Specifying a G99 type-test bench
Every figure below was read out of ENA EREC G99 Issue 2 itself, and carries the clause, table or annex it came from. Nothing here is transcribed from a secondary summary, and nothing is carried over from IEEE 1547.
G99 applies in GREAT BRITAIN. Northern Ireland uses ENA EREC G99/NI, and the texts differ materially — the GB document requires operation between 51.5 and 52 Hz for at least 15 minutes where the NI document says 60. Saying 'the UK' here is a small error that a DNO engineer will notice.
The settings the bench has to exercise
LV values shown; HV settings are percentages of the connection-point nominal.
| Function | Setting (LV) | Delay | Clause |
|---|---|---|---|
| Under-voltageSINGLE stage. See the corrections below — a two-stage under-voltage is G59, not G99. | −20% (184 V) | 2.5 s | Table 10.1 · 10.6.3 |
| Over-voltage stage 1 | +14% (262.2 V) | 1.0 s | Table 10.1 · 10.6.4 |
| Over-voltage stage 2Not specified for Type D or ≥50 MW — that column is blank. | +19% (273.7 V) | 0.5 s | Table 10.1 |
| Under-frequency stage 1 | 47.5 Hz | 20 s | Table 10.1 · 10.6.6 |
| Under-frequency stage 2 | 47.0 Hz | 0.5 s | Table 10.1 · 10.6.6 |
| Over-frequencySINGLE stage. There is no 51.5 Hz trip stage — see the corrections. | 52.0 Hz | 0.5 s | Table 10.1 · 10.6.5 |
| RoCoF (loss of mains)We publish the voltage and frequency settings because we read them in the standard's own text. We do NOT publish a RoCoF figure here, because our extraction of Table 10.1 could not resolve it cleanly and we will not reprint a protection setting we have not seen. Open the source and read the row. What IS verifiable is the behaviour: the delay starts when measured RoCoF exceeds the threshold, resets if it drops below, and the relay must not trip unless the rate stays above it continuously — and a relay implementing that delay as a cycle count is explicitly not acceptable. | see Table 10.1 | see Table 10.1 | Table 10.1 Note (1) |
What that demands of the source
- Voltage step resolution≤ 0.5%Annex A.7.1.2.2 (PPM), A.7.2.2.2 (synchronous)
- Applied in steps of ±0.5% or less of the voltage setting, held longer than the trip delay, starting at least 4 V below or above it.
- Frequency step resolution0.05 Hz steps, or a ramp under 0.1 Hz/sAnnex A.7.1.2 / Form A2-3
- Trip frequency is recorded to the setting ±0.1 Hz; the delay is measured with a 0.3 Hz step.
- Source accuracy the standard impliesvoltage ±1.5% of nominal, frequency ±0.2% of nominal10.6.7.1, after Table 10.1
- This is the relay accuracy G99 permits. A source less accurate than the relay it is testing cannot resolve a pass from a fail — and it is almost never quoted on a datasheet.
- Phase-jump capabilitysingle-cycle ±50 degreesAnnex A.7.1.2.6 / A.7.2.2.6
- Positive shift from 49.5 Hz, negative from 50.5 Hz, by extending or shortening ONE cycle, then held 10 s. A bench that cannot synthesise this cannot complete type testing.
- Fault ride-through profile140 ms dip to 30% (synchronous) or 10% (Power Park Module)Annex B.4.4.2, Table B.4.1
- Plus 140 ms unbalanced faults — phase-phase, two-phase-earth and single-phase-earth.
- RoCoF stability sweep49.0 → 51.0 Hz at 0.95 Hz/s over 2.1 s, and the reverseAnnex A.7.1.2
- Must NOT trip. The bench has to produce a controlled rate of change, not just a step.
- LFSM-O droop verificationmeasured 50.4 → 51.15 Hz; a nominal 10% droop passes between 8.5% and 12.8%Annex A.7.1.3
- Frequency tolerance ±0.05 Hz, power tolerance ±10% of the required change.
- Reconnection timerminimum 20 s after voltage and frequency returnAnnex A.7.1.2
- The sequence has to wait it out, so it is bench time on every iteration.
Which tests apply, and what the bench must do
A routing aid, not advice. Where the standard leaves a value to the DNO, this says so rather than inventing one.
Settings, fault levels and several protection choices are set by your DNO per connection. This tool does not guess at them. Send the connection details and we will size the bench against them.
