Derating

What it is (precise)

A power converter is, at bottom, a current-limited machine. Its semiconductors, magnetics and cabling set a maximum continuous current, and on the three-phase AC side apparent power is S = √3 × V(L-L) × I(line) — so at a fixed current limit, the kVA rating scales directly with AC voltage.

One shipping modular PCS delivers 374 kVA per module at 480 V and 537 kVA at 690 V: same hardware, same current limit, and the power ratio tracks the voltage ratio almost exactly. That is why a PCS power figure means nothing without its AC voltage (0.69 kV is the common grid-scale default), and why a vendor headline quoted on a higher-voltage variant does not transfer to a design built around a lower one.

Temperature derating is the thermal face of the same limit. Rated current holds only while the cooling system can remove the losses it generates, so every rating is pinned to an ambient temperature and falls above it — and the shape is a product fact, not a rule of thumb.

Within a single vendor's range, one model holds full rating to 40 C, another begins derating at 30 C, and a third is rated at 50 C outright; that last family delivers about 7.7% above its rated figure whenever the air is below 40 C, then falls from full output at 50 C to zero at 60 C. The modular unit above derates 2% per degree from 45 to 55 C and 6% per degree beyond. Derating is a capability surface with a cliff in it, and where each vendor plants the reference point is a design choice they make differently.

The other levers: altitude, DC voltage, power factor, age

Altitude bites twice — thin air cools less and insulates less — and the two effects produce opposite product behaviour. One DC/DC converter family runs flat at 100% to 2,000 m and then loses capability linearly to 66.4% at 4,000 m, about 1.7% per 100 m; an air-cooled string PCS holds full output to 5,000 m and instead derates its maximum DC voltage above 4,700 m, because for that machine the binding constraint at height is insulation clearance, not heat.

Two products, two entirely different altitude stories. The product curve is the truth, and any generic percent-per-metre rule substitutes for it badly.

The DC side interacts with everything else. One central-inverter family shows no DC-voltage derate at all when ambient is 40 C or below, yet at 50 C holds full output only to a 1,400 V knee and gives 97% at 1,500 V — the derate exists only where two stresses coincide. That interaction is why single-axis rules fail and the matrix has to be read whole. Power factor spends the same envelope from another direction: a reactive obligation consumes the apparent-power circle through S² = P² + Q², so at 0.95 power factor real power is at most 95% of the MVA rating before any thermal derate is counted.

Age is the derate no dispatch decision can reverse. Capacity fade takes usable energy down toward the 65-70% of beginning-of-life conventionally taken as End of Life, and rising internal resistance trims peak power and efficiency along the way. Oversizing and Augmentation exist to offset it, and the Capacity warranty is written against a defined operating envelope — run the plant outside the warranted temperature or cycling conditions and the guaranteed fade curve stops protecting you.

The battery derates itself, in real time

Everything above is printed as curves; the DC block enforces its derates in firmware instead. Charge power tapers near the top of the SOC window and discharge power near the bottom, so a rating that holds across the middle band does not hold at the ends. Temperature narrows the window further: below roughly 0 C, charging must be derated or blocked outright, because lithium plates onto the graphite anode instead of intercalating — permanent capacity loss and a potential internal-short seed. The BMS applies these limits second by second.

Plant-level available power is therefore the minimum of two schedules: the PCS matrix row for the day's conditions and the battery's state right now. A 100 MW plant at high SOC on a cold morning can accept far less charge than its interconnection limit allows, and an EMS that dispatches against nameplate rather than against the live limit will simply be refused by the layers below it. Local protection wins that argument every time.

How it shows up in specs, studies and contracts

On a datasheet, look past the headline to the matrix: rated kVA tabulated against grid voltage, ambient and altitude, with derating curves above the reference point, and ask three questions — is the figure continuous or a short-time overload, which voltage variant is it quoted on, and at what temperature does it hold.

Transformers carry the same idea as cooling stages: one nameplate can read 75/100/125 MVA for the same unit at ONAN/ONAF/OFAF. The binding number for the project is the continuous rating in the row that matches the site's hottest credible hour, at its altitude, at its grid voltage.

Studies and contracts are written at the derated point whether you modelled it or not. The interconnection study holds the plant to its contracted MW and reactive envelope at the POI; the capacity test demonstrates it at a stated ambient and power factor; an energy model that dispatches the brochure number all year overstates August revenue at any hot or high site. Vendor comparison has the same discipline: a unit rated at 50 C looks smaller on paper than one rated at 40 C and may be the larger machine at your site — comparisons mean nothing until every number is moved to the same row.

Common pitfalls

The first trap is reading the headline as a guarantee. The datasheet's front-page figure is the best row of the matrix — reference ambient, reference voltage, unity power factor — and the firmware will fold output back the moment the heatsink says so.

The second is assuming derates are linear and independent: the 50 C-rated family above loses its entire output across a ten-degree band, and its DC-voltage knee exists only at high ambient. Extrapolating a gentle 2%-per-degree slope past the cliff, or adding single-axis derates that actually interact, both produce numbers the hardware will never deliver.

The third trap is forgetting the flip side. Below the reference point many fleets carry genuine headroom — 7.7% on the family rated at 50 C, more on others — which is real capability for a cool-climate site or a winter peak. But headroom the capacity test never demonstrated and the contract never named is not bankable capability; it is a favourable rounding error. Count in contracts only what is tested at stated conditions, and treat everything above the reference row as operational margin, not revenue.