How to Size a Commercial Battery from 15-Minute Load Data

Size a commercial battery from one row per 15-minute interval, not from a single peak number: subtract the demand cap from each interval, sum only the positive differences into a running kWh total, then size discharge power and nameplate energy separately. In the illustrative worked example, four intervals above a 320 kW cap need 80 kW of AC discharge power and 47.5 kWh of delivered energy. Derating gives 88.3 kWh of nameplate, yet the power-to-energy check pushes a 0.5P cabinet to at least 160 kWh. The worksheet is reproducible in a spreadsheet, so a buyer can hand the same rows to a supplier and compare quotes on equal terms. Templates sit in the resources hub.

Key takeaways

• Size from a 15-minute worksheet: subtract the cap, keep positive differences, carry a running cumulative kWh.
• Divide by usable fraction, one-way discharge-path efficiency and end-of-life SOH, not round-trip efficiency.
• The illustrative example needs 88.3 kWh by energy, but at least 160 kWh by power on a two-hour cabinet.
• Guaranteed peak shaving plus backup means adding reserves; the same kWh cannot serve both duties.

Why a single peak number is not a sizing method

Demand is billed on an averaged interval, not an instantaneous spike. A U.S. Department of Energy guide shows 10 kW held for 15 minutes registers 10 kW of demand and 2.5 kWh, while a 36 kW spike integrates to 28 kW, and the demand register resets monthly (U.S. DOE, 2024). One peak number hides shape and duration.

Ratchets carry that error forward. The same guide bills the greater of actual demand or 85% of the highest June-September peak for the following 11 months (U.S. DOE, 2024). A mild-month reading is no guide to a billed month.

Intervals vary. NREL lists 15, 30 or 60-minute windows (NREL, 2015), and its 2017 survey defines the charge as the highest average use inside a set interval, usually 15 minutes (NREL, 2017). Its tariff counts and bill-share figures are from 2017 and are not used as current values here.

An assumed duration is the weakest shortcut. Table 1 reproduced in a U.S. DOE report puts demand-charge management at 50 kW-10 MW, 1-4 hours and 50-500 cycles per year, citing the Sandia / DOE / EPRI Electricity Storage Handbook, SAND2015-1002 (2015). The illustrative event below lasts one hour.

What you need before you start

Collect inputs before opening a spreadsheet. Twelve months of 15-minute data is the target: an LBNL study of 537 commercial buildings reached median CV(RMSE) under 25% for every baseline model with 12 months of training data (LBNL, 2016). A week of logging gives only a first pass. Export UTC timestamps so clock-change duplicates stay visible. Gather the single-line diagram, the critical-load list and tariff documents including the ratchet clause. Define and document the baseline before the change, as DOE FEMP M&V Guidelines v5.0 requires (DOE FEMP, 2024). IEC 62933-3-1:2025 calls for preliminary and final sizing with a duty cycle at the point of connection, this worksheet’s boundary.

Step 1: Set the cap from the tariff, not from the load

Set the cap at the billed demand threshold in the tariff, then check the ratchet floor. The cap is a commercial input, read from the tariff rather than chosen to make the battery look small. Where a ratchet applies, the floor for the following 11 months is 85% of the highest June-September peak (U.S. DOE, 2024), so a mild-month cap will not hold in a billed month. Confirm the monthly reset, since the worksheet is evaluated per billing period (U.S. DOE, 2024). Confirm the averaging interval too: a 320 kW cap on 15-minute data is not the same constraint on 30-minute data (NREL, 2015). The illustrative cap is 320 kW.

Step 2: Build the interval worksheet

One row per interval is the whole method: compute net load minus cap, keep positive values, multiply by dt in hours, and carry a running cumulative total. All load figures below are illustrative.

Net load comes from the meter, cap from Step 1, and delta is net load minus cap clipped at zero. dt is 0.25 h. Delivered kWh = delta x dt is AC energy at the point of connection. Cumulative kWh is a running sum; keep it running across the billing period, since a second peak after partial recharge can exceed any single event.

NYSERDA Report 20-17 (NYSERDA / Clarkson University, 2020) is the nearest documented precedent: a real 15-minute Excel worksheet converting interval kWh to average kW, computing discharge power and a running state of discharge, and taking the maximum state of discharge as the required energy rating. This is the same running sum on a different column.

