ARTICLES PUBLISHED: XX MONTH XXXX | VOLUME: 2 | ARTICLE NUMBER: 17001

The impacts of storing solar energy in the home to reduce reliance on the utility

Robert L. Fares* and Michael E. Webber

There has been growing interest in using to capture solar energy for later use in the home to reduce reliance on the traditional utility. However, few studies have critically assessed the trade-os associated with storing solar energy rather than sending it to the utility grid, as is typically done today. Here we show that a typical battery system could reduce peak power demand by 8–32% and reduce peak power injections by 5–42%, depending on how it operates. However, storage ineciencies increase annual energy consumption by 324–591 kWh per household on average. Furthermore, storage operation indirectly increases emissions by 153–303 kg CO2, 0.03–0.20 kg SO2 and 0.04–0.26 kg NOx per Texas household annually. Thus, home energy storage would not automatically reduce emissions or energy consumption unless it directly enables .

1 n recent years, there has been growing interest in storing energy lead-acid batteries used with solar panels in the UK would increase 41 25 2 produced from rooftop photovoltaic panels in a home battery both primary energy consumption and carbon dioxide emissions . 42 1 3 Isystem to minimize reliance on the electric utility . A number In this paper we critically assess the trade-offs of using lithium- 43 4 of vendors have sought to capture this emerging market, including ion battery storage to capture solar energy and minimize reliance on 44 5 market leader Tesla and German home energy the utility. We build on previous work by using measured 45 2,3 6 storage provider Sonnenbatterie . Notably, Tesla has partnered use and production data from 99 Texas households to understand 46 7 with Green Mountain Power, one of the largest electric providers how adding energy storage would impact power demand, energy 47 8 in the state of Vermont, to offer home storage to its customers; consumption, electricity service costs, and emissions of CO2, SO2 48 9 and Sonnenbatterie has partnered with Sungevity, the largest private and NOx from the electricity system. We consider two different stor- 49 4,5 10 solar company in the United States . age operation models and compare their impacts. We also perform a 50 11 While there is a growing market for home energy storage for sensitivity analysis considering various storage efficiencies, storage 51 12 rooftop solar panels, storage is not strictly required to integrate energy capacities, and storage power capacities to understand the 52 13 rooftop photovoltaic systems with the grid. A study on the impacts impact of energy storage under different scenarios. 53 14 of rooftop photovoltaic panels in California found that even at 100% 15 penetration (measured as the ratio between and Energy storage system model 54 16 peak system demand), the utility Pacific Gas and Electric (PG&E) The energy storage application considered in this paper is minimiz- 55 17 could maintain adequate voltage levels in its system by increasing ing the interaction between a household and the utility by minimiz- 56 18 the number of tap changing operations at a cost of ing power draws from and injections to the utility grid for the benefit 57 19 US$442,000 annually—or 0.007% of its US$6 billion annual opera- of the electricity customer in terms of increased solar energy self- 58 6,7 20 tion and maintenance budget . These findings align with previous consumption, independence from the utility, and reduced sensitivity 59 21 findings on the impact of high photovoltaic penetration in distribu- to grid outages. This application has been studied extensively in the 60 8 22 tion circuits in California . Furthermore, a number of studies have literature, and is the primary value proposition offered to residential 61 2,3,15–20,22 23 shown that upgrading conductors, upgrading the transformer, or customers by home energy storage vendors . 62 24 incorporating ‘smart’ photovoltaic inverter control could be used We utilize electricity data directly measured from 99 Texas 63 9–14 25 in lieu of storage to maintain adequate system voltage . Even if households over calendar year 2014 to reveal how home storage 64 26 energy storage were needed to integrate rooftop solar panels, it is would operate with solar panels to minimize reliance on the 65 27 not clear that it would have to be installed at the household level. utility. These data track electricity use and solar production with 66 28 Despite the fact that energy storage is rarely required to integrate a one-minute time resolution, allowing us to reveal how storage 67 29 rooftop solar panels, there is significant interest in capturing on- could respond to short-duration power fluctuations. The data were 68 30 site solar generation to minimize reliance on the electricity utility collected on a voluntary basis by the non-profit entity Pecan Street 69 31 and injections of solar energy to the grid. This application has and are freely available to university researchers through an online 70 15–21 26 32 been studied extensively in the literature , and it is the primary portal . Summary statistics for the 99 households in the sample are 71 33 value proposition offered to residential customers by home energy provided in Supplementary Table 1. 72 2,3,22 34 storage vendors . We model home energy storage operation using two different 73 35 While a number of studies have assessed the benefits of energy methods: a ‘target zero’ approach where the battery does not have 74 36 storage that captures rooftop solar energy to mitigate overvoltage information about the future level of solar generation or electricity 75 20,23,24 37 in the distribution grid and hedge utility tariffs , the amount demand and seeks to reduce injections to and demand from the 76 38 of energy consumed by the battery during operation and the grid to zero at all times; and a ‘minimize power’ approach where 77 39 corresponding emissions footprint is typically neglected. One the battery system has perfect information about the future level of 78 Q.1 40 notable exception is McKenna et al.’s 2013 study, which found electricity demand and solar generation over the day, and plans its 79

Department of Mechanical Engineering, The University of Texas, Austin, Texas 78712, USA. *e-mail: [email protected]

NATURE ENERGY | www.nature.com/natureenergy 1 ARTICLES NATURE ENERGY

ab'Target zero' operation 'Minimize power' operation

Calculate target battery power

Pbatt(t) = Puse(t) − Pgen(t) Perfect foresight of

to zero out net demand Puse(t), Pgen(t) over 24 h

Battery energy Yesconstraints No Optimization violated Minimize sum of squares of

Puse(t) − Pgen(t) − Pbatt(t) over 24 h Pbatt(t) = 0 Pbatt(t) = Puse(t) − Pgen(t)

Subject to battery energy and power Battery power constraints constraints violated Yes No

Pbatt(t) = Prated Pbatt(t) = Puse(t) − Pgen(t) Pbatt(t) over 24 h

6 6 Use Battery Use Battery Solar Net Solar Net 4 4

2 2

0 0 Power (kW) Power (kW)

−2 −2

−4 −4

27 July 27 July 27 July 27 July 27 July 27 July 27 July 27 July 27 July 27 July 0:00 5:00 10:00 15:00 20:00 0:00 5:00 10:00 15:00 20:00

Figure 1 | Storage operation model control logic and sample outputs. a, The ‘target zero’ method steps from one minute to the next with no foresight of future electricity demand or solar generation, and seeks to set net grid demand to zero whenever possible without violating the battery’s energy and power constraints. b, The ‘minimize power’ method has perfect day-ahead foresight of electricity demand and solar generation, and uses an optimization program to minimize the sum of squares of net grid demand over the entire day. Sample outputs for each method are shown below the flowcharts, with storage discharging given a negative sign and storage charging given a positive sign.

1 operation to minimize the sum of the squares of net power demand utility as much as possible. Neither operational method explicitly 26 2 from the utility over the entire day. Besides the level of foresight, the considers other grid-level services that could be offered by 27 3 primary distinction between these two methods is that ‘target zero’ distributed energy storage or the potential economic benefits of 28 4 seeks to maximize the number of hours during which the household those services. While we are aware of the fact that the methods of 29 5 is completely independent from the grid, while ‘minimize power’ storage operation selected are not optimal from a purely economic 30 6 seeks to minimize the magnitude of grid power demand over every or system perspective, our objective is to assess the specific impacts 31 7 minute of the day with equal weights placed on each minute, so that of storing solar energy in the home to minimize reliance on the 32 8 the household is resilient to a grid outage regardless of when it oc- utility, because this application is the primary value proposition 33 2,3,22 9 curs. Furthermore, the ‘target zero’ mode restricts the battery system offered to residential customers by storage vendors . The details 34 10 to charge only with , while the ‘minimize power’ mode of each of these operational models are provided in the Methods. 35 11 allows the battery to charge with grid or solar electricity to minimize For both operational models, three parameters define the home 36 12 demand over the day. Figure1 illustrates the control logic of each energy storage system: its power capacity (P rated) in kilowatts, its 37 13 operational method and shows sample outputs for one household. energy capacity (Erated) in kilowatt hours, and its roundtrip (a.c. to 38 14 These two operational methods were selected because they a.c.) energy efficiency (ηrt). For the base case battery system consid- 39 15 represent plausible yet distinct methods for storing solar energy in ered, we set these parameters equal to P rated = 3.3 kW, Erated = 7 kWh 40 16 the home to reduce reliance on the utility. ‘Target zero’ prioritizes and ηrt = 85%, corresponding to the parameters announced for 41 3 17 being as independent as possible during the current minute, while a common home battery system for daily cycle applications . We 42 18 ‘minimize power’ prioritizes being as independent as possible over also consider a range of power capacities P rated = 1–7 kW, energy 43 19 the entire day. By considering these two plausible yet distinct capacities Erated = 1–7 kWh, and roundtrip efficiencies ηrt = 70–100% 44 20 operational strategies, we can show the range of impacts that would in our sensitivity analysis, as discussed in Supplementary Note 1. 45 21 be expected for storage systems that operate somewhere between the 22 no foresight and perfect foresight extremes represented by ‘target Power demand and energy consumption impacts 46 23 zero’ and ‘minimize power,’ respectively. Figure2 illustrates the effect that home energy storage has on the ag- 47 Q.2 24 Note that both of these operational modes take a customer- gregate net power demand (electric load minus solar generation) for 48 25 centric perspective that seeks to minimize interaction with the the 99 households considered. When home energy storage operates 49

