The Optimal Timing of Storage Additions to Solar Power Plants

Abstract

The addition of battery storage to solar plants enhances the ability of those plants to deliver electricity during high-value periods. However, the value proposition of storage improves over time due to falling battery costs and increasing volatility in electricity prices, making it unclear when storage adoption should occur. In this work, we consider a 100 MW solar plant constructed in the year 2022 and build a techno-economic model to determine the optimal system design and timing of storage additions in four locations (CAISO, NYISO, ERCOT, and PJM). We find that the optimal time to add storage is 5–10 years after solar plant construction and that the optimal storage quantity is much higher than the amount selected if storage is included during the initial plant construction. Additionally, the model suggests significant upscaling in inverter capacity, allowing storage to deliver electricity during brief high-price periods. We also consider the effects of temporary and permanent subsidies for batteries, showing that a long-term subsidy encourages economically optimal delays in storage adoption.

1. Introduction

In 2021, the Environmental Protection Agency [1] found that the electric power sector accounted for roughly 25% of total US greenhouse gas (GHG) emissions. Worldwide, electricity use (residential, commercial, and industrial) accounts for almost 42% of total GHG emissions [2]. Despite this, by 2050, global energy demand is expected to triple [3]. Without the use of low- or zero-carbon production, this increase in electricity generation could significantly increase GHG emissions. Large-scale solar power plants currently represent only 3.9% of US generation but are rapidly being deployed, representing 53% of new generation in 2023 [4,5].

Solar power has a notable weakness due to intermittency in power generation due to weather, nighttime, and seasonality. This creates greater volatility for the electricity grid, which must be managed either at the solar facility or by other grid entities [6,7]. One solution to this volatility is the addition of battery energy storage that is co-located with the solar plant, which allows the solar + storage facility to become a more reliable and dispatchable source of renewable energy [8,9]. Historically, the use of battery storage solutions was limited due to high battery costs, but this has been changing as these costs decrease, making battery storage more attractive for solar developers. The National Renewable Energy Laboratory (NREL) predicts that battery costs will continue to decrease through to 2050 [10].

The importance of affordable renewable energy has led lawmakers to institute federal tax credits for solar power. The tax credit program allows for a plant to capitalize on either an investment tax credit (ITC) (reduces tax liability for a percentage of the solar system cost) or a production tax credit (PTC) (credit per kilowatt-hour of electricity generated by solar technologies for the first 10 years a system is in operation). These credits offer substantial assistance to solar owners and investors, having driven extensive investment in the last 10 years [11]. However, the decision of whether to add battery storage to existing or newly built solar plants is more complex. Uncertainty surrounding the operation of solar + storage plants (due to plant characteristics, grid characteristics, and geographic characteristics) and the predicted declining costs of solar and storage assets can make it difficult to evaluate and justify the upfront cost of adding storage to a solar facility.

Existing research on storage additions to solar plants has identified the potential of storage to mitigate intermittency in solar generation and considered the economic costs and benefits of such investment [12,13,14,15]. In 2010, Nair and Garimella [15] found that battery energy storage has high potential revenue but will not be extensively employed until initial capital costs are driven down by policy or demand. Eyer and Corey [13] concurred with this, but also mentioned that this is recognized by policymakers and regulators, who are creating current or emerging incentives that are strong opportunity drivers for storage adoption.

Other work has identified the optimal storage size under different plant characteristics, such as weather, electricity consumption, and electricity prices [16,17,18]. Arun et al. [16] provided illustrative examples of storage sizing and proposed that retaining a high system reliability requires attention to the ambient temperature effect on system sizing. Other work considers where future deployments should be located to ensure that they make sense economically, which tend to be those areas with the best weather patterns and supportive electricity prices, such as California and Texas [19].

Researchers have performed detailed studies of different solar + storage technologies and configurations, determining how to best exploit such technologies both behind and in front of the meter [20,21,22]. Englberger et al. [20] found that including both AC and DC storage can decrease the required amount of inverters and increase revenue, but also increases capital costs since batteries are more expensive than inverters. Montanes et al. [22] concluded that two- to four-hour duration batteries currently seem to be the most economically viable and locations like California and Texas should have a 1:1 ratio of battery to generation capacity to achieve the highest net value. These are just prominent examples from an extensive body of solar + storage research that considers optimal design from many different perspectives. However, one perspective seems to be missing: When should we add energy storage to a solar power plant? In an era when the costs of storage keep falling and the benefits of the better dispatchability of solar output are increasing, investigating the optimal timing of storage additions is highly relevant [10,23].

