Full-power deliverability

What it is (precise)

Deliverable power at a given moment is the minimum of a chain, not a property of any one device. The BMS publishes a charge and discharge power limit derived from present cell voltages, temperatures and estimated resistance. The DC bus sits somewhere in its swing, and where that voltage falls below the converter's full-power floor the PCS becomes a current-limited machine rather than a power-limited one. The converter's own AC rating is itself conditional on ambient temperature, altitude, AC terminal voltage and the power factor being asked of it.

Downstream, the medium-voltage transformer has a cooling-stage rating, the collection system has an ampacity, and the interconnection agreement has an export cap. Full-power deliverability is the statement that the smallest of those numbers still equals rated MW at every point of a named envelope — and the discipline that makes it meaningful is evaluating all of them at the same instant, under the same conditions, rather than each on its own datasheet at its own reference point.

Two qualifiers turn the phrase into something testable. The first is the boundary: rated MW at the PCS AC terminals and rated MW at the point of interconnection are different quantities separated by transformer, cable and auxiliary load, so the obligation has to name the bus it is measured at — the measurement-boundary entry covers that discipline. The second is the date.

The envelope shrinks over a project's life as resistance grows, so a full-power claim carries a year the way a retention figure does; a plant that holds rated MW to the floor of the window in year one may hold it only to a third of the way down by year twelve. Note also what this term does not own: megawatt-hours. Deliverable energy is the MWh axis, and the two fail independently — a fleet can sweep its full window at a gentle rate while being unable to hold nameplate MW through the last stretch of it.

The current-limited corner

Below its full-power DC voltage floor a converter is limited by DC current, so available power falls roughly in proportion to bus voltage — the VDC window entry works that mechanism through. What matters for deliverability is where that floor actually sits, because it is not one number per product.

It tracks the AC output voltage class: across one central-PCS family in this glossary's reference set of datasheets, the minimum DC voltage for full power runs about 870 V on a 615 Vac frame and about 975 V on the 690 Vac frame of the same family at 25 C, and higher AC classes push published knees upward into the roughly 1,050 to 1,300 V region that vendor windows commonly quote. Pick the AC voltage and you have picked the DC floor.

Reactive power moves the floor again, in the direction that hurts. On that same 690 Vac family the minimum DC voltage for full power rises from about 975 V at unity to about 992 V as the demand moves toward a leading power factor of 0.4 (25 C). A 4,200 kVA central unit from another product line, on a 660 Vac frame, shows a larger swing on the same axis: about 933 V at unity, about 1,024 V toward leading 0.4, measured at 40 C ambient.

Most of the gap between the two floors is the AC class each is synthesizing; the swing is the part that belongs to the reactive demand. So a reactive obligation charges the plant twice. It consumes apparent-power headroom on the AC side, which is the MVA-headroom question, and it simultaneously raises the DC voltage the battery must still be holding for the converter to make rated MW at all. The second charge is the one that goes unmodelled, because it lives on a different vendor's curve from the first.

Now put those floors next to the battery. A 1500 VDC-class LFP string sits nominally around 1,150 to 1,330 V, and its protected floor under a cold, loaded discharge is roughly 900 to 1,040 V — the same band the converter's full-power floor occupies. That overlap is the whole subject of this entry. Whether the last portion of a January discharge happens at rated MW or at a taper depends on which of two independently specified numbers, from two suppliers, wins by a few tens of volts at the moment of deepest sag.

The hot end of the envelope has its own version: one PCS family rates 1.00 pu at 50 C and 1.077 pu at or below 40 C, then falls steeply above roughly 52 C toward zero at 60 C, while a modular unit publishes 537 kVA per module at 690 VAC and 45 C, derating 2% per degree to 55 C and 6% per degree beyond it. Neither corner is an anomaly; both are printed, and both are outside the row most reviews read.

Why a capacity test can pass while this fails

A capacity test measures megawatt-hours swept across a defined window at a contract C-rate, temperature-corrected, at a named meter. It is a good instrument for the quantity it measures and a poor proxy for this one, for four reasons that compound. The test is normally scheduled in mild weather with the thermal system given time to precondition the containers, so the cold corner never appears.