Step 3: Read power and energy from the worksheet

Two numbers come out of the same column. The largest delta is the required AC discharge power, 80 kW in interval 2. The final cumulative kWh is the delivered energy, 47.5 kWh. IEC 62933-2-1:2017 defines rated input and output active power with a specified duration at the point of connection, so report both as a pair. A calculator asking for “hours of shaving” collapses them into an assumption.

Step 4: Convert delivered kWh to nameplate kWh

Divide by the one-way discharge-path efficiency, not round-trip efficiency. Round-trip efficiency covers a full charge-discharge cycle (IEC 62933-2-1:2017), so it belongs in the charging-energy and operating-cost calculation. PNNL-33283 Table 4.4 assumes one-way inverter and transformer efficiencies of 0.98 each, giving 0.98 x 0.98 = 0.9604 (PNNL / DOE ESGC, 2022). That report covers 1-100 MW grid systems, not C&I cabinets, so replace 0.9604 with the supplier’s measured value.

Round-trip efficiency still has a job. The NREL 2024 Annual Technology Baseline lists 85% round-trip efficiency, 1-8 hour duration, 100-2,000 kW and 15-year life for commercial storage (NREL, 2024). At 85%, recharging the illustrative 47.5 kWh takes about 55.9 kWh of grid energy, an operating-cost figure, not a nameplate derate.

Three stated inputs convert 47.5 kWh to nameplate. The 15%-95% state-of-charge operating window is a worksheet assumption, leaving 0.80 usable. End-of-life state of health of 0.70 is also a stated warranty assumption, not an industry average: PNNL-33283 uses 0.60 for grid-scale systems (PNNL / DOE ESGC, 2022), so the contract value must replace it. Only the one-way efficiency comes from a published measurement.

Keep efficiency boundaries visible. The round-trip efficiency measurement boundary and usable vs nameplate kWh explainers show why a datasheet maximum system efficiency is not a one-way discharge path. Cycle life also depends on depth of discharge: PNNL-33283 reports 2,400 cycles at 80% DoD, 4,500 at 70% and 8,000 at 60% for grid-scale systems, so its values are not cabinet warranties (PNNL / DOE ESGC, 2022).

Step 5: Check the power-to-energy ratio against the cabinet

Power and energy give different answers, and the power-to-energy ratio decides. Divide the 80 kW required discharge by the 88.3 kWh nameplate to get 0.91P. A two-hour cabinet, meaning 100 kW of continuous rating per 200 kWh of nameplate, is a 0.5P unit whose two-hour duration sits inside the 1-8 hour range in the NREL 2024 Annual Technology Baseline (NREL, 2024). It cannot deliver 80 kW at 88.3 kWh, so a 0.5P unit needs at least 80 / 0.5 = 160 kWh.

The governing value is 160 kWh, not 88.3 kWh. Ask the supplier to state the continuous discharge rating at end of life, since a nameplate satisfying only the energy column will still clip interval 2.

Step 6: Size backup separately, then resolve the reserve conflict

Size backup from the critical-load list, not from the peak-shave event. An illustrative 40 kW critical load held for 2 h needs 80 kWh delivered, or 80 / 0.5378 = 148.8 kWh of nameplate. NFPA 110 lists standby runtimes of 5 min, 15 min, 2 h, 6 h and 48 h (NFPA 110 classes reproduced, 2023), but it does not decide which loads require backup. NEC Articles 700, 701 and 702 separate emergency, legally required standby and optional standby loads, so the critical-load list is a code decision.

Guaranteeing both duties means adding the reserves, not taking the maximum. The same kWh cannot be promised to a scheduled tariff event and to an unscheduled outage. Peak shaving consumes energy on a known schedule; backup must stay untouched until the outage arrives.

Adding rather than maxing gives 127.5 kWh delivered and 237.1 kWh nameplate. If both duties can be called at the same moment, the power check must also cover the combined draw, so put that question in the RFQ rather than guessing.

Four data checks that change the answer

Match the tariff interval. NREL lists 15, 30 and 60-minute windows (NREL, 2015), and its 2017 survey ties the charge to the highest average use inside that interval (NREL, 2017). Resample raw data to the billed averaging window before setting the cap. A 15-minute peak compared with a 30-minute billed average overstates what the tariff will record.

Keep daylight-saving duplicates. Elexon notes a long clock-change day has 50 half-hour settlement periods instead of 48 (Elexon, 2025). UTC timestamps expose the duplicate hour. Never average the two readings away.

