Voltage headroom

What it is (precise)

Four DC-side numbers compete to be the bound, and the smallest positive difference is the headroom that matters. The BMS enforces per-cell limits that refer up to a string ceiling and floor; the PCS publishes a minimum DC voltage for full rated output and a lower minimum at which it stays connected but derated; and the system carries a maximum DC voltage that no operating state may cross.

Comparing a battery's nominal voltage against a converter's window tells you almost nothing about any of these, because nominal is a label, not an operating point — the nominal-voltage entry makes that case in general, and headroom is where it bites hardest.

There are always two headrooms in play, and current direction decides which is under pressure. Loaded terminal voltage is the open-circuit voltage set by state of charge, plus I × R on charge and minus I × R on discharge. Charge current therefore spends top headroom while discharge current spends bottom headroom, and state of charge positions the baseline both are measured from. A plant's headroom is a matrix of corners: hot and full while charging at rated power, cold and low and aged while discharging at rated power, and the reactive-support cases in between.

State the terminal or the number is unusable. The converter's DC bound applies at the PCS DC terminals, not at the rack, so the I × R drop in the DC run between them is spent before the bound is even tested — the voltage-drop entry covers the mechanism. The same discipline applies on the AC side: a grid-code limit binds at the point of interconnection, an equipment rating binds at the equipment, and the transformer and cable impedance between them means those points read different voltages at the same instant.

Where the DC bounds actually sit, and what moves them

The converter's floor is not a marketing choice — it is set by the AC side it has to synthesize. The DC bus must stay above the peak of the line-to-line voltage the inverter is producing, which is √2 × V(L-L), or the modulation runs out and output falls. Power Electronics publishes the resulting full-power DC minimum per AC variant across its Freemaq PCS families: 849 V for a 600 V unit, 934 V at 660 V, 976 V at 690 V, 1,019 V at 720 V.

Compare those against √2 × V(L-L) — 848.5, 933.4, 975.8, 1018.2 — and the floor is the AC peak to within a volt in every case. So the headroom the DC design is working with on a 690 V machine is the span from 976 V up to the ceiling, and the published full-power top is 1500 V on some variants and 1,310 V on others.

The consequence most DC-side checks miss is written in the datasheet conditions: the minimum DC voltage varies proportionally with AC voltage on grid overvoltage. A grid running 5% high raises a 690 V unit's full-power floor from 976 V to roughly 1,025 V. That is about 49 V of DC headroom consumed by an event the battery never experienced, at a moment when nothing on the DC side changed. A headroom figure computed against the nominal AC voltage is therefore an optimistic figure, and the amount of optimism is exactly the grid's overvoltage excursion.

Reactive duty spends the same currency. Power Electronics publishes minimum DC voltage against power factor: at 690 Vac and 50 Hz that curve reads 975 V — the same floor the per-variant figure above rounds to 976 — holding flat from the lagging end across to about cos φ 0.84 at 25 °C, then climbs to roughly 992 V at cos φ 0.4, with the 50 °C curve staying flat to about cos φ 0.78 and reaching 987 V at 0.4.

Other vendors publish a plain pair — SMA's SCS 4400 UP-S-US states an operating DC range of 962 to 1500 V — but the structure is identical. The MVA headroom entry owns the apparent-power side of reactive obligations; the point here is that a reactive instruction is also a DC-voltage demand, so the two headrooms are spent together.

What it costs when it runs out

Above the knee, headroom buys nothing at all; below it, headroom is power. Huawei's published DC-voltage derating for the LUNA2000-213KTL-H0 at 800 Vac, 40 °C and unity power factor makes the shape concrete: the full 213 kW is available from 1,100 V upward and stays flat all the way to 1500 V, but at 1,000 V the unit makes 170 kW, at 900 V it makes 100 kW, and at 800 V it makes 35 kW. The first 100 V below the knee costs a fifth of rated output. The next 100 V costs a third of it.