2 NATURE ENERGY | www.nature.com/natureenergy NATURE ENERGY ARTICLES

No storage 50 'Target zero' operation Storage: 'Target zero' operation 400 'Minimize power' operation Storage: 'Minimize power' operation 40 Sample means Population means 95% CI 200 30 20 0 10

Number of households 0

Net power demand, −200 99 households (kW) 0 200 400 600 800 1,000 −400 Additional annual energy consumption (kWh per household) Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Figure 3 | Energy consumption impact of adding energy storage. The Figure 2 | Aggregate power demand impact of adding energy storage. increase in energy consumption observed across the 99-household sample Energy storage reduces the magnitude of power flows in the local utility is shown using a histogram. The dashed lines indicate the mean increase grid by storing produced solar energy for later use in the home. When across the sample, and the shaded areas indicate 95% confidence intervals storage operates under the ‘target zero’ mode (shown in blue) with no of the expected mean across all households computed using a Student’s foresight of future electricity demand or solar generation, it reduces the t-test. The average additional energy consumption caused by home energy maximum aggregate power demand for the 99 households by 8% and the storage is 338 ± 14 kWh under the ‘target zero’ operating scenario and maximum magnitude of reverse power flows by 5%. When storage 572 ± 19 kWh under the ‘minimize power’ operating scenario. operates under the ‘minimize power’ mode (shown in red) with perfect foresight, it reduces peak power demand by 32% and reduces the Economic impacts 45 maximum magnitude of reverse power flows by 42%. In addition to the impact that home energy storage would have on 46 electricity demand and consumption, we also consider the impact 47 1 according to the ‘target zero’ mode, the aggregate is that storage used to isolate customers from the utility and increase 48 2 reduced by 29 kW or 8% from a value of 378 kW without storage to a solar self-consumption would have on the cost of electricity service 49 3 value of 349 kW with storage. Under the ‘minimize power’ operating to the customer. Note that we did not explicitly consider a particular 50 4 mode, energy storage reduces the level of peak demand by 121 kW utility tariff structure when selecting the objective function for 51 5 or 32%. Likewise, the maximum magnitude of reverse power flows storage operational management, because our goal is to show the 52 6 is reduced by 17 kW or 5% when storage operates in the ‘target zero’ economic impact of storage used to isolate a customer with solar 53 7 mode versus 158 kW or 42% when storage operates in the ‘minimize panels from the utility without binding storage operation to a 54 8 power’ mode. The results shown in Fig.2 are for the base case battery particular utility’stariff. Weconsider the economic impact of storage 55 9 system considered (P rated = 3.3 kW, Erated = 7 kWh and ηrt = 85%). used in Austin, Texas (where the home electricity data used in 56 10 Supplementary Figs 6–8 show how changing the power capacity, this paper were collected), other areas of Texas, and the states of 57 11 energy capacity, and efficiency of the storage systems affects their Hawaii and California, which have seen a significant penetration of 58 12 ability to reduce aggregate peak demand and injections. rooftop solar panels and are strong candidates for early home energy 59 13 The change in aggregate power demand is an important metric storage deployments. 60 14 for the utility, which must size distribution equipment to meet Table1 shows the average annual customer benefit and present 61 15 the expected maximum magnitude of net electricity demand. value calculated across our 99-household sample for the base case 62 16 However, residential electricity customers are not typically billed 3.3 kW, 7 kWh, 85% efficient energy storage system considered. 63 27–37 17 for their demand in kilowatts, but rather for their cumulative Results are shown for seven different Texas utility tariffs , as 64 38–42 18 energy consumption in kilowatt hours. Thus, we consider the energy well as four Hawaiian Electric Company (HECO) tariffs and 65 43–48 19 consumption impact of home storage on a customer-by-customer three California tariffs . Note that Hawaii and California are 66 20 basis. Figure3 illustrates the change in annual energy consumption shown only for illustration, as storage that operates in these regions 67 21 from the addition of storage for each of the 99 households when would see different patterns of electricity use and generation than 68 22 it is operated under the two operating modes considered. Because those observed from our sample of Texas households. For each 69 23 home energy storage consumes some energy every time it charges tariff, the approximate consumption tariff and feed-in tariff are 70 24 and discharges, annual energy consumption increases for every shown in US cents per kilowatt hour. These approximate values 71 25 household. The mean increase in annual energy use across the 99 are based on the average monthly consumption and production 72 26 households is 338 kWh when storage operates in the ‘target zero’ measured across our sample. The values shown for the average 73 27 mode and 572 kWh when storage operates in the ‘minimize power’ annual customer benefit and present value are calculated on the 74 28 mode, illustrated by the dashed vertical lines in Fig.3. This increase basis of each household’s precise monthly electricity use and each 75 29 is equal to 8% and 14%, respectively, of the average annual net energy utility’s precise tariff structures, which typically include volumetric 76 30 consumption of sampled households. We compute 95% confidence tiered rates, seasonal rates, and other subtleties that would be 77 31 intervals for the corresponding population means using a Student’s difficult to summarize here. We refer the reader to the citations 78 32 t-test, resulting in an estimated mean additional annual energy provided for each tariff in Table1. 79 33 consumption of 338 ± 14 kWh under the ‘target zero’ operating In general, using storage to increase solar self-consumption 80 34 scenario and 572 ± 19 kWh under the ‘minimize power’ operating provides a financial benefit when the consumption tariff is 81 35 scenario. Supplementary Figs 9–11 show how changing the power higher than the feed-in tariff. The maximum present value that 82 −1 36 capacity, energy capacity, and efficiency of the storage system could be realized in Texas is US$95 kWh of storage capacity. 83 37 considered affects the mean increase in energy consumption. Note If Texans were exposed to Maui’s electric rates, the maximum 84 −1 38 that the average increase in energy consumption caused by adding would increase to US$287 kWh . The minimum present value 85 −1 39 storage is small compared with the average decrease from adding observed under Texas electricity tariffs is −US$60 kWh . If 86 40 solar panels in the first place, as discussed in Supplementary Note 2 Texans were exposed to California’s electric rates, the minimum 87 −1 41 and illustrated in Supplementary Fig. 32. Thus, energy storage that would fall to −US$143 kWh . The installed cost of a lithium- 88 42 directly enables rooftop photovoltaic panels could lead to a decrease ion battery system used for residential applications ranges from 89 −1 43 in net household energy consumption, although energy storage is approximately US$700–US$1,800 kWh of storage capacity, where 90 6–8 −1 44 typically not required . US$700 kWh represents a low cost estimate for a market-leading 91

NATURE ENERGY | www.nature.com/natureenergy 3 ARTICLES NATURE ENERGY

Table 1 | Economic impacts of adding energy storage.