Log missing intervals instead of inventing values. No Tier 1-3 source was found that prescribes a numeric rule for filling missing intervals in BESS sizing. Record the gap, flag the billing period, and ask the supplier to state its assumption in writing.

Cover a full season and fix the baseline first. An LBNL study of 537 commercial buildings reached median CV(RMSE) under 25% for every model with 12 months of training data (LBNL, 2016). DOE FEMP M&V Guidelines v5.0 requires the baseline documented before the change (DOE FEMP, 2024).

Common mistakes

• Dividing by round-trip efficiency. It double-counts losses on a discharge-only path.
• Sizing from one event. The running sum must span the whole billing period.
• Taking the max of the two duties. Guaranteed peak shaving plus backup needs reserves added.
• Filling gaps with an invented average. No sourced rule supports it; log the gap.

Frequently asked questions

Why not divide by round-trip efficiency?

Round-trip efficiency covers a full charge-discharge cycle, so it double-counts losses on a discharge-only path. Use the one-way discharge efficiency instead: 0.98 x 0.98 = 0.9604 (PNNL / DOE ESGC, 2022). Put round-trip efficiency in the charging-energy and operating-cost column, where the NREL 2024 baseline value of 85% applies (NREL, 2024).

Can one cabinet guarantee peak shaving and backup?

Not from the same kWh. Peak shaving spends energy on a tariff schedule, while backup must stay unused until an outage. If both are guaranteed, add the reserves: the illustrative 47.5 kWh event plus 80 kWh of backup gives 127.5 kWh delivered, or 237.1 kWh nameplate at the same divisor. Taking the maximum leaves one duty unfunded.

What if the meter uses 30-minute intervals?

Resample to the billed averaging window before setting the cap. NREL lists 15, 30 and 60-minute windows (NREL, 2015), and its 2017 survey confirms the charge follows the highest average use inside that interval (NREL, 2017). A 15-minute peak compared with a 30-minute billed average overstates what the tariff will record. Keep the raw data and record the conversion in the RFQ.

How should duplicate daylight-saving intervals be handled?

Log them, do not merge or delete them. Elexon notes a long clock-change day carries 50 half-hour settlement periods instead of 48 (Elexon, 2025), and UTC timestamps make the duplicate visible. No Tier 1-3 source was found that prescribes a numeric rule for missing-interval filling in BESS sizing, so record the gap and ask the supplier to state its assumption.

What to send a supplier

Attach the worksheet, not just a target kWh. Send the interval CSV with UTC timestamps, the cap and ratchet clause, the running cumulative column, the derate table and the critical-load list. IEC 62933-3-1:2025 calls for preliminary and final sizing with a duty cycle at the point of connection, so ask the supplier to restate your inputs at that boundary. The 12-field RFQ checklist gives the field list; the quote comparison worksheet normalises quotes built on different efficiency assumptions.

This is general engineering information, not a substitute for a site survey. Final configuration must be confirmed with the supplier and the authority having jurisdiction.

Sources

• U.S. DOE, Understanding Your Utility Bill: A Guide for Business in Ohio (posted 2024)
• IEC 62933-2-1:2017
• IEC 62933-3-1:2025
• NREL 2024 Annual Technology Baseline, Commercial Battery Storage (hosted at NLR)
• PNNL-33283, 2022 Grid Energy Storage Technology Cost and Performance Assessment
• Sandia / DOE / EPRI, SAND2015-1002 (2015), Table 1
• LBNL, Granderson et al., Applied Energy 173 (2016)
• DOE FEMP M&V Guidelines v5.0 (2024)
• NFPA 110 standby classes, reproduced 2023

NYSERDA Report 20-17 (NYSERDA / Clarkson University, 2020), NREL/TP-6A20-64850 (NREL, 2015), NREL/FS-6A20-68963 (NREL, 2017) and Elexon (2025) are cited by institution and year in the text without a live link. NEC Articles 700, 701 and 702 are cited by article number. The 0.80 usable fraction and 0.70 end-of-life state of health are worksheet assumptions stated in Step 4, not measurements from a vendor.

Reproduce the worksheet, keep power and energy as separate results, and make every efficiency assumption visible before comparing quotes. Send the interval file with the enquiry so the supplier restates the same rows.