Free · Control modes

Grid-forming vs grid-following, and when each one is right.

Which control law the converter runs sets its fault behaviour, how weak a node the plant can live at, and which clauses bind it.

3 of 36Parameter tables naming a mode
~1.1–1.5 puTypical fault ceiling, both modes
10 msClause-stated GFM reaction
EMT / RMSModel class, named per claim

The two control modes

Grid-forming vs grid-following — the two control modes, side by side.
GRID-FORMING (GFM)GRID-FOLLOWING (GFL)voltage sourcesets its own V, angle & frequencycurrent sourcetracks the grid through a PLLresponse: inherent, typ. sub-cycle (<10 ms)response: measured loops (~100 ms and up)can hold SCR ≈ 1 grids — if designed for itPLL typically loses grip below SCR ≈ 2–3fault current: inverter-limited, typ. 1.1–1.5 pufault current: inverter-limited, typ. 1.1–1.5 publack-start: only where designed & ratedneeds a live voltage to lock ontoinertia & system-strength servicestoday's fleet workhorse — cheap & provenIn one line: grid-forming sets the voltage and frequency;grid-following injects current into a waveform something else is holding up.

Numbers from the term's Key Facts, read as typical of the class, not guarantees of the mode: a GFM design responds inherently — usually inside ~10 ms — and can hold SCR ≈ 1 grids where a GFL PLL typically loses its grip below SCR ≈ 2–3. Sub-cycle response, weak-grid stability and black-start are designed, rated capabilities of a specific implementation; none comes automatically with grid-forming control. The fault-current ceiling (typically ~1.1–1.5 pu, briefly) is a hardware thermal limit shared by both modes — the difference is what the control does at that limit. The droop mechanics behind the GFM card are in the interactive Grid-Forming Droop visual in the resources.

The note's Key Facts and droop visual live on the grid-forming entry.

At the terminals

What is the difference between grid-forming and grid-following?

Grid-following locks a phase-locked loop onto the voltage already there and injects a commanded current into it, so it cannot energize a dead network. Grid-forming imposes its own voltage magnitude, angle and frequency from an internal reference, so power falls out of the difference against the network, as for a machine.

Roles, not generations

Is grid-forming better than grid-following?

No. Two control laws with different jobs, neither replacing the other; the upgrade framing is the commonest error in public writing. Grid-following runs most of the utility-scale fleet and stays correct at a strong connection point, where its assumption of a stiff voltage holds. Grid-forming answers a different problem. A phase-locked loop estimates the grid angle from a voltage the converter is itself moving; as system strength falls that estimate degrades, and converters sited close together start oscillating against one another. Finland names the band — 1 to 15 Hz voltage oscillations in grid-following converters in a weak network — as one a plant must not amplify. A grid-forming converter imposes an angle instead of estimating one. It is not free either: a rated grid-forming response is backed by converter current headroom, and whether that headroom must be bought as oversizing belongs to the code and the contract rather than to the mode.

In a fault

Both modes hit the same wall

A grid-following loop is already commanding current, so the limit clips a command it was issuing anyway. A grid-forming controller is commanding a voltage, so how it limits is a design choice — saturating the current reference hands voltage control away while the limit binds; a virtual impedance holds it as a voltage source through the fault.

All three codes below legislate that choice, and disagree about how far the guarantee reaches. Finland writes it flat: restricting the current must not interrupt the grid-forming controller, and the limits must match the plant's real capability and use its short-term overload. Switzerland writes nearly the same sentence and scopes it — once the converter reaches the limit, it says, no grid-forming control can be guaranteed, so the requirement holds only below that point. Chile writes it as two thresholds: a phase jump of at least 30 degrees ridden with the limiting functions off, then stable operation with them on. In none of the three is it the vendor's call.

Every number above is about a simulation until someone names the model class that produced it.

What the codes ask for

Where is grid-forming actually mandatory?