‘Target zero’ ‘Minimize power’ Electric utility Tari name Approximate Approximate Average Ten-year Average Ten-year consumption feed-in tari annual present value annual present value 1 −1 −1 tari (US kWh− ) benefit (US$ kWh ) benefit (US$ kWh ) − − − (US kWh 1) ¢ (US$ yr 1) (US$ yr 1) Austin Energy Value of Solar27,28 9.0¢ 10.9 −31 −32 −57 −58 San Antonio − CPS Energy Net Energy Metering29,30 9.8 9.8 −27 −27 −49 −50 MP2 Energy + Oncor Net Energy Metering31–33 10.1 10.1 −34 −35 −58 −60 MP2 Energy + Oncor Net Energy Buyback33,34 8 3.6 78 80 57 58 MP2 Energy + Centerpoint Net Energy Buyback33,34 8.8 3.6 93 95 70 72 TXU Energy + Oncor Clean Energy Credit33,35,36 11.4 7.5 50 51 24 25 TXU Energy + Centerpoint Clean Energy Credit33,35,37 9.9 7.5 21 21 −1.9 −1.9 HECO − Hawaii/Oahu Customer Grid Supply38,39 22.4 15.1 107 110 58 59 HECO − Maui Customer Grid Supply38,40 33.8 17.16 280 287 207 212 HECO − Molokai Customer Grid Supply38,41 39.0 24.07 251 258 165 169 HECO − Lanai Customer Grid Supply38,42 43.4 27.88 233 239 140 143 California − PG&E Net Energy Metering43,44 28.1 28.1 −72 −74 −140 −143 California − SDG&E Net Energy Metering45,46 27.9 27.9 −71 −73 −139 −143 California − SCE Net Energy Metering47,48 20.9 20.9 −55 −57 −105 −107 We use utility taris from Texas, Hawaii and California to show how storage operating as described in this paper would aect the cost of electricity service to customers. Note that the Hawaii and California cases are shown only for illustration, because storage would see dierent load and generation profiles than those observed from our sample of Texas households. The installed price of a home energy storage system would have to fall below US$100 kWh−1 to provide a benefit under current Texas electricity taris.

−1 1 storage vendor in 2016 and US$1,800 kWh represents the high 0.24 ± 0.02 kg NOx per household per year. The complete range of 40 2 cost estimate reported to the US Department of Energy in 2013 emissions impacts across the 99-household sample is illustrated in 41 49,50 3 for its Energy Storage Handbook . Thus, under no scenario Fig.4. A sensitivity analysis of emissions impacts to energy storage 42 4 considered here could storage provide sufficient direct economic system parameters is provided in the Supplementary Information. 43 Q.3 5 benefit to the customer to offset its upfront cost. The installed Note that while adding storage to homes with existing solar panels 44 −1 6 price would have to fall below US$100 kWh of storage capacity leads to an increase in emissions on average, the observed increase 45 7 to provide a benefit under current Texas electricity tariffs. Details is smaller than the average decrease in emissions caused by adding 46 8 of the economic calculations carried out to obtain the data given solar panels in the first place, as discussed in Supplementary Note 2 47 9 in Table1 are provided in the Methods. The complete range of and illustrated in Supplementary Figs 33–35. However, energy 48 6–8 10 customer benefits calculated across our 99-household sample for storage is typically not required to integrate solar panels . 49 11 each utility tariff is provided in Supplementary Figs 14–27. The change in grid emissions from the addition of home battery 50 energy storage is caused by two separate factors: the additional 51 12 Electricity system emissions impacts energy consumption required to cover storage inefficiencies, and the 52 13 In addition to the impact that home energy storage has on electricity fact that storage shifts electricity demand in time, and alters which 53 14 demand and consumption, we also consider the indirect impact it generators are used to provide energy not produced from rooftop 54 15 would have on electricity system emissions. As the households in solar panels. 55 16 the data set are located in Texas, we use marginal emissions factors To gauge how much of the emissions impact of home energy 56 17 calculated for the Texas electricity system from US Environmental storage is caused by its energy consumption versus its temporal 57 18 Protection Agency emissions monitoring data to approximate the impact on electricity demand, we test the sensitivity of the CO2, SO2 58 19 change in emissions associated with the hourly changes in electricity and NOx emissions impact to storage system a.c.–a.c. roundtrip effi- 59 51,52 20 demand caused by home energy storage . These data report the ciency. The results of this analysis are illustrated in Fig.5. The mean 60 21 emissions in kilograms of CO2, SO2 and NOx per megawatt hour emissions impacts calculated across our sample of 99 households are 61 22 of marginal change in electricity consumption at 5% quantiles of illustrated by the solid lines, and 95% confidence intervals of the 62 23 fossil generation online measured in gigawatts. We use these data corresponding population means are illustrated by the shaded areas 63 24 to approximate marginal emissions factors for each hour of the around each line. We also calculated the sensitivity of the emissions 64 25 year by comparing them with the measured hourly level of fossil impacts to the storage system’s power capacity and energy capacity. 65 53 26 generation in the Texas electricity system over 2014 . The resulting These results are provided in Supplementary Figs 12 and 13. 66 27 hourly marginal emissions factors are used to calculate how adding Under the ‘target zero’ operating scenario, storage could reduce 67 28 home energy storage would impact annual emissions for each SO2 and NOx emissions if its a.c.–a.c. roundtrip efficiency exceeds 68 29 household considered. The Methods discusses these calculations in 90%. However, it could not reduce CO2 emissions unless its ef- 69 30 detail. Marginal emissions factors for the Texas electricity system are ficiency approaches 100%. The sensitivity of emissions to energy 70 31 provided in Supplementary Figs 28–31. efficiency changes under the ‘minimize power’ operating scenario 71 32 We find that the addition of energy storage to a household with because the battery charges and discharges at different times of day. 72 33 existing rooftop solar panels in the Texas electricity system would Under this operating scenario, storage could reduce SO2 emissions 73 34 increase annual emissions of CO2, SO2 and NOx for an average if a.c.–a.c. roundtrip efficiency exceeds 95%, but it could not reduce 74 35 household. When storage operates under the ‘target zero’ mode, NOx emissions even if efficiency equals 100%, because it shifts more 75 36 its mean emissions impact is 160 ± 7 kg CO2, 0.05 ± 0.02 kg SO2 energy production to combustion turbines. For refer- 76 37 and 0.05 ± 0.01 kg NOx per household per year. When storage ence, reported a.c.–a.c. roundtrip efficiencies from Li-ion energy 77 38 operates under the ‘minimize power’ mode its mean emissions storage vendors range from 80–93%, although these estimates might 78 39 impact increases to 290 ± 13 kg CO2, 0.16 ± 0.04 kg SO2 and be optimistic because they are reported values and not measured 79

4 NATURE ENERGY | www.nature.com/natureenergy NATURE ENERGY ARTICLES

a 35 ‘Target zero‘ operation from storage leads to an increase in customers’ utility bills, as 33 30 ‘Minimize power‘ operation shown in Table1. However, in areas where the tariff charged for 34 25 Sample means 20 Population means 95% CI consumption is higher than the feed-in tariff for solar energy, 35 15 the addition of storage can provide an economic benefit to the 36 10 Number of households 5 customer, despite the fact that it leads to higher consumption overall. 37 0 These instances mirror the current situation in Germany, where 38 0 100 200 300 400 500 600 the feed-in tariff decreased in 2012 below the rate charged for 39 23 Additional annual CO2 emissions (kg per household) electricity consumption, creating an incentive for home storage . 40 Under current Texas utility tariffs, the base case energy storage 41 b 35 −1 ‘Target zero’ operation system has a maximum present value of US$95 kWh for a typical 42 30 ‘Minimize power’ operation 25 Sample means household. This value is much lower than the current installed 43 20 −1 Population means 95% CI cost of a home energy storage system (US$700–US$1,800 kWh of 44 15 49,50 10 storage capacity ), so storage could not provide a direct economic 45 Number of households 5 benefit to Texas customers under current tariffs. It is worth noting 46 0 that the customer benefit of adding energy storage was found to be 47 −0.2 0.0 0.2 0.4 0.6 0.8 higher when it operates according to the ‘target zero’ approach, even 48 Change in annual SO2 emissions (kg per household) though this approach provides less benefit to the utility in terms 49 of reduced power demand and injections. This finding indicates 50 c 50 ‘Target zero’ operation 51 40 ‘Minimize power’ operation that it might be useful for utilities to institute demand charges for Sample means residential customers with energy storage to explicitly incentivize 52 30 Population means 95% CI 20 reducing power demand and injections. 53