Seven things this market repeats that G99 does not say
Most of these are G59 settings that outlived the document they came from. A bench specified around them is specified around a superseded standard, and a DNO engineer will read it that way immediately.
Commonly said — 01
“Under-voltage has two stages — 207 V at 2.5 s and 184 V at 0.5 s”
What Issue 2 actually says: G99 specifies a SINGLE under-voltage stage: −20% with a 2.5 s delay. Clause 10.6.3 says so explicitly, to maximise transmission fault ride-through capability. 207 V appears once in the whole document, in Table 10.2, which governs Infrequent Short-Term Parallel Operation — a different regime.
Why it matters: This is a G59 setting. A bench specified around it is specified around a superseded document, and a DNO engineer will read it as such immediately.
Commonly said — 02
“Over-frequency trips at 51.5 Hz”
What Issue 2 actually says: Over-frequency is a single stage at 52.0 Hz with a 0.5 s delay. 51.5 Hz appears only as a ride-through boundary — the module must keep operating between 51.5 and 52 Hz for at least 15 minutes — never as a trip setting.
Why it matters: Confusing a ride-through floor with a trip setting inverts the requirement: it turns something the equipment must survive into something it must react to.
Commonly said — 03
“Vector shift trips at about 6 degrees”
What Issue 2 actually says: Vector shift is NOT permitted as loss-of-mains protection in Great Britain. Clause 10.4.11 says so, and it was removed by modification DC0079 in 2018. There is no vector-shift trip setting in degrees anywhere in Issue 2.
Why it matters: And yet a ±50 degree vector-shift STABILITY test is mandatory for type verification — the module must not trip. Banned as a protection, required as a test. Getting that backwards means buying the wrong bench.
Commonly said — 04
“Fault ride-through applies to every G99 module”
What Issue 2 actually says: For Type A it is conditional. Clause 11.3.1 applies it only where the DNO and the Generator have specifically agreed the facility will contribute to network security.
Why it matters: Specifying an FRT-capable bench for a Type A product can be money spent on a requirement that was never going to apply.
Commonly said — 05
“LFSM-U applies alongside LFSM-O”
What Issue 2 actually says: LFSM-U appears only in Section 13 — Types C and D. Types A and B have LFSM-O and no LFSM-U.
Why it matters: It is a whole test regime that either does or does not apply to your product.
Commonly said — 06
“The droop setting is 10%”
What Issue 2 actually says: 10% is the MAXIMUM permitted droop, which is the MINIMUM required response — at least 2% of output per 0.1 Hz. Lower droop means more response and is permitted, down to 2%.
Why it matters: Stated as a fixed value it reads as a target rather than a limit, and the direction is easy to get backwards.
Commonly said — 07
“The trip time is the delay setting”
What Issue 2 actually says: Note (2) to Table 10.1 puts total operating time to circuit-breaker opening of the order of 100 ms LONGER than the delay setting. G98 states the tolerance as −0 s +0.5 s.
Why it matters: A bench that records the delay setting as the measured trip time is measuring the wrong thing.
What the standard leaves to your DNO
A specification that pretends these are fixed will overpromise. They are not ours to state, and they are not a supplier’s to guess.
- HV settings are percentages, not volts — a value to suit the nominal voltage of the connection point, so 11 kV, 6.6 kV and 33 kV give different volts from the same clause.
- The 2.5 s under-voltage delay may need reducing where auto-reclose times are under 3 s.
- The DNO may specify additional over-voltage protection at the connection point.
- Droop anywhere between 2% and 10% is a design choice.
- Fault levels for ride-through come from the DNO. The 50 MVA in Annex B.4.4.3 is a generic fallback, not your number.
- Alternative settings can be agreed in exceptional circumstances and recorded in the Connection Agreement — after which they cannot be altered without the DNO's written agreement.
Sources
- ENA Engineering Recommendation G99 Issue 2 (2025) · Energy Networks Association, published under the GB Distribution Code · retrieved 2026-09-06
- ENA Engineering Recommendation G98 Issue 2 (2025) · Energy Networks Association · retrieved 2026-09-06
- ENA document catalogue — G99 Issue 2 status · Energy Networks Association · retrieved 2026-09-06
- DCRP/MP/26/02 — Distribution Code proposed changes · Ofgem · retrieved 2026-09-06
Sizing one against a real connection
Send the connection details and the device under test. Several of the values above come from your DNO rather than the standard, and the bench has to be sized against those, not against a generic table.