The net benefit of adding energy storage to a solar plant increases over time due to the following two factors: the falling cost of batteries and the increasing benefit of electricity dispatchability. The first factor is well-known, but a smaller and growing body of literature is showing how high-renewable systems will affect electricity system economics. Average electricity prices are likely to fall as the deployment of renewable energy increases, introducing a “revenue cannibalization” effect for investors in wind and solar [15,19,24]. There is a wide body of research that studies future pricing, and while this is important, it lacks substantive conclusions as to how to mitigate this phenomenon or how revenue cannibalization can increase the value of storage over time. The focus on decreasing energy prices, especially during periods when solar/wind energy is being produced, indicates the need for future research to understand how we can continue decarbonizing the grid while ensuring that it makes both social and business sense [24,25].

Prior work on solar + storage technology has intensively investigated many important questions but has not addressed the issue of the optimal timing for the purchase and construction of battery energy storage for new and existing solar plants. In this work, we directly address this question, using a techno-economic model that combines future energy prices, infrastructure costs, and solar plant design to identify the optimal timing and design of storage additions to solar plants. While investment decisions may currently consider the fact that storage will be lucrative in the future, our work additionally considers how initial solar plant design might prepare for later storage additions by scaling the inverters in an anticipatory way. It may be optimal to construct storage capacity in one or more phases several years after the initial construction of the solar plant. This model identifies the optimal inverter sizing, storage capacity, and construction timing for solar + storage plants. We run the model with data from four Independent System Operators (ISOs) to compare the economics of differing locations across the US. Furthermore, this model is used to assess policy options, such as the effect of different subsidy quantities and phase-out, and can aid in the design of policies that support the adoption of solar + storage facilities.

2. Materials and Methods

Figure 1 presents an overview of the modeling approach. The model begins by producing a forecast of future electricity prices, using historical prices from 2017 along with NREL forecasts of wind/solar generation through to the year 2050. This results in hourly prices for four ISOs every five years from 2017 through to 2047. These future electricity prices are used as inputs for a linear programming-based operational model of a solar + storage plant. This model produces the optimal hourly operation of a solar + storage generator, given the size of the solar, storage, and inverter components, as well as electricity prices and solar generation patterns, then calculates the annual revenue. A Net Present Value (NPV) model calculates the NPV of a given solar + storage buildout strategy, given the scale and timing of solar, inverter, and storage additions and the estimated annual revenue of operation in each 5-year period. A higher-level optimization module uses the solar + storage operational module and the NPV module to search for NPV-maximizing buildout strategies. For this work, we assume that a 100 MW solar plant will be built in the initial year and allow the model to optimize its storage quantity, the timing of storage additions (including in year 0), and the initial scale of inverters. Thus, the modeling perspective asks the following: if you are building a solar plant, how should inverters be initially scaled, when should storage be added, and how much storage would you want to add?

2.1. Data Inputs

Each Independent System Operator (ISO) in the US has its own electricity generation mix, energy price patterns, and solar resource. These factors can contribute to differing optimal solar + storage strategies between locations. For this reason, we perform the analysis for locations at the following four US ISOs: New York ISO (NYISO), PJM Interconnection (PJM), Electric Reliability Council of Texas (ERCOT), and California ISO (CAISO). For each ISO, we utilize the locational marginal price (LMP) in two different years, 2017 and 2022. The LMP represents the hourly electricity prices in different geographical zones as defined by each ISO. The zonal pricing system is employed by ISOs to accurately reflect the differences in electricity generation, transmission, and distribution costs in different physical areas. Additionally, we collect data on renewable generation capacity from the National Renewable Energy Laboratory (NREL) Electrification Futures Study for each ISO and historical capacity in both of the aforementioned years [26,27]. Finally, we use demand and seasonal regression coefficients from prior research by Das et al. [24] to aid in calculating future LMP values up to the year 2050.

The next level of analysis, the solar plus storage module, takes the hourly electricity pricing from the prior level, as well as hourly solar generation patterns from Das et al. [24] and the NREL renewable generation capacities. Additionally, this module receives the initial inverter size, solar capacity, and storage capacity from the higher-level optimization module.