The mechanism is the converter's DC current limit. Below the knee the machine is already at its current ceiling, so available power tracks bus voltage almost linearly — the derating entry owns condition-based capability loss in general, and this is its voltage axis. What the shape means practically is that a compliance check reporting "inside the window" is close to useless, because it gives the same answer for 5 V of margin and for 300 V. Report the number of volts, at the corner, or the check has not been done.

The ceiling fails in a different currency. Charging into the top does not trip anything in normal operation; it forces a taper, with the BMS cutting current to hold the highest cell under its limit, so lost top headroom shows up as charge duration rather than as missing megawatts. The ceiling itself is also the least negotiable bound on the plant: IEC 62477-1 covers converter systems up to 1000 V AC or 1500 V DC, which is why 1500 V is the top of the low-voltage-classified band a grid-scale PCS is normally certified in. No firmware setting moves it.

The quantity that shrinks with resistance

Every term that grows in I × R takes volts off the bottom. Cell internal resistance rises over life, contact and joint resistance drifts with thermal cycling, and cold multiplies both. The site's canonical numbers show how little slack there is to start with: a 1500 VDC-class string of roughly 360 to 416 LFP cells sits near 1,150 to 1,330 V nominal with a protected floor around 900 to 1,040 V, and that floor is a cold, loaded number rather than a sum of open-circuit voltages.

Set the converter's 976 V full-power floor beside it and the overlap is obvious — at the cold, low-state-of-charge, full-current corner the string can be entirely legal by its own protection limits while sitting below the voltage the converter needs for rated output.

This is why the headroom check belongs at end of life rather than at delivery. Same dispatch, same ambient, more millivolts of sag per amp: bottom headroom erodes year over year at a rate the degradation model usually does not report, because degradation models report capacity. The efficiency-fade entry treats resistance growth as a loss story; here it is a volts story, and it is the one that ends a discharge early. Cell imbalance takes a further slice, since the BMS enforces per-cell bounds and the string stops when its worst cell arrives, not when the average does.

The DC cable run between rack and converter is the last slice, and it is asymmetric. On discharge the drop subtracts, so the PCS terminals read lower than the rack and bottom headroom is smaller where the bound is tested. On charge the same drop adds, so the PCS terminals read higher and it is top headroom that shrinks. One conductor run, two different penalties, both worst at full current — which is precisely when the bound is being approached.

AC-side headroom, and how it shows up in specs and contracts

AC headroom is measured against two unlike bounds. The first is the grid-code band, and it is jurisdictional. ERCOT's Nodal Protocols section 3.15 defines the lagging reactive obligation over voltage set points of 0.95 to 1.04 pu and the leading obligation over 1.0 to 1.05 pu.

In Great Britain, ERE G99 Issue 2 clause 13.5.5 defines the capability for connection points at or below 33 kV across a 0.95 to 1.05 pu envelope in which full lagging capability exists only at or below 1.00 pu and full leading capability only at or above it. Read either as headroom and they say the same thing: the obligation is written only inside a band, and where the bus sits inside that band decides how much of the obligation you can still meet.

The second bound is insulation and thermal rating, and it does not care about compliance bands. IEC 60038 pairs every nominal system voltage with a highest voltage for equipment — 36 kV for a 33 kV system, about 9% above nominal — and it is that pairing, not the nominal label on the drawing, that the collection system's headroom is measured against.

Transformer off-load taps are the coarse adjustment on this axis, and each tap position trades top headroom for bottom. A tap chosen at commissioning to prop up a weak feeder becomes the wrong choice once that feeder is reinforced and the bus runs high.

In specifications, ask for the DC bound with its conditions attached rather than as one number: the full-power minimum at rated AC voltage, at the grid overvoltage the interconnection permits, and at the power factor the code requires. In studies, take POI voltage from the load flow across the operating range instead of at nominal, then refer it back through the transformer to the converter bus.

Contractually, a performance guarantee at rated power should name the DC corner at which it is demonstrated, and the capacity-test procedure should record grid voltage on the test day — two plants with identical nameplates and different tap positions do not test the same.