developed by Kotzur et al. kotzur_time_2018 for four representative days. The top row shows LDES state of charge for each representative period. Operations of the full system are modeled for these representative periods at hourly resolution. The middle row shows the sequence that the representative periods are arranged in to develop an approximated version of the full year (in reality this would have 365 periods, but only a few are shown for simplicity). The bottom row shows LDES state of charge across this approximated representation of the full year, which is calculated using the change in state of charge for each period in the sequence in the middle row. Note that this formulation allows LDES to both shift energy between seasons, and to discharge for longer than the length of a representative period, neither of which would be possible if the periods were unlinked.
• Figure 2: LDES value as a function of number of operational hours modeled, with and without the Zero Carbon CT resource, and with a breakout of the total value into capacity and energy value. Each dot represents one run of GenX with a particular number and length of representative periods.
• Figure 3: Left: price duration curves for the ERCOT zone with and without the Zero Carbon CT resource. Right: change in other resources when LDES is forced in (summed across all zones), with and without the Zero Carbon CT resource.
• Figure 4: LDES value as a function of number of operational hours modeled, broken out by representative period length.
• Figure 5: LDES value as a function of number of operational hours included, for multiple zonal aggregations. The two orange dots present for 8760 hours are a result of running both a full year without time domain reduction, and 52 representative weeks, which are not identical because in the latter case only LDES is able to shift energy between weeks. 8760 results are not included for all zones because the 8760 model was not computationally feasible for more granular zonal aggregations.