Efficiency fade

What it is (precise)

Efficiency fade is what remains after you hold everything else still. Round-trip efficiency moves hour to hour with dispatch — C-rate, depth of discharge, ambient temperature, partial-load PCS operation, idle time carrying auxiliaries — and none of that is fade, because the conditions give the points back when they return.

Fade is the residual: the same cycle, at the same boundary and the same reference conditions, returning less energy in year eight than it did at commissioning. That is why proving it is a comparison problem rather than a measurement problem, and why a retest run to a different protocol than the commissioning test cannot settle the question. The Round-trip efficiency entry owns the ledger and the boundary discipline; this entry is about the age term inside it.

Not every line of that ledger has an age term, and separating the ones that do is the whole diagnosis. Cell ohmic loss grows, because internal resistance grows. Auxiliary energy grows, partly because the extra ohmic heat has to be rejected and partly because a faded fleet delivers fewer megawatt-hours per hour of conditioning.

PCS conversion loss and transformer load and no-load losses, by contrast, carry no meaningful electrochemical aging term — a ten-year-old transformer has much the same loss curve it shipped with. So when a net AC number at the POI drifts down over a decade, the movement sits almost entirely in the battery and in station service, which is where any investigation should start rather than in the conversion chain.

The magnitude is modest and the visibility is early. A drift of one to two points from beginning of life to end of life is the working expectation for utility-scale LFP, small enough that it hides inside ordinary operational scatter and large enough to matter commercially. It also surfaces before its siblings: an RTE retest will show the drift while the energy capacity test still passes comfortably, because a plant needs to lose twenty-odd points of retained capacity to reach its contractual floor and only one or two points of efficiency to breach a typical guarantee tolerance.

The three channels that move the number

The first channel is resistance in the energy account. The ohmic loss fraction per direction is roughly I × R over cell voltage — linear in resistance at a given current, and paid twice per cycle, once charging and once discharging — so the same growth curve that the Resistance growth entry tracks as heat shows up here as percentage points of round trip.

The mechanism belongs to that page; what belongs here is the compounding. Resistance growth commonly steepens late in life, and the same growth that adds loss also adds heat, which the cooling system spends auxiliary energy rejecting, which lands on the same net AC number a second time.

The second channel is where in the window the fleet ends up working. Holding nameplate MW on a faded pack raises the C-rate as a normalised figure — a system built at 0.25C is running near 0.3C once retention reaches about 80 percent, because the same power is divided by a smaller capacity.

The current does not follow the capacity, though. At fixed power it follows the voltage, and what the fade changes is where the discharge sits: the same delivered megawatt-hours now come out of a smaller pack, so more of every cycle runs in the low-SOC region where terminal voltage is lowest and sag is worst.

The ohmic loss fraction there goes as R over V squared, rising with aged resistance and with the square of whatever voltage the plant gives up. That is the channel — not more current at the same conditions, but more of each discharge spent at the conditions where current is highest, with aged resistance carrying it. It is also why a short-duration, high-C-rate asset fades in efficiency faster than a four-hour arbitrage asset on identical cells, and why an efficiency figure quoted from a gentle test cycle travels badly to an aggressive duty.

The third channel is the denominator. Auxiliary consumption typically runs 1 to 3 percent of annual discharged energy, and much of it — thermal conditioning, controls, communications, transformer no-load loss — accrues by the hour rather than per megawatt-hour. Let the fleet fade to 80 percent retention with the hour-based burn unchanged, and a 2 percent auxiliary share becomes about 2.5 percent of a smaller delivered total, with no change in the auxiliary equipment at all.

Augmentation does not escape the arithmetic, it relocates it: restoring the delivered megawatt-hours means adding enclosures, and each one brings its own conditioning load. Either the throughput shrinks under a fixed burn or the burn grows to hold the throughput, and both land on the same net AC line.

Why the efficiency guarantee is thinner than the capacity warranty

The two instruments are shaped differently. Capacity gets a year-by-year retained-energy table, an operating envelope that conditions it, a test procedure and a remedy, all of it standard commercial furniture. Efficiency usually gets one guaranteed net RTE at stated conditions, verified at commissioning and sometimes at periodic retests, with a tolerance around one percentage point as common negotiating ground.