The high-level optimization relies on cost data for solar (per MW) and inverters (per MW) and the time-varying cost of storage (per MWh). Utility-scale solar and inverter costs are taken from NREL’s 2022 annual technology baseline, while the current and future battery storage costs are derived from NREL’s storage futures study [27,28]. These inputs are given as parameters for the model and are not altered through the optimization.

2.2. Electricity Prices

The modeling approach requires estimates of future hourly electricity price patterns, which are challenging to locate or derive. For this work, we use a simple method to estimate future hourly electricity prices developed by Das et al. [24]. The electricity prices we use include both past and future electricity prices, reflecting model years of 2017 (historical), 2022 (historical), 2027, 2032, 2037, 2042, and 2047. Historical electricity prices are monitored and published by ISOs and other third parties, which are the sources for our collection of 2017 and 2022 data [29,30,31]. We decide to examine both 2017 and 2022 as reference historical years due to the concern that 2022 might include undesirable effects from COVID-19. We recognize that the pandemic may have affected electricity utilization and pricing, as well as renewable energy projects, which could skew the pricing data and expected patterns from normal after the year 2019.

After collection, the 2017 and 2022 electricity pricing data are organized into an hourly time series for each ISO. In the case of locational marginal prices (LMPs) that are reported at a higher resolution than hourly, the inter-hour pricing is averaged. Any missing data is interpolated from adjacent data points.

An important input to the model is estimates of future hourly electricity prices. We use the method presented in Das et al. [24] along with forecasts of future wind/solar deployment from the NREL Electrification Futures Study [27]. To predict future electricity prices, we model renewable generation as negative demand and assume that there is a linear relationship between price and renewable generation capacity, as determined by Das et al. [24]. To calculate future pricing, the following equation is utilized:

LMP_new = LMP_ref − (Change in USD per MW of Renewable Generation × (RenewableGeneration_current − RenewableGeneration_ref)).

(1)

The benefit of this method is that it predicts future prices while retaining the realistic hourly fluctuations of observed/historical data. However, this method does not consider large-scale shifts in demand patterns or traditional generation, meaning that it is more accurate for near-term predictions and is likely to be structurally incorrect as the forecast goes further into the future.

The reference year of 2022 is used for the LMP and renewable generation data in Equation (1). The change in USD per additional MWh of renewable generation represents the linear price decrease informed by the amount of installed renewable energy. These relationships are taken from Das et al. [24], which show the effects of changes in demand on price, calculated on a seasonal basis (seasons are defined as February–April, May–July, August–October, and November–January). Estimated hourly wind/solar generation is calculated by multiplying the future capacity in each region (using data from NREL’s Electrification Futures Study),

Renewable Generation per Hour = Solar Capacity × Capacity Factor,

(2)

and hourly capacity factors from Das et al. [24], originally derived from the Eastern Wind Integration and Transmission Study (EWITS), the Western Wind dataset, and the Typical Meteorological Year (TMY3) dataset.

It is important to note some benefits and limitations of this price forecasting method. As previously mentioned, by utilizing past pricing data, the realistic hourly pricing patterns are preserved into the future, which is not the case for price patterns derived entirely from dispatch models. Additionally, the hourly capacity factors allow our predictions to follow generation patterns observed in recent years. On the other hand, these factors become less pertinent over time, so the estimates will provide diminishing returns over the long term. Given the importance of near-term periods in our NPV calculations, due to discounting, the use of a method that is more relevant for near-term periods seems appropriate.

2.3. Solar Plus Storage Linear Programming Model

The linear programming model generates hourly energy dispatch in a solar + storage plant configuration. This provides us with an optimized dispatch model that captures the dynamic relationship between hourly electricity price fluctuations and plant storage characteristics. The linear programming model is built upon a previous model used to optimize a wind-powered storage-only system, with some key changes to adapt the model for our application [32]. The optimization is performed over a one-year period (8760 h).