Number of 54 households 10 Because energy storage decreases kilowatt power flows in the 0 utility grid but increases kilowatt hour sales of electric energy, it 55 −0.1 0.0 0.1 0.2 0.3 0.4 0.5 would be in the utility’s interest for consumers with solar panels to 56 install home storage. While adding storage does reduce a customer’s 57 Change in annual NOx emissions (kg per household) reliance on the utility, previous analysis has shown that mass 58 Figure 4 | Emissions impacts of adding energy storage. a–c, The change in defection from the utility is unlikely due to the high cost required 59 electricity system CO2 (a), SO2 (b) and NOx (c) emissions observed across for sufficient photovoltaic panels and battery storage to be 100% 60 the 99-household sample shown using histograms. The dashed lines independent, and the advantages of an interconnected grid that can 61 indicate the mean impacts observed across the sample, and the shaded balance customers’ generation with demand and enable customer- 62 56 areas indicate 95% confidence intervals of the expected mean across all producers to sell excess electric energy . Thus, it is possible that 63 households computed using a Student’s t-test. energy storage could provide a solution to the disruption of utility 64 57 business models with rising use of . Future 65 49 1 from the field . A roundtrip efficiency exceeding 90% would be work should investigate the potential for energy storage to benefit 66 2 difficult to achieve due to losses in the battery pack itself, additional utilities in this way. 67 3 losses in the power conditioning system that converts the battery’s Our findings on the emissions impact of adding energy storage 68 4 d.c. electricity into a.c. electricity suitable for the grid, and supple- to Texas households with existing photovoltaic panels indicate that 69 5 mental energy required for thermal controls to maintain an accept- it would increase overall electricity system CO2, SO2 and NOx 70 6 able battery pack temperature. Thus, adding home energy storage to emissions due to time-shifting of electric demand and the additional 71 7 households with existing photovoltaic panels in Texas would most electric energy required to cover storage system inefficiencies. 72 8 likely lead to an increase in CO2, SO2 and NOx emissions. Changing the local mix would alter the 73 emissions impact of home energy storage, but it is unlikely that 74 9 Discussion and conclusions storage could decrease emissions unless it directly enables new 75 10 Our findings on the power demand impact of home energy storage installation of non-emitting generators or enables production from 76 11 show that it could be a useful tool to reduce the magnitude of non-emitting sources that otherwise would have been curtailed. 77 12 power flows in the utility grid. This reduction would benefit the This finding aligns with previous work that has examined the 78 13 utility in two ways: it would reduce the required capacity of electric system impacts of bulk electricity storage used for price arbitrage 79 54 58 14 delivery equipment such as substations and , and it in wholesale electricity markets . 80 15 would reduce the need for new generation capacity to reliably meet While rooftop photovoltaic systems deployed on the US grid to- 81 55 16 peak electricity demand . These benefits are greater when home day do not require home energy storage, there are limited instances 82 17 energy storage operates in a way that minimizes the magnitude of where storage might enable wider use of photovoltaic panels. For 83 18 households’ individual power flows (our ‘minimize power’ scenario) example, in October 2015 the Public Utilities Commission of Hawaii 84 19 versus when home energy storage operates in a way that seeks ended its net-energy metering program due to concerns about the 85 20 to reduce power flows between the household and the utility to impact of growing use of rooftop solar panels on electric grid opera- 86 59 21 zero whenever possible (our ‘target zero’ scenario). Note that these tions and utility rates . The net-metering tariff was replaced with a 87 22 benefits arise even though storage operates from a customer-centric ‘self-supply’ tariff for customers that use all of their produced solar 88 23 perspective that seeks to minimize customers’ reliance on the energy on site, and a ‘grid-supply’ tariff that pays a price lower than 89 59 24 electric utility. If storage operated to explicitly benefit the utility the retail electric price for energy sent to the grid from solar panels . 90 25 without trying to isolate customers, there would be a greater benefit There is no limit to the number of customers that can install solar 91 26 to the utility. panels under the self-supply tariff, but installations under the grid- 92 59 27 While home energy storage is a useful tool to reduce power supply tariff are capped . Thus, for the case of Hawaii, home storage 93 28 flows in the distribution system, our findings indicate that it could enable more customers to connect solar panels under the self- 94 29 would increase net energy consumption due to energy storage supply tariff, and indirectly decrease electricity system emissions. 95 30 inefficiencies. Under common net-metering tariffs, which credit However, it is worth noting that the rule change in Hawaii was 96 31 customers for solar energy at a rate equal to the rate charged driven by both technical issues associated with integrating solar 97 32 for energy consumption, the increase in energy consumption panels and economic solvency issues related to net-metering tariffs. 98

NATURE ENERGY | www.nature.com/natureenergy 5 ARTICLES NATURE ENERGY

‘Target zero’ operation ‘Minimize power’ operation

600 CO2 0.6 CO2 x SO2 SO2 2

400 , NO 0.4 2 NOx NOx

200 0.2 Change in CO Change in SO

(kg per household−year) 0 (kg per household−year) 0.0

70 75 80 85 90 95 100 70 75 80 85 90 95 100 Battery roundtrip efficiency (%) Battery roundtrip efficiency (%)

Q.4 Figure 5 | Sensitivity of emissions impacts to storage roundtrip eciency. We test the sensitivity of the CO2, SO2 and NOx emissions impact of home energy storage versus its roundtrip eciency by repeating our analysis for values of roundtrip energy eciency ηrt ranging from 70–100%. The solid lines are the mean emissions impacts observed across the 99-household sample, and the shaded areas are 95% confidence intervals calculated for the corresponding population means.

1 Further, Hawaii is a unique case where high electricity prices, excep- recorded value of electricity use is equal to zero is calculated for each of the 50 2 tional solar resources, and a progressive energy policy make rooftop households, and households with greater than 1,440 min (one cumulative day 51 over the year) of electricity use equal to zero are excluded from the data set (22 52 3 solar very attractive to consumers, yet the islands’ small grid and 9 households in total). Finally, we generate time series plots showing the minutely 53 Q.5 4 isolation from the mainland exacerbate integration challenges . level of electricity demand and solar generation measured from each of the 54 5 In the future, regulators, policymakers and other decision- remaining households over 2014. We examine each of these plots visually, and 55 6 makers should seek to evaluate the impacts of energy storage eliminate an additional 13 households from the data set on the basis of 56 7 separately from the impacts of photovoltaic panels or other anomalous solar generation or electricity consumption behaviour. 57 8 renewable energy sources. Because energy storage is an energy After the screening process is complete, the verified data set contains 58 residential electricity use and solar generation data for 99 households, a sufficient 59 9 consumer and not a producer, it would most likely not reduce sample to examine the expected impact of home energy storage. The data set is 60 10 emissions or primary energy consumption unless it directly enables organized into data files that can be queried and then entered into the 61 11 intermittent renewable energy. It would be useful for future research optimization program that is used to plan the operation of home battery energy 62 12 to assess the cost-emissions trade-off associated with using energy storage. Summary statistics for each of the 99 homes from which data were 63 13 storage to integrate intermittent renewable resources versus other collected for this study are provided in Supplementary Table 1. 64 14 sources of flexibility such as , transformer Energy storage operational management. We model home energy storage 65 15 tap-change operations, advanced inverters, controllable loads, and operation using two different methods: a ‘target zero’ approach where the battery 66 16 other sources of power and voltage control. does not have information about the future level of solar generation or electricity 67 demand and seeks to reduce the power flows between the household and the grid 68 17 Methods to zero without regard for how its present actions might affect future reliance on 69 18 Summary. This paper approximates the impacts of home energy storage using a the utility; and a ‘minimize power’ approach where the battery system has perfect 70 19 sequence of five steps. First, electricity use and solar production data for a information about the future level of electricity demand and solar generation over 71 26 20 particular household are downloaded from Pecan Street’s Dataport website . the day, and plans its operation to minimize the magnitude of power flows 72 21 Second, the data are entered into a program that plans the operation of home between the home and the utility grid over the entire day. The details of each of 73 22 energy storage based on the observed level of electric load and solar generation. these operational models are provided in the following sections. 74 23 Third, the revealed charging and discharging behaviour of home energy storage is 24 used to calculate its impact on electrical power demand and energy consumption. ‘Target zero’ operational model. The ‘target zero’ model for operational 75 25 Fourth, electricity tariffs from Texas, Hawaii and California are used to calculate management of home energy storage considers variables defined over three sets: 76 26 the economic impact of home energy storage. Fifth, a data set of marginal H:{1,2, ..., 99}, representing the numerical identifier of the household where the 77 27 emissions factors for the Texas electricity system is used to approximate the energy storage system is operating; d:{1,2, ..., 365}, representing the day of the 78 51,52 28 emissions impact of adding home energy storage . The following subsections year; and m:{1,2, ..., 1440}, representing the minute of the day. 79

29 explain the methods associated with each of these steps in detail. The model considers an energy storage device with a rated power P rated, a 80

rated energy capacity Erated, and a roundtrip efficiency ηrt. P rated and Erated are 81 30 Electricity use and production data. To show how energy storage could respond selected to correspond to the specifications announced for Tesla’s ‘Powerwall’ 82

