P-Q capability test

What is actually demonstrated

The test walks a matrix of commanded operating points and records what the plant did at each. The shape of that matrix is the first thing worth arguing about, because a default vendor procedure and a grid-code obligation rarely want the same points.

A defensible matrix reaches all four quadrant corners at rated real power — discharge with reactive injection, charge with injection, charge with absorption, discharge with absorption, in the generator-convention numbering the four-quadrant-operation entry sets out — plus both reactive extremes at zero real power, plus both reactive extremes at intermediate real-power levels.

The intermediate rows are the ones most often dropped and the ones a rectangular P-Q requirement most needs, because a converter's circular boundary offers more reactive headroom at partial output than at full output while a rectangular code envelope asks for the same reactive figure across most of the real-power range. Testing only the full-power corners proves the tightest point on the circle and says nothing about whether the plant controller can actually hold the requirement in the middle of the range, where a partly derated fleet or a block out for maintenance changes the answer.

A corner reached is not a corner held. Converter, transformer and enclosure temperatures rise over minutes, so a point touched for a few seconds proves the current loop hit its setpoint and nothing more; whether the plant can sustain that point long enough for thermal limits to bind is a separate question, and the dwell time and averaging window that settle it belong in the test procedure rather than in the operator's judgement on the day. State of charge moves under the test at the same time.

A four-hour plant discharging at rated power traverses its whole usable window in four hours — about one percent of the window every 2.4 minutes — so the full-power corners have a short budget before state of charge, not the converter, becomes the binding limit. Sequencing matters: the charging corners need room at the top of the window and the discharging corners need room at the bottom, and a matrix written without that constraint in mind ends up half-tested with a battery at the wrong end of its range.

The scope boundary is worth writing down as well. A P-Q capability test is a steady-state demonstration of where the operating point can sit and how long it can stay there. It is not a ride-through test and does not establish what the converter does at depressed voltage, where the current limit binds and firmware decides how the available current is split between real and reactive components.

Transition behaviour sits in between: most procedures record the ramp and settling between commanded points, which is useful evidence about the controller, but reaching a setpoint quickly and holding an envelope corner for its full dwell are different claims and should be reported as different results.

Which boundary — the whole game

The vendor draws its curve at the inverter terminals under stated conditions because the terminals are the only boundary it controls. The obligation is written somewhere else.

In the United States, FERC Order 827 removed the reactive-power exemption for non-synchronous generation and set the requirement for newly interconnecting non-synchronous resources as a ±0.95 power-factor range at the high side of the generator substation; IEEE 1547-2018 for distribution-connected resources and IEEE 2800-2022 for transmission-connected inverter-based resources each define a reference point of applicability at which their requirements are evaluated. Grid codes in other jurisdictions name their own point, and neither its location nor its name travels between markets.

Between the two boundaries sit the unit transformers, the collection cables and the main power transformer, and they take reactive power in proportion to loading squared times their own reactance — one nameplate in hand carries a reactive component of impedance voltage running 8.48 to 9.14 percent, which is roughly 340 to 370 kvar out of a 4 MVA-class block at full load, and the sign of that absorption does not reverse when the plant swings from discharge to charge.

A separate 4,140 kVA / 690 V medium-voltage skid transformer publishes 4.6 kW of no-load loss and 43 kW of load loss at rated power, about 1.15 percent of its own rating with both counted, which comes off the real-power axis. The poi-capability-envelope entry covers that translation in full; the point here is that a test measured at the wrong bus produces a number nobody can use.

What the test uses to measure decides how much of the result is real. A revenue-class metering point is specified to an accuracy class, and the classes come from different documents that are not interchangeable: 0.2S and 0.5S are IEC 62053-22 classes for active energy, reactive energy has its own IEC document in 62053-24, and the comparable ANSI classes 0.2 and 0.5 come from ANSI C12.20. Reactive measurement is also intrinsically harder than real measurement near unity power factor, and the reason is geometry.

With P = S cos φ and Q = S sin φ, a small phase-angle error in the instrument transformers barely moves P when φ is near zero, while it lands almost in full on Q: a 0.1° error is 1.75 mrad, which puts on the order of 0.17 percent of apparent power into the reactive reading and leaves the real reading essentially untouched. The acceptance band on the reactive result should reflect that asymmetry rather than borrowing the band written for the capacity test.

Who witnesses is a contract question, not a default. Owners send their own engineer, lenders send an independent engineer, and the utility attends where its agreement gives it the right — but the right has to be located.

Under the FERC pro forma Large Generator Interconnection Agreement the Article 6.3 right to observe testing is scoped to each party's Interconnection Facilities, so in that document the transmission provider has no pro forma right to witness testing of the battery plant itself; a witness right over the plant's own capability test comes from elsewhere in the agreement or from the utility's specification. Settle it before the test window is booked, because a result nobody with standing observed tends to get retested.

The conditions you get versus the conditions you owe

The grid does not take instructions from a test procedure. Point-of-interconnection voltage on the day will be close to whatever the network is holding, usually near nominal, and no amount of scheduling produces a 0.90 pu bus on demand — yet the requirement is generally owed across a voltage band, and a current-limited converter delivers less apparent power as voltage falls.

One string-PCS characteristic set publishes the same converter as a 236 kVA circle at and above nominal, 225 kVA at 0.95 Un and 213 kVA at 0.90 Un. Measuring at 1.0 pu and declaring the band proven is a claim the measurement does not support.