• The model uses the following decision variables:
  • $E_t\in \mathbb{R}$: Net energy dispatched at time t (MW); positive for discharge, negative for charge. Charge is from solar generation only—no buyback from the grid.
  • $S_t\in \left[0,S_{max}\right]$: State of charge at time t (MWh)
• The model uses the following input parameters:
  • $P_t$: Electricity price at time t (USD/MWh)
  • $G_t:$ Hourly solar generation available at time t (MW)
  • $S_{max}:$ Storage capacity (MWh)
  • $I_{max}:$ Inverter capacity
  • $R_{max}:$ Maximum battery charge/discharge rate (MW)
  • ${\eta }_{rt}\in \left(0,1\right]:$ Round-trip efficiency of the battery system
• P_t and G_t are time-varying inputs and S_max, I_max, R_max, and η_rt are constants. The model assumes perfect foresight of electricity prices and solar generation values in the optimization year.

The model maximizes the total revenue produced by storage, as follows:

$$\mathrm{Max}\sum P_t{E}_t$$

(3)

This is modeled over a one-year (8760 h) period. The model is subject to the following constraint equations:

$${\forall }_t,0\le S_t\le S_{max}$$

(4)

$$S_{t=1}={\displaystyle \frac{S_{max}}{2}}$$

(5)

$$S_t=\left\{\begin{array}{cc}{S}_{t-1}-E_{t-1}·\sqrt{{\eta }_{rt}},& if{E}_{t-1}\ge 0\\ S_{t-1}-{\displaystyle \frac{E_{t-1}}{\sqrt{{\eta }_{rt}}}},& if{E}_{t-1}<0\end{array}\right.$$

(6)

$${\forall }_t,E_t\le min(R_{max},G_t)$$

(7)

$${\forall }_t,E_t\ge -min(R_{max},max(0,I_{max}-G_t\left)\right)$$

(8)

The state of charge for the system cannot be negative and cannot exceed S_max, as shown in Constraint (4). The initial state of charge for the storage system S_t is set to 50% of its maximum capacity at t = 1, as shown in Constraint (5). The state of charge in hour t is defined by the prior state of charge at hour t − 1 minus the change in energy dispatch during that hour, as shown in Constraint (6). The change in state of charge is subject to a round-trip efficiency loss. Constraint (7) ensures that discharging is limited by the lesser of the battery’s discharge capacity R_max and the available solar power output G_t, preventing exporting more power than the system can physically provide. To prevent charging beyond physical system limits, (8) imposes a lower bound on E_t, ensuring that the battery can only charge at a rate up to the lesser of the inverter’s remaining capacity (I_max − G_t) and the battery’s maximum charge rate R_max. The negative sign reflects that charging is represented by negative values of E_t. Equation (8) also prevents the battery from charging from the grid.

To manage computational demands and numerical stability in hourly optimization, the model divides the 8760 h of each simulated year into overlapping 500 h segments, with a 50 h (10%) overlap between successive chunks. This overlapping region ensures continuity in dispatch and state-of-charge trajectories. For each new segment, the initial battery state of charge is set as equal to the final state from the previous segment. Optimization is executed using deterministic prices and solar generation inputs specific to each year.

Hourly plant operation is optimized via linear programming only for the following set of representative “key” years spaced at five-year intervals: 2022, 2027, 2032, 2037, and 2042. For years between key year simulations, annual gross revenue is estimated using linear interpolation. Prior to the storage build year, revenue interpolation is performed using solar-only dispatch results. After storage is built, it is interpolated from the storage-inclusive result, which tends to increase revenue due to greater dispatch flexibility.

2.4. Net Present Value Model

The middle layer of the model functions to communicate between the user-facing optimization model and the annual solar + storage operation LP module to generate a lifetime NPV estimate for a given plant construction scenario. This layer is where the plant investment, battery prices, operations and maintenance, battery salvage, and discounting are handled. Final cash flows are combined by converting all values to real 2022 USD for NPV calculations.

2.4.1. Solar Plant

While the model is capable of varying solar scale and timing, all scenarios in this work assume that a 100 MW fixed-axis solar plant is built in year 0 (2022) of the project. Thus, all build scenarios have the same estimated total cost of USD 94 M in 2022 prices [28].

2.4.2. Inverter Pricing

Inverters are estimated to cost USD 0.045/Wdc [28]. Inverters are always built during plant construction and are not expanded in later periods. This allows us to see how the anticipation of later storage additions affects the sizing of initial inverter selection.