31 to a household’s electricity consumption and solar production to minimize home battery system for daily cycle applications: P rated = 3.3 kW, Erated = 7 kWh 83 32 interaction with the utility, we use electricity data collected by Pecan Street from (ref.3). We assume a roundtrip a.c.–a.c. efficiency ηrt = 85% based on the 92% 84 Q.6 26,60 33 residential electricity customers . Each of the customers provides their data to d.c.–d.c. efficiency announced for the Powerwall and an assumed d.c.–a.c. 85 3,49,61,62 34 Pecan Street on a voluntary basis. Many of the participants in the data collection converter efficiency of 96% . 86 35 program are knowledgable about energy and take active steps to reduce their The model steps from one minute of the year to the next, and uses the 87 36 energy use and environmental impact. Previous analysis of the study participants following algorithm to decide the level of battery power. First, the target battery 88 37 revealed a negative correlation between the score obtained on an energy power for household H during the day of the year d and the minute of the day m 89 60 38 knowledge quiz and the homeowner’s overall electricity consumption . We is calculated as the instantaneous net power demand of the household during the 90 39 believe the demographics of the participants resemble the demographics of likely current minute, which is equal to the measured level of electricity use 91

40 home energy storage early adopters. P use(H, d, m) minus the measured level of electricity generation P gen(H, d, m) as 92 41 Electricity use and production data are downloaded from Pecan Street’s given in equation (1). The parameters P use(H, d, m) and P gen(H, d, m) come 93 26 42 Dataport website . The data track electricity use and production in kilowatts at a from the home electricity data set discussed in the previous section. Second, the 94 43 one-minute time resolution, allowing us to examine how energy storage could target power is compared with the rated power of the battery system. If the target 95 44 respond to short-duration fluctuations in electricity demand and solar power exceeds the rated power, then the magnitude of the target power is 96 45 production. The data collected from each household are screened according to reduced to the rated power level. Third, the level of stored energy at the end of 97 46 the following algorithm. First, the number of minutes over the year for which no the current minute as a result of the applied target power is calculated. If the level 98 47 recorded value of electricity use or solar generation exists is calculated for each of stored energy violates the battery system’s limits, then the battery power during 99 48 household, and households with missing data points are excluded from the data the minute is set equal to zero. Otherwise, the battery power during the minute is 100 49 set (38 households in total). Second, the number of minutes for which the set equal to the target power level. 101

6 NATURE ENERGY | www.nature.com/natureenergy NATURE ENERGY ARTICLES

1 This control logic is defined in equation (2). Positive values of P bat(H, d, m) An important constraint on the battery system’s operation is that the 5657 2 indicate discharging while negative values indicate charging. The relationship instantaneous amount of energy stored in the battery system must be within its 58

3 between the applied battery power and the level of stored energy during minute rated energy storage capacity, Erated. We impose an equality constraint within the 59 4 m is defined by equations (3)–(6). Equation (3) estimates the instantaneous optimization program to define a dependent variable Ebat(H, d, m), which 60 5 amount of stored energy as a function of the initial energy stored at the represents the amount of energy stored in the battery at the end of minute m. 61

6 beginning of the day, Ebat,i(H, d), and the change in the amount of energy stored The relationship between the decision variable P bat(H, d, m) and the dependent 62

7 during each prior minute of the day, 1Ebat(H, d, m). Equation (4) defines variable Ebat(H, d, m) is defined as given in equations (3)–(6) discussed in the 63 8 Ebat,i(H, d). The initial amount of energy stored at the beginning of the first day previous section. As equation (6) has a conditional definition based on the sign of 64 9 of the year is set equal to one half the system’s rated energy capacity. For the decision variable P bat(H, d, m), it must be defined within the optimization 65 10 subsequent days, it is set equal to the amount of energy stored at the end of the program using either an integer variable or a smooth functional approximation of 66

11 prior day. Equation (5) defines 1Ebat(H, d, m) as a function of the amount of the discontinuous conditional constraint so that the solver can traverse the 67 12 power applied to the battery. The constant 1t represents the duration of the solution space. We approximate equation (6) using the smooth, continuous 68 13 one-minute time step, and is set equal to (1/60) to integrate the kilowatt power hyperbolic tangent function given in equation (8). Supplementary Fig. 1 69 14 flow to/from the battery into kilowatt hours of energy. The constant κ represents compares equation (6) and (8). 70 15 the energy losses during charging and discharging, and is defined in equation (6).   √   √   16 When the battery is discharging, more energy is extracted from the storage device 1 1 1 1 √ κ ≈ √ + ηrt − √ − ηrt tanh(−50Pbat(H,d,m)) (8) 71 17 than is delivered to the grid, so κ is equal to 1/ η rt and greater than 1. When 2 ηrt 2 ηrt 18 the battery is charging, less energy√ enters the storage device than is extracted 72 19 from the grid, so κ is equal to η rt and less than 1. Put together, the values Within the optimization program, the battery power P bat(H, d, m) and the 73

20 defined for κ in equation (6) set the net roundtrip energy storage efficiency to be amount of energy stored in the battery Ebat(H, d, m) are constrained according to 74 21 equal to ηrt and equally impose energy losses on charging and discharging. the technical limits of the energy storage system. The magnitude of the battery 75 58,63 22 Previous work has used a similar formulation to model storage energy losses . power is constrained to be less than or equal to the rated power capacity P rated 76 according to equation (9). The amount of energy stored is constrained to be 77 greater than or equal to zero and less than or equal to the rated battery energy 78 23 Ptarget(H,d,m)=Puse(H,d,m)−Pgen(H,d,m) (1) capacity Erated according to equation (10). 79

−Prated ≤Pbat(H,d,m)≤Prated (9) 80 24 Pbat(H,d,m)

0≤E (H,d,m)≤E (10) 81  ( ( ) ) ( )> bat rated min Ptarget H,d,m ,Prated if Ptarget H,d,m 0 (net consumer)  ≤ 82  and 0 Ebat(H,d,m) 83 25 = max(Ptarget(H,d,m),−Prated) if Ptarget(H,d,m)<0 (net producer) (2) We implement the optimization using the General Algebraic Modeling 64 84  and Ebat(H,d,m)≤Erated System (GAMS) . Within GAMS, the interior point nonlinear optimization  65 0 Otherwise solver is used . 85 The values of the parameters P use(H, d, m), P gen(H, d, m), P rated, Erated and ηrt 86 66 m are passed to GAMS using the R programming package gdxrrw . The parameters 87 X 26 Ebat(H,d,m)=Ebat,i(H,d)+ 1Ebat(H,d,µ) (3) P use(H, d, m) and P gen(H, d, m) come from the data set discussed in the previous 88 = µ 1 section. The parameters P rated and Erated are selected to correspond to the 89 specifications announced for Tesla’s Powerwall home battery system for daily 90  Erated/2 if d =1 cycle applications: P rated = 3.3 kW, Erated = 7 kWh. We assume a roundtrip a.c.–a.c. 91 27 Ebat,i(H,d)= (4) Ebat(H,d −1,m=1,440) if d 6=1 efficiency ηrt = 85% based on the 92% d.c.–d.c. efficiency announced for the 92 3,49,61,62 Powerwall and an assumed d.c.–a.c. inverter efficiency of 96% . 93 R is used to obtain and store the solution to the optimization program 94 28 1Ebat(H,d,m)=−Pbat(H,d,m)κ1t (5) computed by GAMS. Sample input parameters and results for the optimization 95 program are provided in Supplementary Figs 2–5. We solve the optimization 96 program over the set of all households H in parallel. For each household, the 97  √ 1/ ηrt if Pbat(H,d,m)>0 (discharging) optimization problem is solved for each of the 365 days of the year d in series. 98 29 κ = √ (6) ηrt if Pbat(H,d,m)<0 (charging) Calculation of power demand and energy consumption impacts. Once the 99 30 charge–discharge pattern of home energy storage has been calculated for each 100 31 The control logic is applied using a for loop, which steps from one minute to day of 2014 for each of the 99 households in our data set, we can calculate the 101 32 the next and assigns the battery power during minute m according to impact that home energy storage would have on electrical power demand and 102 33 equation (2). Sample input parameters and results from the operational model are annual energy consumption. 103 NS 34 provided in Supplementary Figs 2–5. The aggregate power demand with no storage P grid(d, m) is calculated by 104 summing the electricity use and solar generation measured from each household 105 35 ‘Minimize power’ operational model. Like the ‘target zero’ model, the ‘minimize H according to equation (11). Likewise, the aggregate power demand with 106 S 36 power’ model considers variables defined over three sets: H:{1,2, ..., 99}, storage P grid(d, m) is calculated according to equation (12). The calculated values 107 NS S 37 representing the numerical identifier of the household where the energy storage of P grid(d, m) and P grid(d, m) are illustrated in Fig.2. 108 38 system is operating; d:{1,2, ..., 365}, representing the day of the year; and m:{1,2, 99 39 ..., 1440}, representing the minute of the day. A nonlinear optimization program NS =X − Pgrid(d,m) Puse(H,d,m) Pgen(H,d,m) (11) 109 40 is used to plan the operation of home energy storage. H=1 41 The decision variable for the optimization program is the level of battery 99 S X 42 power P bat(H, d, m) during each minute m of operating day d for household H, = − − Pgrid(d,m) Puse(H,d,m) Pgen(H,d,m) Pbat(H,d,m) (12) 110 43 with negative values of P bat(H, d, m) indicating charging and positive values H=1 44 indicating discharging. 111 45 The objective of the optimization program is to minimize the magnitude of As residential electricity customers are typically billed for their kilowatt hour 112 46 power flows between a household and the grid by scheduling the battery charging consumption, we calculate the impact that the addition of home energy storage 113 47 and discharging power over the day. Thus, we define the objective function to be would have on annual energy consumption for each of the households in our 114 48 minimized as the sum of the squares of the net power flow between household H data set. The change in net energy consumption over the year for each household 115

