Metering Accuracy & Billing Error

The class is not the system accuracy

A meter marked class 0.5S is accurate to ±0.5 %. That describes the meter. What you are billed on is the meter and the current transformer feeding it, and on most installations the CT is the weaker of the two.

Two independent errors do not simply add. Adding them gives the worst case — both wrong in the same direction at the same moment — which is possible but pessimistic. The realistic figure combines them in quadrature:

worst case = e_meter + e_ct
likely     = √(e_meter² + e_ct²)

A class 1.0 meter on a class 0.5 CT is ±1.5 % at worst and ±1.12 % realistically. On a large industrial supply, that band is a substantial amount of money.

Accuracy is a function of load, and this is where it breaks

Every class is quoted at or near rated current. Below a threshold the permitted error widens, and below a lower one it is not specified at all.

Look at the CT rows. A plain class 0.5 CT is specified from 20 % of its rating; the same class with an S is specified from 1 %. On a plant that idles overnight and at weekends, a large share of the year is spent below 20 % — and for those hours the plain CT has no stated accuracy whatsoever.

This is why an oversized CT ratio is expensive twice over. It pushes normal load down the range, and the range is exactly where the accuracy stops applying. Pick the ratio properly with the CT ratio calculator.

When the number actually matters

Accuracy costs money to buy, so it is worth being clear about when it earns that back:

• Money changes hands between two parties — tenant billing, a landlord recharging a shop, a captive plant selling to a neighbour. Use 0.5S on both meter and CT, and be prepared to defend it when somebody disputes a bill.
• Reconciling against the utility meter. If your submeters total 3 % away from the DISCOM bill, you cannot tell whether that is a real unmetered load or just accumulated class error. Better metering removes the ambiguity.
• Measuring a saving. If you are claiming a 4 % improvement using instruments with a ±1.5 % band, the claim is barely outside the noise.
• Plain monitoring — trending, spotting a failing motor, comparing shifts — is entirely fine on class 1.0. The comparison is against yesterday's reading from the same instrument, and a consistent error cancels out.

What this calculation leaves out

The class limits are laboratory figures at unity power factor and rated conditions. Real installations add more:

• Power factor. Class limits widen at low power factor — typically to 1.5 times the unity-pf figure at 0.5 lagging. An inductive plant is being metered at the wider limit all day.
• Burden. A CT outside its VA rating saturates and reads low, by far more than its class. That error is not in any of the tables here, and it is the commonest real cause of a metering discrepancy.
• Harmonics. Class limits assume a sine wave. Distorted current is measured differently by different meter designs, and a plant full of drives can push a meter outside its class without anything being faulty.
• Drift and temperature. Class is an as-new figure at reference temperature. Meters drift; a panel runs hot.

So treat the output here as a floor on the uncertainty, not a complete budget. If the number matters commercially, the answer is a calibration certificate for the actual devices, not a class marking on a datasheet.

A CT-operated meter inherits the transformer’s error

Above about 100 A the meter no longer sees the load current; it sees whatever the current transformer gives it. The class printed on the meter describes the meter alone, and the accuracy of the measurement is the combination of both.

Errors add, and they add in two dimensions. A CT has a ratio error — the secondary current is not exactly the primary divided by the ratio — and a phase displacement, which shifts the current waveform relative to the voltage. Ratio error affects every reading. Phase displacement affects energy measurement specifically, because energy depends on the angle between voltage and current, and its influence grows as power factor falls. At unity power factor a phase error contributes almost nothing; at 0.5 lagging it dominates.

So a class 0.5S meter behind a class 1.0 CT is a class 1-and-a-bit measurement, not a class 0.5S one. For billing-grade metering the CT must be at least as accurate as the meter, and the burden calculation on the CT ratio page is part of achieving it — an over-burdened CT drifts out of its class regardless of what was paid for it.

What a calibration certificate is actually claiming

A certificate records how the instrument behaved on one day, on a test bench, at a small set of specified points — typically a few currents at unity and at 0.5 lagging power factor, at nominal voltage and 50 Hz, at around 23 degrees.

It is a genuine and useful statement, and it is narrower than it looks. It says nothing about behaviour at 4 per cent of rated current, at 45 degrees inside a metering cubicle, on a distorted waveform, or after three years of drift. Electronic meters are stable but not immune; reference voltage sources and shunt resistors both age.

Two things make a certificate worth having. Check that the calibration points bracket the load you actually run — a certificate taken at 50 and 100 per cent of rating tells you very little about a feeder that spends its life at 8 per cent. And check the stated measurement uncertainty of the test itself, because a bench that is only three times better than the meter under test cannot meaningfully confirm the meter's class. Where the reading settles a commercial dispute, periodic recalibration against a traceable reference is the only answer, and most utility agreements specify an interval for exactly that reason.