Start at the tariff, not at the battery
A demand charge is billed against the highest averaged interval in the billing period, and the averaging window is usually fifteen or thirty minutes. That window decides which events are worth shaving at all: a two-minute inrush that barely moves a thirty-minute average may cost nothing, while a forty-minute plateau at a modest level sets the bill for the month.
Get interval data at the resolution the meter bills on. Data averaged more coarsely hides the peaks; data averaged more finely exaggerates them. Then establish whether the charge is set monthly, seasonally, or ratcheted from an earlier month, because a ratchet changes what a single missed event costs.
Two numbers, computed separately
PCS power in kW decides how deep a peak can be cut. Usable battery energy in kWh decides how long the cut can be held. Sizing one and assuming the other is the most common way a peak-shaving system ends up unable to do the job it was bought for.
The kW figure is the worst observed interval minus the target threshold. The kWh figure is the area above the threshold across the longest single event, which then has to be grossed up before it becomes a nameplate number.
- Divide by the usable state-of-charge window, since a pack is not cycled from full to empty
- Divide by round-trip efficiency, so the losses are covered
- Add the end-of-life capacity you are contracting to, not the day-one capacity
- Add any energy held in reserve for backup, which is not available for shaving
A worked example
Take a site billed on thirty-minute demand that peaks at 800 kW and wants to hold 600 kW. The PCS has to supply 200 kW. If the worst event runs forty-five minutes at that depth, the energy above the threshold is 150 kWh.
Grossed up for an eighty percent usable window and ninety percent round-trip efficiency, that is roughly 210 kWh of nameplate capacity before degradation allowance. Check the rate that implies: 200 kW out of 210 kWh is close to a 1C discharge, which is a real constraint on cell selection and on cooling. If the numbers land above what the chosen cells will sustain, the battery has to grow even though the energy calculation said it did not need to.
These figures are illustrative. The method transfers; the numbers must come from the site.
The recharge is the second peak
A battery that discharged during the afternoon peak has to recharge, and recharging inside the same billing period at full power can set a new demand record higher than the one that was just avoided. This is the failure mode that turns a correctly sized system into a project that did not save anything.
Constrain charge power against the measured site load rather than against the clock, hold the recharge until the site has headroom, and confirm that the control still completes the recharge before the next expected event.
When shaving shares the battery with something else
Time-of-use arbitrage wants the pack empty before the cheap window. Backup wants it full. Solar self-consumption wants it empty at midday. Each objective claims the same energy at a different time, so the controller needs an explicit priority order rather than three rules running at once.
Write down which objective yields when they conflict, and what state of charge is reserved and never spent on the lower-priority one.
What the duty cycle does to cooling
A peak-shaving duty is short, deep and repeated, which produces a different thermal profile from a daily energy-shifting cycle at the same throughput. Size the cooling against the worst event and the recovery period between events, not against the daily average.
Questions this raises most often
Separately in kW and kWh. kW is the worst billed interval minus the target threshold; kWh is the energy above the threshold in the longest event, grossed up for the usable state-of-charge window, round-trip efficiency, degradation and any backup reserve.
Demand is charged on an averaged interval, usually fifteen or thirty minutes. A short spike may barely move that average, while a long plateau at a lower level can set the bill.
Yes, and it regularly does. Recharging at full power inside the same billing period can set a demand record above the one that was avoided, so charge power has to be constrained against measured site load.
The objectives claim the same energy at different times. Define an explicit priority order and a reserved state of charge that the lower-priority function is never allowed to spend.
Take it to a part number
Ratings vary between codes that read almost the same, so settle the decision here and then confirm the figures on the product page for the exact code.