49 and the grid during each minute m of day d, as given in equation (7). The net 1Econs(H) from the addition of home energy storage is calculated by integrating 116 50 power flow during each minute is equal to the measured level of electricity use the flow of power in and out of the storage device according to equation (13), 117

51 P use(H, d, m), minus the measured level of electricity generation P gen(H, d, m), where 1t is set equal to (1/60) to convert the kilowatt power flows in/out of the 118

52 minus the battery power decision variable P bat(H, d, m). The optimization battery during each minute m into kilowatt hour of energy. Note that a negative 119

53 program is executed separately for each household H and each day d to minimize sign is added to the summation because positive values of P bat indicate 120 54 each household’s interaction with the utility during each day of the year. discharging and negative values indicate charging. 121

1,440 365 1,440 X 2 XX 55 fObj(H,d)= (Puse(H,d,m)−Pgen(H,d,m)−Pbat(H,d,m)) (7) 1Econs(H)= −Pbat(H,d,m)1t (13) 122 m=1 d=1 m=1

NATURE ENERGY | www.nature.com/natureenergy 7 ARTICLES NATURE ENERGY

365 24 21 Figure3 illustrates the change in energy consumption from the addition of XXX 73 1 ( )= − ( )1 ( ) 74 3 home energy storage for each of the 99 households in the data set. The mean SO2 H Pbat H,d,m tMEFSO2 h (16) d=1 h=1 m∈h 4 additional energy consumption is calculated by taking the average of 1E(H) over 5 all H, and a 95% confidence interval for the mean is calculated using a Student’s ◦ 365 24 6 t-test with 98 of freedom. =XXX− 1NOx (H) Pbat(H,d,m)1tMEFNOx (h) (17) 75 d=1 h=1 m∈h 7 Calculation of economic impacts. To assess the impact that adding energy 8 storage would have on the annual cost of service under different electricity tariffs, 76

9 we first calculate the monthly energy consumption from the grid and energy The result of equations (15)–(17) is an estimate of the annual change in CO2, 77 10 injection to the grid for each of the 99 households in our data set with and SO2 and NOx emissions in kilograms caused by the addition of home energy 78 11 without energy storage added. Then, we create custom functions using the R storage. Note that no emissions credit is given for grid energy consumption offset 79 12 programming package that calculate the annual cost of service as a function of by the solar panels because our objective is to analyse the impact of adding 80 13 monthly consumption and production for each of the 14 utility tariffs identified energy storage to a household with existing solar panels. The emissions impact of 81 67 14 in Table1 . While some of the tariffs charge a fixed US cent per kilowatt hour home energy storage for each household in our sample is shown in Fig.4. The 82

15 rate for energy consumed and energy produced, most of the utility tariffs include mean emissions impact is calculated by taking the average of the 1CO2(H), 83

16 volumetric tiered charges, seasonal peak charges, or other subtleties that would be 1SO2(H) and 1NOx(H) over each of the 99 households in our sample H. We 84 17 difficult to compactly summarize here. We refer the reader to the utility tariffs calculate a 95% confidence interval on the population mean using a Student’s 85 27–48 ◦ 18 cited in the references and identified in Table1 . We use the custom R t-test with 98 of freedom. 86 19 functions to calculate the annual cost of service for each of the 99 households in 20 our data set with and without energy storage, and then calculate the annual Data availability. To calculate the results presented in this paper, we draw on the 87 21 benefit from energy storage as the difference between the cost of service with no database of household electricity use and production available through Pecan 88 26 22 energy storage installed and the cost of service with energy storage installed. The Street’s Dataport website . Supplementary Table 1 provides the unique data 89 23 annual benefit for each of the 99 households under each of the utility tariffs identifiers and summary statistics for each of the 99 households considered in 90 24 considered is illustrated in Supplementary Figs 14–27. this study. The marginal emissions factors data used to calculate the emissions 91 25 Once the average annual benefit from the addition of energy storage impacts presented in Fig.4 are available online from the Carnegie Mellon Center 92 51,52 26 is calculated for each one of the utility tariffs considered and for both storage for Climate and Energy Decision Making . Additionally, the US Environmental 93 27 operational modes considered, we calculate the present value of the energy storage Protection Agency Continuous Emissions Monitoring System (CEMS) data 94 28 system assuming a ten-year lifetime, a 10% discount rate, and a 2.5% inflation originally used to calculate marginal emissions factors are available online 95 3,68 69 29 rate . These assumptions result in a present worth factor of 7.17, as given through the Air Markets Program Data website . Any intermediate data not 96 30 in equation (14). We multiply this present worth factor by each average annual available from the sources described above, and not included in this article 97

31 benefit in Table1 and then divide by the base case energy capacity Erated = 7 kWh or its Supplementary Information, are available from the authors 98 32 to obtain the present value in US dollars per kilowatt hour for each utility tariff. on request. 99

10 y−0.5 X (1+0.025) Received 8 March 2016; accepted 3 January 2017; 100 33 PW = =7.17 (14) factor y−0.5 (1+0.10) published XX Month XXXX 101 34 y=1

35 Marginal emissions data and estimation of emissions impacts. To calculate the References 102 36 emissions impact of adding home energy storage to households with existing 1. Walton, R. Residential Energy Storage: The Industry’s Next Big Thing (2015); 103 37 solar panels, it is important to consider which generating units would respond to http://www.utilitydive.com/news/residential-energy-storage-the-industrys- 104 38 a change in electricity demand at the particular times when energy storage next-big-thing/406789 105 39 charges and discharges, because those generators might have different emissions. 2. Sonnenbatterie Sonnenbatterie Enters US Market with First Distribution Deal 106 40 Electricity generators are scheduled by the grid operator according to their (2014); http://www.sonnenbattery.com/en/press/news/news/article/ 107 41 marginal operating cost, with the least costly generators dispatched first to sonnenbatterie-enters-us-market-with-first-distribution-deal 108 42 minimize overall production costs. As a rule of thumb, wind, solar, hydroelectric 3. Tesla Motors Powerwall Tesla Home Battery (2015); 109 43 and nuclear generation are dispatched first, followed by , and then natural gas https://www.teslamotors.com/powerwall 110 44 generation. The precise order that generators are dispatched depends on a 4. Green Mountain Power Green Mountain Power Files First in the Country 111 45 number of factors, such as their individual efficiency, fuel price, maintenance Innovative Plan to Offer Vermonters the Home Battery (2015). 112 Q.7 46 requirements, electricity transmission constraints, and other factors. 5. Sungevity & Sonnenbatterie Sungevity and Sonnenbatterie Announce 113 47 To estimate the emissions associated with the marginal change in electricity Partnership to Offer Home Energy Storage (2015); 114 48 demand caused by home energy storage, we use marginal emissions factors http://www.sungevity.com/article/?Id=a6MU00000004DQLMA2&_ga= 115 49 calculated for the Texas electricity system from US Environmental Protection 1.97172890.465704836.1444095635 116 51,52 50 Agency Continuous Emissions Monitoring System (CEMS) data . These data 6. Cohen, M. & Callaway, D. Effects of distributed PV generation on California’s 117 51 estimate the change in CO2, SO2 and NOx emissions associated with a change in distribution system, Part 1: engineering simulations. Sol. Energy (2016). 118 Q.8 52 electricity demand at 5% quantiles of fossil generation online. These data show 7. Cohen, M., Kauzmann, P. & Callaway, D. Effects of distributed PV generation 119 53 that when the total fossil generation online is at its minimum, the marginal CO2 on California’s distribution system, part 2: economic analysis. Sol. Energy 120 54 and SO2 rates are at their maximum because more of the generators responding (2016). 121 55 to a marginal change in electricity generation are coal steam units. Likewise, 8. Nguyen, D. A. et al. Impact Research of High Photovoltaics Penetration Using 122 56 when the total fossil generation online is at its maximum, the marginal NOx rate High Resolution Resource Assessment with Sky Imager and Power System 123 57 is at its maximum because more of the generators responding to a marginal Simulation (2015); http://calsolarresearch.ca.gov/images/stories/documents/ 124 58 change in electricity demand are natural gas combustion turbines. The data are Sol3_funded_proj_docs/UCSD/CSIRDD-Sol3_UCSD_Task4_3_20151121.pdf 125 59 available online from the Carnegie Mellon Center for Climate and Energy 9. Braun, M. et al. Is the distribution grid ready to accept large-scale photovoltaic 126 52 60 Decision Making, and provided in Supplementary Fig. 28 . We estimate hourly deployment? State of the art, progress, and future prospects. Prog. Photovolt. 127 61 marginal emissions factors by comparing the recorded hourly level of fossil Res. Appl. 20, 681–697 (2012). 128 62 generation over 2014 in Texas to the marginal emission factors calculated for 10. Smith, J. W., Sunderman, W., Dugan, R. & Seal, B. Smart inverter volt/var 129 63 various levels of fossil generation and interpolating linearly. The hourly marginal control functions for high penetration of PV on distribution systems. in 2011 130 64 emissions factors are illustrated in Supplementary Figs 29–31. IEEE/PES Power Systems Conference and Exposition, PSCE 2011 1–6 (2011). 131 Q.9 65 Once hourly marginal emissions factors for CO2, SO2 and NOx have been 11. Stetz, T., Kraiczy, M., Braun, M. & Schmidt, S. Technical and economical 132 66 estimated, we calculate the annual emissions impact of adding home energy assessment of voltage control strategies in distribution grids. Prog. Photovolt. 133 67 storage to households with existing solar panels according to equations (15)–(17), Res. Appl. 21, 1292–1307 (2013). 134 68 where MEF(h) is the marginal emissions factor in kilograms per kilowatt hour at 12. Von Appen, J., Braun, M., Stetz, T., Diwold, K. & Geibel, D. Time in the sun: the 135 69 hour h, P bat(H, d, m) is the battery power for household H on day d at minute m challenge of high PV penetration in the German electric grid. IEEE Power 136 70 in the corresponding hour h, and 1t is set equal to (1/60) to convert the kilowatt Energy Mag. 11, 55–64 (2013). 137 71 power flow in and out of the battery system into kilowatt hours. 13. Büchner, J. et al. Smart Grids in Germany: How Much Costs Do Distribution 138 Grids Cause at Planning Time? 224–229 (2015). 139 365 24 XXX 14. Stetz, T. et al. Techno-economic assessment of voltage control strategies in low 140 72 1CO (H)= −P (H,d,m)1tMEF (h) (15) 2 bat CO2 voltage grids. IEEE Trans. 5, 2125–2132 (2014). 141 d=1 h=1 m∈h