2.4.3. Battery Pricing

Battery pricing is calculated based on NREL’s 2021 Battery Energy Storage System (BESS) table [27]. Battery types include 2 h, 4 h, 6 h, 8 h, and 10 h. Future pricing conditions include low, mid, and high scenarios. Our base case assumes the mid scenario and 2 h batteries, though the power to energy ratio is investigated in sensitivity analysis. Battery additions to the solar plant are limited to a single year, including allowing for year 0 additions, where batteries are installed in the initial project with the solar technology.

2.4.4. Operations and Maintenance

Yearly solar plant O&M is estimated to be 1.1% of the capital investment costs. Yearly battery O&M is estimated to be 2.5% of the purchase price of the current active batteries [28].

2.4.5. Capital Payments and Battery Storage

All capital costs are handled as annualized payments over the expected lifetime of the asset, equivalent to an amortized loan with an interest rate of 5.5% for the solar plant and 4.5% for battery storage. This approach is used to address “salvage value” issues relating to the end of the modeling period. The operational lifetime for the solar plant is 25 years, resulting in an annual Capital Recovery Factor (CRF) of 8.55% (7.45% Loan CRF + 1.1% Operations and Maintenance (O&M)). Battery payments are made over the expected lifetime of 15 years, which results in a higher CRF of 11.81% (9.31% Loan CRF + 2.5% OM). Because the battery lifetime is shorter than the study period, storage is replaced at end of life with new storage at the prevailing current year’s (2037 or 2042) market price. Because all asset costs are calculated as a fixed annualized rate over their lifetime, a salvage value does not need to be estimated.

2.4.6. Discounting

A discount rate of 5% and an inflation rate of 3% are used to calculate NPV.

2.5. Optimization Model

The high-level optimization model searches for NPV-maximizing choices for inverter capacity (built in year 0), storage capacity, and the storage addition year. This optimization is performed in MATLAB version R2023b, using the global optimization and parallel computing toolboxes. Storage capacity and inverter capacity are constrained to specific ranges, and the possible storage addition years are constrained to 5-year increments between 2022 and 2047, as described in

Table 1. Optimization parameters used in the high-level optimization.

Optimization Parameter	Parameter Definition
Utility-Scale Solar	100 MW Fixed Axis Solar
Storage Capacity	[0–600] MWh (NY, CA, PJM) [0–6000] MWh (TX)
Inverter capacity	[0–400] MWh (NY, CA, PJM) [0–4000] MWh (TX)
Storage Addition Years	2022, 2027, 2032, 2037, 2042
Solar + Inverter Capital Costs, Operation, and Maintenance	Paid as a fixed rate loan over the plant operation period.
Battery Costs	Paid over a 15-year battery lifespan. 1

¹ New loan rate is calculated when replacement batteries are purchased after 15 years.

. A “build” is defined as a single combination of storage capacity, inverter capacity, and storage addition year for a solar plant at a given ISO location. To determine the appropriate range of searchable values for storage capacity and inverter capacity, preliminary testing is performed at a low resolution (~50 builds per storage addition year, 250 total) until a 25% increase in maximum potential storage capacity does not influence the optimal build conditions. For NY, CA, and PJM, 600 MWh is determined to be appropriate for base case analysis. ERCOT requires a significantly higher threshold at 6000 MWh in order to not impede storage optimization. The maximum search range for inverter capacity is set at two-thirds of the maximum potential storage capacity, as this provides more than adequate coverage while improving the optimization resolution.

Optimization within the specified solution space is performed with a genetic algorithm search, available within MATLAB’s global optimization toolbox. Each condition (e.g., base case and subsidy) begins with a 25-generation optimization over the total range of storage, inverters, and storage addition years. Since the genetic algorithm favors higher-performing regions of the solution space over lower-performing areas, storage addition years with a relatively lower NPV are underrepresented by the search (~10:1 in number of builds). To improve the resolution within these suboptimal storage addition years, an additional 5–10 generations of optimization are run in each underrepresented storage addition year. Final NPV estimates for a given set of conditions are not influenced by the initial conditions, suggesting good convergence (see Supplementary Data). Over 1.2 million potential build combinations are possible for each set of conditions, making a brute-force search computationally infeasible. However, validation testing finds that each dimension is smooth and contains only one maximum. Given the smooth nature of the optimization landscape, 0.5% of potential combinations (~6000) is sufficient for the genetic algorithm to obtain precise results that are reliably chosen as optimal during repeat runs with the same conditions.