Three jurisdictions, and all three mandates are recent. Chile requires every storage installation interconnecting to the national system to be grid-forming, with the operator able to instruct a change back, and transitional carve-outs that turn on the date the plant energises. Switzerland requires converter-based transmission-connected storage to use grid-forming-capable converters, active whenever connected. Finland requires storage to operate constantly in grid-forming mode, Type C and, by inheritance, Type D. Those are 3 of the 36 codes whose parameter tables are published here — a fact about these tables, not a finding that the other 33 permit both modes. The grid-forming entry names what Great Britain and North America specify instead, and how binding each one is.

Chile

Grid-forming is the default for storage, and the fault clause behind it asks for a reaction inside 10 ms, reactive current above the pre-contingency value without exception, and negative-sequence injection on asymmetrical faults. Transitional articles carve out plant already declared in construction — except storage energising more than six months after publication, which must comply unless the Comisión grants a justified exemption.

AT-IBR, Enero 2026 — approved by Resolución Exenta CNE N°45 of 28 January 2026

Chile — the clause and the quote

Finland

An operating mandate, not a capability: the control must be live whenever the plant is connected, and the mode is barred from changing on a network disturbance. It sits in the Type C requirements and the Type D section inherits it: from 10 MW below 110 kV, and at any size from 110 kV up. Which clauses you answer is decided by type and never by mode: Type C is excepted from the reactive-current supply written for Type B; Type D inherits both sections and owes both.

SJV2024 (storage), not VJV2024 (generating facilities) — Fingrid’s unofficial English translation, filed on this site under its VJV2024 record

Finland — the clause and the quote

Switzerland

Grid-forming capability must be active whenever the plant is connected, in generation and in demand mode and regardless of state of charge, and it is written as behaviour rather than as a label: counteract a phase jump by supplying active and reactive power, raise local voltage stiffness by supplying reactive power on an amplitude change, within 10 ms of the change.

Technische Mindestanforderungen kESS (Swissgrid), Version 1.1 of 29 May 2026

Switzerland — the clause and the quote

For most projects the other direction matters more: 65 of the 66 codes in the ride-through atlas set their envelopes without reference to mode — Chile, which draws a figure for each, is the exception.

Procurement

How a buyer decides

  1. Is black start or islanded operation in scope? Then it is decided: a grid-following converter cannot energize a dead network or hold an island alone. Black start is the one binary here, and it must be specified and tested, never inferred.
  2. Otherwise start at the connection point. Short-circuit ratio is fault MVA at the node over the plant's own MW rating, so resizing changes it — double the rating and you halve the ratio. Get it for your point of interconnection from the operator, and ask which definition.
  3. Read the code before the datasheet: most write duties both modes must meet, and a few write the mode into the requirement.

That is enough to decide. Code names the mode, or the ratio is at or below the vendor's stated minimum: grid-forming, with the project-specific stability study budgeted. Code silent and the ratio still clear of that minimum in the weakest credible case — after the outage that thins the network, and with the neighbouring converters counted against you rather than ignored — grid-following. Chile writes that second test into law for a grid-following plant, which must verify a short-circuit ratio of at least 3 and an equivalent ratio, counting neighbouring converters against you, of at least 1.5. Between the two it is a study question before a mode question: the weak-grid one. Which model class answers it, and what the answer has to say, is not a detail. A service you intend to sell that prices the capability overrides the ratio.

Then the question that outlives it. Grid-forming is often — not always — the same hardware running different control software; Chile assumes as much, limiting an instructed mode change to software and communications. So ask each bidder whether the unit is grid-forming capable or field-upgradeable, what it costs, what headroom backs the rated response at your ratio, and what type test and model delivery stand behind the answer: a capability answer is worth what its evidence is, and the grid-forming and grid-following entries carry what to demand. A plant studied in one mode must be re-studied for the other.

Whether that headroom must be bought as oversizing belongs to the code and the contract, and the three are not answering about one quantity. Chile makes at least 1.3 pu of rated current for 5 seconds a design requirement, with reactive and active overcurrent margins held against it. Finland and Switzerland say only that the storage capacity need not be overdimensioned for grid-forming control — a different quantity, and no waiver of the current headroom. A reserved state-of-charge band is a market-and-contract matter too, and belongs in the capacity warranty.

Go deeper

The controls, the physics, the contracts.

The control hierarchy, the PCS, interconnection and the market products run as modules in the Complete Guide.