8 NATURE ENERGY | www.nature.com/natureenergy NATURE ENERGY ARTICLES

1 15. Castillo-Cagigal, M. et al. PV self-consumption optimization with storage and 41. Hawaiian Electric Company HECO Molokai Electric Rate Schedule 78 2 Active DSM for the residential sector. Sol. Energy 85, 2338–2348 (2011). R—Residential Service (2016); https://www.hawaiianelectric.com/Documents/ 79 3 16. Braun, M., Büdenbender, K., Magnor, D. & Jossen, A. Photovoltaic my_account/rates/maui_electric_rates_molokai/molokai_rates_sch_r.pdf 80 4 self-consumption in Germany—using lithium-ion storage to increase 42. Hawaiian Electric Company HECO Lanai Electric Rate Schedule R—Residential 81 5 self-consumed photovoltaic energy. in 24th European Photovoltaic Solar Energy Service (2016); https://www.hawaiianelectric.com/Documents/ 82 6 Conference 2009 3121–3127 (2009). my_account/rates/maui_electric_rates_lanai/lanai_rates_sch_r.pdf 83 7 17. Mulder, G., Ridder, F. D. & Six, D. Electricity storage for grid-connected 43. Pacific Gas and Electric Understand Net Energy Metering (NEM)and Your Bill 84 8 household dwellings with PV panels. Sol. Energy 84, 1284–1293 (2010). (2016); http://www.pge.com/en/myhome/saveenergymoney/solar/ 85 9 18. Lang, T., Ammann, D. & Girod, B. Profitability in absence of subsidies: nembill.page?WT.mc_id=Vanity_nem 86 10 a techno-economic analysis of rooftop photovoltaic self-consumption in 44. Pacific Gas and Electric Residential Electric Rates (2016); 87 11 residential and commercial buildings. Renew. Energy 87, 77–87 (2016). http://www.pge.com/tariffs/electric.shtml#RESELEC 88 12 19. Hoppmann, J., Volland, J., Schmidt, T. S. & Hoffmann, V. H. The economic 45. San Diego Gas and Electric Net Energy Metering Program (2016); 89 13 viability of battery storage for residential solar photovoltaic systems—a review http://www.sdge.com/clean-energy/overview/overview 90 14 and a simulation model. Renew. Sustain. Energy Rev. 39, 1101–1118 (2014). 46. San Diego Gas and Electric Schedule DR—Residential Service (2016); 91 15 20. Resch, M., Ramadhani, B., Bühler, J. & Sumper, A. Comparison of control https://www.sdge.com/sites/default/files/regulatory/1-1- 92 16 strategies of residential PV storage systems. in 9th Int. Renewable Energy 16 Schedule DR Total Rates Table.pdf 93 17 Storage Conf. (IRES 2015)18 (2015). 47. Southern California Edison Net Energy Metering (2016); 94 18 21. Santos, J. M., Moura, P. S. & de Almeida, A. T. Technical and economic impact https://www.sce.com/NEM 95 19 of residential electricity storage at local and grid level for Portugal. Appl. Energy 48. Southern California Edison Schedule D—Domestic Service (2016); 96 20 128, 254–264 (2014). https://www.sce.com/NR/sc3/tm2/pdf/ce12-12.pdf 97 21 22. 247 How It Works (2016); http://www.powerstation247.com/ 49. Akhil, A. A. et al. DOE/EPRI 2013 Electricity Storage Handbook in Collaboration 98 22 how-it-works/storing-solar-energy-with-your-powerstation.html with NRECA Tech. Rep. (Sandia National Laboratories, 2013). 99 23 23. von Appen, J., Braslavsky, J. H., Ward, J. K. & Braun, M. Sizing and grid impact 50. Randall, T. Tesla’s New Battery Doesn’t Work That Well With Solar (2015); 100 24 of PV battery systems—a comparative analysis for Australia and Germany. in http://www.bloomberg.com/news/articles/2015-05-06/tesla-s-new-battery- 101 25 2015 Int. Symp. Smart Electric Distribution Systems and Technologies (EDST) doesn-t-work-that-well-with-solar 102 26 612–619 (IEEE, 2015); 51. Siler-Evans, K., Azevedo, I. L. & Morgan, M. G. Marginal emissions factors for 103 27 http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=7315280 the U.S. electricity system. Environ. Sci. Technol. 46, 4742–4748 (2012). 104 28 24. Moshövel, J. et al. Analysis of the maximal possible grid relief from 52. Siler-Evans, K., Azevedo, I. M. L. & Morgan, M. G. Marginal Emissions Factors 105 29 PV-peak-power impacts by using storage systems for increased Repository (2012); 106 30 self-consumption. Appl. Energy 137, 567–575 (2015). http://cedmcenter.org/tools-for-cedm/marginal-emissions-factors-repository 107 31 25. McKenna, E., McManus, M., Cooper, S. & Thomson, M. Economic and 53. U.S. Environmental Protection Agency Regional Data Files for 2014—Texas; 108 32 environmental impact of lead-acid batteries in grid-connected domestic PV http://www3.epa.gov/avert/avert-download.html 109 Q.10 33 systems. Appl. Energy 104, 239–249 (2013). 54. Willis, H. L. Power Distribution Planning Reference Book 2nd edn (Marcel 110 34 26. Pecan Street Incorporated Pecan Street Dataport (2015); Dekker, 2004). 111 35 https://dataport.pecanstreet.org 55. Eyer, J. & Corey, G. Energy Storage for the Electricity Grid: Benefits and Market 112 36 27. Austin Energy Residential Solar Energy Rate—Value of Solar (2016); Potential Assessment Guide A Study for the DOE Energy Storage Systems 113 37 http://austinenergy.com/wps/portal/ae/rates/residential-rates/residential-solar- Program Tech. Rep. (Sandia National Laboratories, 2010). 114 38 energy-rate 56. Khalilpour, R. & Vassallo, A. Leaving the grid: an ambition or a real choice? 115 39 28. Austin Energy City of Austin Electric Tariff (2016); Energy Policy 82, 207–221 (2015). 116 40 http://austinenergy.com/wps/wcm/connect/ab6d045c-643e-4c16-921f- 57. Kind, P. Disruptive Challenges: Financial Implications and Strategic Responses to 117 41 c76fa0fee2bf/FY2016aeElectricRateSchedule.pdf?MOD=AJPERES a Changing Retail Electric Business Tech. Rep. (Edison Electric Institute, 2013); 118 42 29. CPS Energy Solar Billing Facts (2016); https://www.cpsenergy.com/content/ http://www.eei.org/ourissues/finance/documents/disruptivechallenges.pdf 119 43 dam/corporate/en/Documents/EnergyEfficiency/solar_billing_facts.pdf 58. Hittinger, E. S. & Azevedo, I. M. L. Bulk energy storage increases United States 120 44 30. CPS Energy Residential Service Electric Rate (2016); electricity system emissions. Environ. Sci. Technol. 49, 3203–3210 (2015). 121 45 https://www.cpsenergy.com/content/dam/corporate/en/Documents/ 59. Public Utilities Commission of Hawaii Distributed Energy Resources (Docket 122 46 Rate_ResidentialElectric.pdf No. 2014-0192) (2015); http://puc.hawaii.gov/wp- 123 47 31. MP2 Energy 12-Month Solar Offer for Customers (2015); content/uploads/2015/10/DER-Phase-1-DO-Summary.pdf 124 48 https://www.mp2energy.com/pdf/12 MONTH EFL_RESI_NEM_ 60. Rhodes, J. D. et al. Experimental and data collection methods for a 125 49 SCTY_093015.pdf large-scale smart grid deployment: methods and first results. Energy 65, 126 50 32. MP2 Energy 24-Month Solar Net Metering Offer for Customers (2015); 462–471 (2014). 