The contract C-rate is often the energy-duration rate — 0.25C for a four-hour system — while the market duty that matters may call for several times that current in a frequency event. Most protocols count all the energy delivered, including whatever came out below the full-power floor after the converter began tapering, so a discharge that ends in a derate can still book a full pass. And the test is run at or near unity power factor, while the interconnection agreement obliges reactive support that raises the DC floor the discharge has to stay above.

The consequence is that a plant can hold a clean run of annual capacity certificates and still be unable to do the thing its capacity contract or frequency-response registration was written around. The failure surfaces as an event, not as a trend: a cold morning where output settles a few percent under instruction in the tail of the discharge, or a reactive dispatch that clips real power at a state of charge nobody had flagged.

Because the settlement record shows an energy shortfall, the diagnosis usually starts on the energy axis and the mechanism sits one axis over. The commissioning and capacity-testing article covers what the standard test does establish; the useful addition is a short set of corner points beside it.

The reverse case is worth naming because it exonerates hardware that looks faulty. A plant that holds rated MW perfectly can still fail a capacity test if the SOC window in force was narrowed by a firmware change or a warranty condition. Power and energy have separate failure modes, separate evidence and, frequently, separate responsible parties — which is why the test evidence for each should be separately specified rather than inferred from the other.

Writing it into a contract

A full-power obligation is only enforceable if the envelope is written down. That means five statements, not one number: the state-of-charge band over which rated MW must hold, expressed against the window actually in force rather than against nameplate; the ambient and cell temperature range; the AC voltage band, taken from the connection agreement for the jurisdiction in question rather than from a generic figure; the power factor range over which the MW figure survives; and the contract year the obligation applies in.

A megawatt guarantee without those five carries whatever conditions the supplier's datasheet happened to assume, and those are the mild ones.

The test regime follows from the envelope. Two corner points added to the annual capacity test do most of the work: a rated-MW hold at the bottom of the guaranteed SOC band on the coldest day the test window offers, and a rated-MW hold with the contracted reactive output applied simultaneously. Both should be measured at the contractual boundary and both should record the DC bus voltage and the BMS available-power limit alongside the meter reading, because without those two traces a shortfall cannot be attributed to a link in the chain.

That attribution question is the commercial heart of the clause: the battery supplier owns the DC side, the PCS supplier owns the converter, the EPC owns the collection system, and a corner failure is a chain failure that no single warranty covers unless one party has been made responsible for the intersection.

Two drafting details reward attention. The first is how availability treats a derate — an hour in which the BMS caps available power below nameplate is not an outage, and a definition of availability written around breaker position will score it as fully available, which removes the only routine incentive anyone has to watch the number. Define availability against capability, or accept that the metric will not see this.

The second is telemetry access: the BMS available-power limits and the per-string resistance history are the evidence base for every later argument, and they are far easier to obtain as a contracted data feed at procurement than as a request during a dispute. Power fade covers why explicit power warranties remain rare; the practical answer while that stays true is to buy the measurement even where the guarantee is unavailable.

Common pitfalls

The dominant error is checking each link at its own reference condition and assuming the intersection survives. The PCS sheet is read at 25 C, unity power factor and nominal DC; the battery sheet is read at 25 C and mid-window; the transformer is read at its top cooling stage. Every one of them passes, and the plant still cannot make rated MW at a corner where three of those assumptions are false at once.

The corner is not exotic — cold, low state of charge and a reactive instruction co-occur routinely on winter evenings — it is simply the one condition set no individual document describes. The related habit is quoting the AC nameplate as the plant's MW: the converter sets a ceiling, and the ceiling is only reachable while the DC side supplies the current behind it.

Three narrower traps round the list out. Reading a full-power knee as a single number ignores its dependence on AC voltage class and power factor, which between them can move it by tens of volts into the band the string floor occupies. Running the corner check only at beginning of life proves the least interesting year, since resistance growth pushes the DC floor and the string floor toward each other over time.

And aggregating block ratings to a plant figure hides limits that bind above the block — transformer cooling stages, collection ampacity, the export cap itself — so a plant can be short of rated MW at the meter while every container reports full output. The uncomfortable feature of all of these is that nothing is broken when they bite: every device sits inside its own published limits, and the intersection of those limits is simply smaller than the number on the contract.