Energy capacity test

What the test actually integrates

The measured quantity is energy — the time integral of power at the named meter between the two endpoints the procedure defines — not capacity in the amp-hour sense and not a property of the cells. A four-hour plant discharged at rated power crosses its whole usable window in about four hours, roughly one percent of the window every 2.4 minutes, so the test is short and every minute of it is conditioned by something.

What comes out is one point on the deliverable-energy surface, taken at one temperature, one rate, one window and one date. Deliverable energy is a state of the plant; a capacity test is how you read it.

Two things can end the discharge, and the procedure has to say which governs. Cells in series carry the same current, so the string stops when the first cell reaches its lower cutoff — a physical end condition, and the one the plant will actually hit. A displayed state-of-charge floor or a DC undervoltage threshold at the PCS are different end conditions layered on top of it. Where the contract names a minimum operating SOC, establish what enforces it, because displayed SOC is bookkeeping over voltage limits and the two do not have to agree.

Five definitions, five different answers

Start with the starting condition. A charge terminated at the BMS upper cutoff and a charge terminated at a displayed 100 percent are not the same stock of energy, and charge power tapers near the top of the window, so the last few percent take disproportionate time — a procedure that caps charge duration quietly caps the result. Then the rest. Terminal voltage relaxes after a charge, and a discharge begun immediately starts from a sagging bus and reaches the lower cutoff sooner. The rest period is not padding; it is what makes the measurement repeatable.

Rate does the largest damage. Ohmic loss in cells, busbars and cables grows with the square of current, and the deeper sag brings the weakest rack to its floor earlier, so the same window yields fewer megawatt-hours at rated power than at a gentle test rate. A test at the contract duration and a test run slowly are two measurements of one plant, and only one of them is the duty the offtaker will dispatch. Temperature moves it from both ends: a cold fleet sags harder and terminates high, a hot one raises the HVAC draw the POI meter sees.

The plant's own state is the definition most often left out of the record. How many blocks were in service, whether any container was isolated, what the SOC and temperature history had been in the preceding days, and whether the auxiliary supply was the permanent feed or a temporary one. A result recorded without those is a number that cannot be reproduced, and reproducibility is most of what the test is for.

The boundary decides what has already been subtracted

A capacity test is a single integral rather than a ratio, so it escapes the numerator-and-denominator problem an efficiency test has. It has a simpler and larger one: the meter has to be the meter the guarantee names. Read at the POI revenue meter, the result is already net of PCS conversion loss, medium-voltage transformer no-load and load losses, collection-cable heating, and — if station service is tapped on the plant side of that meter — the auxiliary load that ran throughout the discharge.

Read at the DC terminals, none of that has been taken off. Deliverable AC at the POI commonly lands around 85-92 percent of DC nameplate at beginning of life, so the same discharge reads on the order of ten percent lower at one meter than the other.

Station service is the line that decides borderline results. Tapped on the plant side of the named meter, the HVAC and controls burn is netted out of export and sits inside the number; fed from a separate grid-side supply, it sits outside it. Neither wiring is wrong and both are common, but a guarantee written on one and a test executed on the other will disagree about a perfectly healthy plant. Find where the auxiliary transformer lands on the single-line diagram before anyone argues about megawatt-hours.

Normalize before you compare

The test happens on the day's weather and the day's grid; the guarantee was written at reference conditions. There are two honest ways to close that gap — correct the measurement to the reference conditions using published coefficients, or redraw the guaranteed figure at the day's conditions and compare there. Either is defensible.

Choosing between them once the raw numbers are in is not, which is why the correction method, its data source, the reference conditions it targets and the acceptance band belong in the procedure both parties sign. Performance normalization covers the mechanics; what belongs in the test procedure is the commitment to one of them.

The same procedure should say what happens when the measurement cannot be taken cleanly: a market window that closes mid-discharge, an operator instruction that interrupts it, a block that trips at hour three. A rule for aborting and re-running, and a rule for whether a partial discharge may be extrapolated, are cheap to write in advance and expensive to negotiate at two in the morning with a witness standing on site.

What the number becomes

It decides acceptance first. COD capacity tests commonly require demonstrating roughly 95 to 100 percent of the contracted MWh at the POI, with buy-down damages below the guarantee — and that threshold is a contract term, not a standard, so the same measured result can clear a 95 percent floor and trigger a payment against a 97 percent one. What the test cannot do is settle the remedy. It supplies the measurement and is silent on what a shortfall costs.

Then it becomes the baseline. Where the warranty is baselined on as-measured commissioning capacity, this number anchors the retention curve for the life of the asset; where it is baselined on nameplate instead, a test that comes in above nameplate hands that surplus to the supplier as headroom rather than to the owner as energy.

Either way the record has to survive a decade: the measurement point, the logged cell-temperature trace, the resting SOC, the rate actually held, the correction applied and the source of its coefficients. A baseline nobody can reproduce lasts exactly as long as nobody challenges it.

Three failures recur. Comparing a DC-terminal result against a POI obligation, which is the whole loss chain expressed as a dispute. Running the periodic retest as a lighter version of the commissioning test — different rate, different rest, warmer day — so the trend line ends up measuring the procedure rather than the plant. And testing at flattering conditions to clear the acceptance gate, which buys a pass at commissioning and charges for it at every retention test afterwards, because the baseline it sets is the number the plant now has to keep beating.