3. Results

3.1. Base Case by Region

The base case analysis assumes certain fixed parameters across the four studied regions, as follows: 100 MW of solar built in year 0 (2022), 2 h power-to-energy ratio for batteries, and the NREL mid-price scenario for battery cost projections. The decision variables that are being optimized are the initial inverter scaling in year 0, the quantity of energy storage to add, and the year in which to add it. The base case results are shown in

Table 2. Optimal build results—base case. The last column shows the number of systems tested by the optimization to identify the optimal solar + storage design.

Region	Optimal Conditions
	Plant NPV	Inverter Capacity (MW)	Storage Capacity (MWh)	Storage Build Year	Number of Cond. Simulated
CAISO	USD 118 M	237	445	2032	1271
ERCOT	USD 350 M	1094	2100	2027	970
NYISO	USD 53 M	143	207	2032	1095
PJM	USD 82 M	180	298	2032	1296

.

In the base case scenario, all four regions show a positive NPV over the 25-year operational period of the plant, meaning that a solar + storage project is expected to be profitable. Additionally, all four regions choose to add storage at some point, showing that storage is an appropriate investment that improves upon the solar-only NPV. In anticipation of later storage, all four locations install excess inverter capacity. In practice, the economically optimal design for most solar facilities uses an undersized inverter system (i.e., 100 MW of solar panels would be linked to less than 100 MW of inverters) [33]. However, our model chooses to significantly oversize the inverters for the 100 MW solar facility, but this only lasts from 5 to 10 years until storage is added and a higher inverter capacity is needed. This choice makes financial sense because of the low cost of inverters relative to both solar and storage, meaning that the preparation cost is small. The storage addition year in the base case is in the 2027–2032 timeframe in all four locations, suggesting that storage additions are more preferable in future years than upon solar plant construction, at least with the base case assumptions. The ERCOT results suggest adding storage in the year 2027, the earliest timing in our analysis, which may be because of the higher value of storage under ERCOT price patterns. The optimal storage and inverter capacities for ERCOT are significantly larger than those of the other regions, showing that the model sees this as an opportunity to build a profitable and large storage facility co-located with the solar plant. Finally, regions with decreased storage utilization tend to have a lower plant NPV and quantity of inverters, presumably because the limited market benefits of storage limit both factors.

3.2. Forced Storage Build Year Analysis

The model can be used to consider the possibility of adding storage in any year in the studied timeframe. By comparing and understanding how the optimal storage and inverter capacities change with build year and how these changes can affect the NPV of the plant, we can better understand why the model identifies 2027 to 2032 as the optimal timeframe choice for storage additions. For this set of results, we force particular build years for storage and let the model choose the optimal quantity of storage and the initial inverter scale. For comparison, we also include a solar-only scenario where no storage is permitted in any year, though the inverter scale is optimized. This “forced storage build year” analysis is performed for all four regions under otherwise base case conditions. Figure 2 depicts the results for CAISO, with Figure 3 showing the cumulative cash flow for different build years in CAISO. Figure 4 depicts the results for ERCOT. Figure 5 depicts the results for NYISO and Figure 6 shows the results for PJM.

For CAISO, all storage scenarios have a higher NPV than the solar-only comparison. The optimal amount of storage is approximately four times higher if storage is added 5 or 10 years after solar installation. Even with cheaper storage, this still represents an increased total investment in batteries, but represents a kind of tipping point where the marginal benefit of more storage becomes slightly higher than the marginal cost until much higher levels of storage are reached. In this way, moderate changes in the cost of storage can result in a significant shift in the optimal system design, as observed previously for other electricity systems [34]. The optimal inverter capacity has a slight upwards trend, following the optimal quantity of storage. Inverters are inexpensive compared with solar and storage, so this increased inverter capacity is built in anticipation of later storage and allows storage to better time-shift electricity production to specific valuable hours. Figure 3 helps us to understand these trends, as it demonstrates that battery additions tend to increase annual revenue to the plant but tend to have lower total costs (despite higher amounts of storage) in later years.

ERCOT also shows an increased NPV for storage added in any operational year, with an optimal build year of 2027. ERCOT electricity prices are the most volatile among the four studied locations, presumably due to the energy-only structure of the electricity market, and storage is naturally profitable under these conditions. Thus, the model chooses much larger amounts of storage than in other regions, with storage able to hold several days of solar production. This is true even though the storage can only be charged from the co-located solar panels, and it uses storage to save up solar energy and deliver it in bulk during specific high-price hours. These observations are consistent with the real-world deployment of a large number of battery facilities in ERCOT, both co-located with solar panels and as stand-alone sites [35].