That difference is the practical definition of the exposure: the aging axis of efficiency generally has no contractual curve to fall below, so the years between retests — and every year after the last one — are uncovered by anything except the owner's own trending.

The reason is testability, and the arithmetic explains why the drafting is cautious. RTE is a ratio of two large metered numbers, so a one-point tolerance is a very tight budget. On a cycle delivering 400 MWh at the POI at 88 percent, about 455 MWh went in; one percentage point of RTE is roughly 4.5 MWh of that, which is a little over 1 percent of the discharge.

Any state-of-charge endpoint that does not close — the cycle finishing a percent of usable energy away from where it started — consumes the entire tolerance on its own, before metering error, temperature correction, or the question of which auxiliaries were energized during the test. A capacity test has no such symmetry requirement: it meters one direction and compares against a table.

Everything that makes a single RTE test hard makes a fade measurement harder, because fade is the difference of two such tests years apart. The conditions have to be reproduced: same boundary, same auxiliary treatment, same C-rate and depth of discharge, same temperature reference and correction curve, same dwell between charge and discharge.

The IEC 62933 series describes test methods for electrical energy storage systems, including how auxiliary consumption is accounted in efficiency measurement, which is the right hook to name in a protocol. The contractual questions follow from all of this: is there a retest obligation at all, on what interval, at whose cost, against which baseline value, and does the plant retain the raw commissioning test data needed to reproduce the comparison a decade later.

What it costs in an arbitrage model

Efficiency enters the trade as a multiplier on the purchase, not as a haircut on the sale. The effective charging price is the market price divided by RTE, so the break-even condition is a discharge price of at least 1/RTE times the charge price — about a 14 percent premium at 88 percent, about 16 percent at 86.

A two-point drift over project life therefore widens the spread the plant needs before a cycle is worth taking, and it does so in markets where spreads compress as more storage arrives. Marginal cycles are the ones that disappear first: the deep evening peak still clears easily, while the shoulder trades that were barely profitable at commissioning stop clearing at all.

The volume view is blunter. Hold a 400 MWh delivered cycle and step net AC RTE from 88 to 86 percent, and the purchase rises from about 455 to about 465 MWh — roughly 10.5 MWh more bought per cycle for the same energy sold. An arbitrage-weighted asset commonly performs 250 to 365 equivalent full cycles a year, so at 300 cycles that is on the order of 3,000 MWh a year of energy bought and never resold, recurring for the rest of the life and priced at whatever the charging hours cost.

The number belongs in the model as a declining-RTE profile rather than a constant, alongside the retained-capacity curve, because the two curves bend the revenue line in different places: capacity fade shrinks what the plant can sell, efficiency fade raises what it must buy to sell it.

Common pitfalls

The most common error is calling a duty change a fade. A fleet-average RTE computed over a quarter of shallow cycling, partial-load dispatch and long idle hours will sit below the commissioning result on healthy hardware, and comparing that figure to the test value produces a fade that does not exist.

The inverse error is equally common: a favourable quarter masking real drift. Trending only comparable cycles — matched C-rate, matched depth, matched temperature band, closed state-of-charge endpoints — is the only way the operational record answers the question, and it is worth defining that filter in the EMS historian while the commissioning data is still fresh.

Two more traps are contractual. The first is assuming the capacity test covers efficiency: a plant can pass its retained-energy table for years while the RTE guarantee is already breached, which is precisely the asymmetry the four-fade taxonomy exists to separate. The second is assuming augmentation is the remedy.

New racks restore megawatt-hours and pull the fleet-average resistance down, but the aged racks keep their resistance, the added enclosures add auxiliary load, and mixed-age blocks bring their own operating-window constraints. There is no augmentation clause that resets efficiency the way it resets energy, so the sensible protection is written up front: a stated baseline, a defined retest protocol, and access to the per-rack resistance history the BMS is already recording.