Ambient temperature moves the answer by more than most acceptance bands allow. One 690 V 5 MVA-class block reaches about 5.38 MW at zero vars and the same 5.38 Mvar at zero real power at 40 °C and 1,000 m, and roughly 4.84 MW and 4.84 Mvar at 50 °C — the same hardware, about ten percent apart.

Another converter family on a 1,200 Vdc bus publishes 3,960 kVA at 25 °C, 3,610 kVA at 40 °C and 3,365 kVA at 50 °C, a fifteen percent spread across the range a site might see across a year. A test run on a mild morning and compared to an obligation owed on the site's worst afternoon is comparing two different plants unless something corrects between them.

The DC side carries its own condition, and it is easy to leave out of the test record. One PCS datasheet states ±2,305 kvar at zero power factor with the figure explicitly annotated as valid at 40 °C design temperature, 1,159 Vdc minimum battery voltage, 0.9 pu AC voltage and 0 to 1,000 m altitude. On a 1500 VDC string with a nominal band of roughly 1,150 to 1,330 V, that 1,159 Vdc floor sits at the very bottom of the nominal band and far above the protected floor of about 900 to 1,040 V, which is itself a cold, loaded number rather than a sum of open-circuit cell voltages.

Under load at low state of charge the bus reaches the vendor's stated minimum long before any protection acts, so the reactive figure is conditioned on where in the window the test was run. Record the state of charge at every point in the matrix, not just at the start.

Two further conditions belong in the same record: how many converter blocks were actually in service, since an isolated container shrinks the envelope directly, and what voltage excursion the system operator permitted, because on a weak connection pushing full reactive injection moves the point-of-interconnection voltage enough that the operator may cap the test before the converter reaches its own limit.

Normalizing the result before it meets the guarantee

There are two honest ways to compare a measurement taken at the day's conditions with a guarantee written at reference conditions. Correct the measurement to the reference conditions using the vendor's published derate curves, or redraw the guaranteed envelope at the day's conditions and compare there. Either is defensible.

What is not defensible is choosing between them after the numbers are in, and that is the usual failure: the correction method, its data source and the reference conditions it targets should be named in the test procedure that both parties sign, alongside the boundary, the dwell, the averaging window and the acceptance band.

The corrections above are large enough to decide the outcome on their own — around ten percent between 40 and 50 °C on the 690 V frame, about fifteen percent across 25 to 50 °C on the other family, roughly ten percent of apparent power between 1.0 and 0.90 Un on the string-PCS set — so the method is not a formality attached to a settled result.

Some corners cannot be reached physically, and for those the evidence is a model rather than a measurement. The standard practice is to validate the plant model against the points that were reached and then use the validated model to carry the corners the network would not allow, which is why the capability test and the model-validation campaign are usually run from the same data set.

In North America the model-verification obligations are separate NERC standards from the capability ones — MOD-026 covers verification of models and data for generator excitation control or plant volt/var control functions and MOD-027 covers the turbine/governor and active-power/frequency control side — so a plant that passes its witnessed corners can still have an unfinished model deliverable.

Note also that verification is not a one-off in that jurisdiction: NERC MOD-025-2 requires registered Generator Owners on the bulk power system to verify real and reactive power capability periodically, at least once every five calendar years, and report the data. A commissioning protocol that cannot be repeated cheaply becomes an operating cost the fifth year it is needed, so write it to be re-run.

The last piece of normalization is arithmetic rather than physics: what is being compared to what. A guarantee expressed as a power factor at the meter, an envelope expressed in Mvar at the terminals, and a fleet capability expressed in per unit of rated apparent power are three different quantities, and per unit means nothing without its base.

Fix the base — apparent power of what, defined where, at which reference voltage and ambient — before the comparison, because the same measured megavars can clear or miss the requirement depending on which of those three forms the guarantee was written in.

Common pitfalls

The most common substitution is a certificate for a demonstration. Equipment conformance testing under IEEE 1547.1 and UL 1741 SB is type testing done on a converter in a laboratory; it establishes what a model of inverter does under defined conditions and says nothing about the assembled plant's capability measured at its own meter with its own transformers, cables and auxiliary load in the path.

A related substitution runs the other way at the reporting stage: taking the reactive figure from plant-controller or SCADA telemetry instead of the metering point named in the agreement. Those values come from a different location, a different accuracy class and a different filtering and update regime, and where they are read for convenience during the test they should be reconciled against the meter, not reported in its place.

The charging quadrants are the ones dropped most often, usually for a practical reason — the state of charge was wrong, the market window closed, the operator would not accept the load — and the omission is not benign, because transformer reactive absorption does not change sign when real power reverses, so the charge-side corners are genuinely different tests rather than mirror images of the discharge-side ones.

The zero-real-power corner is skipped for a similar reason and matters for the same one: standby reactive support is a real obligation in several markets, and it is the corner most likely to be limited by a firmware enable state rather than by hardware.

Two reporting traps close the list. The first is declaring a voltage band proven from a single voltage, or an ambient range proven from a single morning, without saying so in the report — the honest version records what was measured, what was corrected, by which method, and which parts of the envelope rest on the model rather than on the meter. The second is letting the test report stand as the whole record.

The plant's baseline capability, the conditions it was measured under and the correction method are the reference every later dispute and every periodic re-verification will be argued against, so they belong in the document set alongside the capacity and round-trip-efficiency baselines from the same commissioning campaign, not in an engineer's laptop.