NYISO has the overall lowest NPV results across the board. The addition of storage only improves NPV starting in 2027, with an optimal build year of 2032. Like other locations, the optimal amount of inverters increases in the first few periods and then falls towards the end of the study period. This is because of the end-of-study cutoff for NPV: across all scenarios, a greater quantity of inverters is associated with a greater quantity of storage, as it enables the faster discharge of batteries in high-value hours. But if storage is added in the last 5–10 years of the solar plant’s lifetime, the overall benefit of that synergy with storage is limited to a short period and overbuilding inverters in the initial plant design is less desirable.

PJM performs similarly to NYISO but has a higher overall NPV. PJM offers a greater NPV than NYISO without storage, and the model recommends building more inverters and storage than in NYISO and generates a greater NPV.

3.3. Simple Fixed Subsidy

We also analyze the impact of a scenario with a fixed subsidy rate on the plant NPV to determine its effectiveness. Currently, the US investment subsidy provides a 30% discount on equipment, as described in Section 1. Figure 7 shows the simulated effect of a 30% fixed subsidy on storage investment costs.

The following three scenarios are tested in CAISO over the standard operational period: no subsidy (control), an expiring 30% battery subsidy that only provides benefits in 2022, and a fixed 30% battery subsidy that lasts the entire operational period. As expected, the presence of the subsidy increases the projected NPV and optimal storage capacity of the solar plus storage system. The increase in both the projected NPV and storage capacity is larger in the persistent subsidy case than the expiring subsidy case, despite the subsidies having an equivalent magnitude during the initial build in 2022. This is due to the replacement costs of the batteries also being subsidized over the operational period in the fixed subsidy case, whereas the batteries are later replaced at full price in the expiring case. Since the plant cannot downsize its storage capacity in future years, a more modest quantity of batteries is built in 2022 because the subsidy is expected to expire soon afterwards.

3.4. Battery Cost Projection Sensitivity Analysis

We also assess the effect of future battery prices on the optimal scale of deployment, as shown in Figure 8. Future battery pricing is separated into the following three levels offered in the NREL projections used: low, mid, and high. Mid corresponds to the middle future pricing prediction by NREL (our base case), low represents a low-cost scenario, and high represents their high-cost scenario.

As expected, given the results from the subsidy analysis, the low-pricing condition (which corresponds to lower future battery prices) significantly increases the optimal quantity of storage in future build years compared to the mid and high conditions. The mid- and high-price scenarios have a similar amount of storage, demonstrating the general value of storage for future solar plants even under higher prices. However, the high-price scenarios have a notably lower NPV than the mid-price results because of the cost of batteries.

3.5. Battery Charge/Discharge Rate Sensitivity Analysis

A sensitivity analysis was performed for the CAISO location to assess the effect of battery charge/discharge rate on optimal build parameters, as shown in Figure 9. The 4 h discharge rate performed the best in terms of maximized NPV, closely followed by the 6 h condition. The 2 h batteries performed better than the 8 h batteries in the near years (2022–2027), but the 8 h batteries overtook the 2 h batteries from 2028 onwards. Predictably, with faster battery charge/discharge rates, more inverters and less total storage capacity were required to capitalize on hourly electricity price fluctuations.

4. Discussion and Conclusions

4.1. Key Lessons

In this work, we used a combined linear programming and Net Present Value approach to determine optimal solar plant design choices as a function of battery properties and the timing of battery additions. While each scenario had its own results, there were several consistent themes that we observed. First, the addition of batteries at some point in time improved the NPV of the solar plant across scenarios and assumptions. If storage was added during the construction of the plant (in 2022), only two of the four locations—CAISO and ERCOT—showed an NPV benefit. This is consistent with observed market behavior, as recent electricity price patterns and anticipated growth in renewables made California and Texas the top two states for storage adoption in 2023 [34]. While the PJM and NYISO solar plants did not show a net benefit for storage in 2022, all four ISOs enjoyed net benefits from storage starting in the year 2027.