127 51 https://www.mp2energy.com/pdf/24 MONTH EFL_RESI_NEM_ 61. Gyuk, I. & Eckroad, S. & U.S. Department of Energy, and The 128 52 SCTY_093015.pdf Research Institute EPRI-DOE Handbook of Energy Storage for Transmission and 129 53 33. Public Utility Commission of Texas Summary of Current Commission-Approved Distribution Applications EPRI-DOE, Tech. Rep. (2003). 130 54 Charges for ERCOT TDUs (2016); https://www.puc.texas.gov/industry/electric/ 62. Ideal Power Ideal Power 30 kW Battery Converter Specification (2015); 131 55 rates/Trans/TDArchive/TDGenericRateSummary_030116.pdf http://www.idealpower.com/wp-content/uploads/2015/11/30kW-battery- 132 56 34. MP2 Energy Solar Buyback—24 Month (2015); converter-RevC-3Nov2015.pdf 133 57 https://www.mp2energy.com/pdf/NEB_EFL_Oct 2015_60 mos_FINAL.pdf 63. Hittinger, E., Whitacre, J. F. & Apt, J. What properties of are 134 58 35. TXU Energy TXU Energy Clean Energy Credit Program for Surplus Distributed most valuable? J. Power Sources 206, 436–449 (2012). 135 59 Renewable Generation (2015); https://www.txu.com/savings- 64. GAMS Development Corporation General Algebraic Modeling System (GAMS) 136 60 solutions/renewable-energy/renewable-buyback.aspx Release 24.2.1 (2013); http://www.gams.com 137 61 36. TXU Energy TXU Energy Simple Rate 12—Oncor Service Area (2016); 65. Wächter, A. & Biegler, L. T. On the implementation of an interior-point filter 138 62 https://www.txu.com/Handlers/PDFGenerator.ashx?comProdId= line-search algorithm for large-scale nonlinear programming. Math. Program. 139 63 ONXSIMRTND12AE&lang=en&formType=EnergyFactsLabel&custClass= 106, 25–57 (2006). 140 64 3&tdsp=ONCOR 66. Jain, R. & Dirkse, S. gdxrrw: An Interface between GAMS and R (2014). 141 65 37. TXU Energy TXU Energy Simple Rate 12—CenterPoint Energy Service Area 67. R Core Team R: A Language and Environment for Statistical Computing 142 66 (2016); https://www.txu.com/Handlers/PDFGenerator.ashx?comProdId= (R Foundation for Statistical Computing, 2015); https://www.r-project.org 143 67 CPXSIMRTND12AG&lang=en&formType=Disclaimer&custClass= 68. Schoenung, S. M. & Eyer, J. Benefit/Cost Framework for Evaluating Modular 144 68 3&tdsp=CENTERP Energy Storage Tech. Rep. (Sandia National Laboratories, 2008); 145 69 38. Hawaiian Electric Company Customer Grid Supply and Self Supply Programs http://prod.sandia.gov/techlib/access-control.cgi/2008/080978.pdf 146 70 (2016); https://www.hawaiianelectric.com/clean-energy-hawaii/producing- 69. U.S. Environmental Protection Agency Air Markets Program Data (2016); 147 71 clean-energy/customer-grid-supply-and-self-supply-programs https://ampd.epa.gov/ampd 148 72 39. Hawaiian Electric Company HECO Hawaii and Oahu Electric Rate Schedule 73 R—Residential Service (2016); https://www.hawaiianelectric.com/Documents/ 74 my_account/rates/hawaiian_electric_rates/heco_rates_sch_r.pdf Acknowledgements 149 75 40. Hawaiian Electric Company HECO Maui Electric Rate Schedule R—Residential This work was sponsored by Pecan Street, the Electric Reliability Council of Texas 150 76 Service (2016); https://www.hawaiianelectric.com/Documents/ (ERCOT), and the University of Texas Energy Institute. Special thanks to R. Baldick for 151 77 my_account/rates/maui_electric_rates_maui/maui_rates_sch_r.pdf his questions, which helped shape this work. 152

NATURE ENERGY | www.nature.com/natureenergy 9 ARTICLES NATURE ENERGY

1 Author contributions How to cite this article: Fares, R. L. & Webber, M. E. The impacts of storing solar energy 10 2 R.L.F. identified the research question, curated the data used, designed the research in the home to reduce reliance on the utility. Nat. Energy 2, 17001 (2017). 11 3 methods, analysed the numerical results, and prepared the manuscript. M.E.W. 4 contributed to identifying the research question, interpreted the results, prepared the Competing interests 12 5 manuscript, and provided institutional and material support for the research. This work was sponsored in part by the University of Texas Energy Institute, which has a 13 number of internal and external funding sources. External funding sources include oil 14 6 Additional information and gas producers, investor- and publicly owned electric utilities, and environmental 15 7 Supplementary information is available for this paper. non-profits that might be perceived to influence the results and/or discussion reported in 16 this paper. A complete list of Energy Institute sponsors is available online 17 8 Reprints and permissions information is available at www.nature.com/reprints. (http://energy.utexas.edu/mission/sponsors-financial-support). Only the authors were 18 9 Correspondence and requests for materials should be addressed to R.L.F. directly involved with developing the manuscript. 19

10 NATURE ENERGY | www.nature.com/natureenergy Queries for NPG paper nenergy20171

Page 1 Query 1: According to style, author names (and associated dates) should be avoided in the text unless the contribution of an author (or the date) must be emphasized. Could ‘McKenna et al’s 2013 study, which’ be changed to ‘a study that’ or ‘a 2013 study that’ here? (Note, if the name must be retained, the text should be amended to avoid the possessive apostrophe after ‘et al.’.)

Page 2 Query 2: According to style, we should use ‘program’ for brain, computer, genetics or cells, and ‘programme’ otherwise (unless these are part of capitalized proper names). As ‘program’ seems to be used in different contexts throughout, please check/amend (using the search function in Adobe Reader) the spelling to follow this rule.

Page 4 Query 3: Please amend the text to specify where the sensitivity analysis can be found in the Supplementary Information.

Page 6 Query 4: Panel letters a and b removed from figure 5, as these are not mentioned in the caption or the text. OK? Query 5: Would ‘minutely’ be better as, for example, ‘per minute’ or ‘minute by minute’? Query 6: Please note that reference numbers are formatted according to style in the text, so that any reference numbers following a symbol or acronym are given as ‘ref. XX’ on the line, whereas all other reference numbers are given as superscripts.

Page 8 Query 7: Please provide publisher for ref. 4. Query 8: Please provide volume and page range for refs 6 and 7. Query 9: Please provide publisher for refs 10, 13, 16, 20, 61 and 66.

Page 9 Query 10: The URLs in refs 53 and 62 seem to generate errors; please check.