The second general result was that the optimal time to add storage to a solar plant built in 2022 was consistently 5 to 10 years in the future. These results suggest that there is value, sometimes significant, for solar developers and plant owners in planning for future storage additions and integration. This may imply physical preparation, such as leaving space for battery enclosures within the facility or sizing inverters based on future needs rather than for optimal solar-only design. Preparation could also be financial or contractual, such as ensuring funding during plant construction for future battery additions.

A third general result was that the addition of DC-connected batteries prompted the addition of more inverters, which we were not expecting. Because batteries are able to absorb excess solar energy, they reduce or eliminate “inverter clipping”, the curtailed energy lost during periods when solar DC output exceeds the available inverters. Thus, our initial hypothesis was that batteries would allow for lower investment in inverter capacity and decrease inverters relative to a solar-only facility. However, due to the low cost of inverters and the high value of electricity in certain hours, the opposite strategy has greater net benefits: batteries allow you to deliver larger quantities of energy concentrated in periods of high prices, which requires the scaling up of inverter capacity.

4.2. Caveats

The base case results suggest that waiting to install batteries is consistently preferable to adding them when a plant is built, but this is somewhat contradicted by the fact that current solar plants sometimes include storage. There are practical reasons for this choice that we do not capture in our modeling. In standard solar development practice, a solar developer specifies a project design, contracts with an off-taker for the energy, and secures project finance before construction. This process necessitates a specific “bankable” project, and any uncertainty about future project modifications would complicate or threaten the ability to close on such a project.

A second practical issue is that our model does not apply additional costs to having multiple project phases. In reality, there would likely be additional soft costs to delaying battery construction by 5 or 10 years, such as identifying new financing or engaging a second engineering firm. These practical considerations push developers to design and build the entire project at one time and can explain why our model outputs might differ from current practice. However, there are also additional benefits in terms of real options. By deferring the decision to add batteries, the plant owner will be able to choose the preferred scale and timing of battery addition based on improved future information about electricity trends and battery prices. For example, waiting 5 years to add batteries could allow the owner to either add a larger-than-planned battery if electricity prices are more volatile than expected or to defer battery addition by another 2 years if battery prices do not fall as quickly as expected.

Another relevant caveat is that battery capacity is modeled as constant over time, while actual lithium-ion batteries experience capacity fade over years of use. For our results, this implies either an underestimation of the quantity of batteries that should be purchased or an overestimation of later-year revenue from battery services. We also do not consider the transmission line connection to the solar + storage facility, which may act to enhance the value of storage. In this work, it is assumed that transmission capacity is sufficient to accept any electricity from any solar + storage plant (though we note that, due to our electricity price modeling, the market value of that energy can sometimes be zero and, thus, unsold for economic reasons). If we expanded the analysis to include the design of the transmission equipment, storage value could increase because storage is able to smooth out the delivery of solar energy and, thus, reduce the required maximum power transfer. However, even this is uncertain: recall that all four locations tend to increase their inverter capacity when storage is permitted in order to rapidly deliver large amounts of energy during peak electricity price periods. This behavior actually increases the maximum power flow to the grid. Hence, the economically optimal solution to solar + storage + transmission optimization would be sensitive to both the volatility of electricity prices and the distance/cost of transmission itself.

4.3. Practical Implications for Developers and Policymakers

The results from this work can be useful for both solar + storage developers and policymakers. For developers or investors, this work suggests that many solar plants would benefit from (or benefit more from) added energy storage in 5–10 years. Explicitly including later additions of storage in the analytical and design process would allow developers to determine whether or how this affects their project’s design and increases its overall profitability. For policymakers, this work suggests that energy storage may be most beneficial if installed in the future rather than today, suggesting the benefits of incentivizing “storage-ready” solar projects and offering assurance that energy storage subsidies will still be available in the future.

In the CAISO-focused analysis of subsidies, we observe that both temporary and permanent subsidies result in increased quantities of storage adoption. However, the optimal quantity of storage is higher (550 MWh vs. 440 MWh) and occurs sooner (2027 vs. 2032) when storage subsidies are expected to persist in the future. This demonstrates the value of long-term policy certainty in supporting the storage and solar industries. In the case where a solar developer expects that policy support will end soon, they are more likely to add storage immediately, which is suboptimal and results in significantly less storage in total (300 MWh vs. 550 MWh). Long-term policy certainty would enable developers to plan for later additions of storage, which our analysis finds to be economically optimal.