Roughly 40% of new utility-scale BESS capacity ordered globally in 2025 was specified grid-forming, up from under 15% in 2022 — and the reason is not policy fashion. It is the diminishing system strength of inverter-dominated grids. When synchronous generation drops below a critical share of dispatched capacity, a purely grid-following (GFL) fleet loses the voltage and frequency reference it needs to lock onto. The inverters that were once passive followers must become the reference themselves.
This is an engineering problem with a specific arithmetic. A grid-forming (GFM) converter is not a bigger GFL converter — it is a different control loop, with different specifications for power rating, energy headroom, and response time. Getting the architecture wrong shows up as frequency nadir excursions below the trip threshold, not as a line item in fail-to-perform. This guide covers the three control families, the sizing rules that follow from them, and how to represent each correctly in a simulation.
Key figure: A grid-forming BESS must reserve roughly 20-30% of its PCS rating as headroom above its energy rating. A 100 MW / 400 MWh system with a 1.0 C-rate PCS (100 MW) cannot deliver a 100 MW inertial response and simultaneously sustain its contracted discharge — the transient share and the steady-state share draw from the same converter. Overplanting the PCS to 125-130 MW is the standard remedy, and it moves CAPEX by 4-7% on a typical 4-hour system.
Contents
Why Grid-Following Broke the Specification
A grid-following inverter synchronises to the grid using a phase-locked loop (PLL). It measures the voltage angle at its terminals and injects current in phase with it. That is a closed-loop system with a reference it does not generate. The consequences appear as two failures that compound:
- No natural inertia. A synchronous machine's rotating mass resists frequency change as df/dt. Inertia constant H is typically 3-6 seconds for a large thermal unit and 2-4 seconds for a hydro unit. Modern inverter-based resources contribute H ≈ 0 unless explicitly emulated. As synchronous share falls, so does system H, and frequency nadir after a contingency deepens proportionally.
- PLL instability in weak grids. As the short-circuit ratio (SCR) at the point of connection falls below roughly 3, the PLL's dynamic response interacts with grid impedance and can oscillate. Below SCR ≈ 2, many GFL inverters become unstable and trip. This is the mechanism behind the multiple solar-farm oscillations documented in weak-grid networks.
The IEEE 1547-2018 standard moved in this direction earlier than most markets: it defines normal-frequency ranges of 59.3-60.5 Hz in 60 Hz systems and 49-51 Hz in 50 Hz systems, and mandates frequency ride-through rather than fast trip. Grid codes in Australia (AEMO), the UK, and increasingly the Middle East go further, requiring a defined active power response to frequency deviation. A GFL inverter can provide frequency-watt droop — a slow, deliberate curtailment response — but it cannot provide the pre-contingency voltage reference or the immediate inertial injection.
Three Grid-Forming Control Architectures
Grid-forming control is not one thing. The three families in commercial deployment differ in how they derive the internal voltage reference and how they synchronise to neighbouring machines.
Active power droop (P-f droop)
The converter synthesises an internal voltage phasor and varies its frequency as a linear function of active power error around a setpoint:
Δf = -kp × (Pmeas - Pset)
Droop coefficient kp is expressed as a per-unit frequency change per per-unit power change, typically 2-5% for a GFM battery. A 5% droop on a 50 Hz system means full rated power output corresponds to a 2.5 Hz frequency offset — a large number in isolation but the correct order of magnitude when multiple units share load according to their droop settings.
Droop is fast — its response is limited only by the inner current loop and measurement filter, typically 20-100 ms. But it provides no inertia: frequency moves immediately in proportion to power imbalance, exactly as it would with no rotating mass. It is a steady-state sharing mechanism, not a transient one.
Virtual synchronous machine (VSM)
VSM embeds the swing equation directly in the converter controller. The measured power error drives a virtual rotor:
2H × dω/dt = Pset - Pmeas - D × (ω - ωref)
Here H is an adjustable inertia constant in seconds and D is a damping coefficient. Because H is a software parameter, a 100 MW battery can be configured with H = 10 s — inertia equivalent to a 1,000 MVA synchronous machine at H = 1 s, or a 250 MVA machine at H = 4 s. That flexibility is the single largest advantage of VSM: the system operator can specify the synthetic inertia it needs, and the plant can be tuned to deliver it.
The cost is complexity. The swing equation introduces a second-order dynamic that must be damped against the PLL-free synchronising loop, and tuning H against D against the inner current loop is not a trivial commissioning exercise. Set H too high and the converter fights its own power limiter; set D too low and the response rings.
Synchronverter
The synchronverter approach models the converter as a synchronous machine including magnetising inductance and the electrical torque equation, rather than approximating with the swing equation alone. It is the most complete emulation and the least common in commercial BESS — the additional fidelity buys little in a battery system where the "prime mover" is a DC bus with no meaningful dynamics. Where synchronverters appear, it is usually in research-grade or specialised multi-port configurations.
| Dimension | P-f Droop | VSM | Synchronverter |
|---|---|---|---|
| Inertia emulation | None | Configurable H | Configurable H + M |
| Typical response time | 20-100 ms | 100-300 ms | 150-400 ms |
| Tuning complexity | Low | Medium-high | High |
| Weak-grid stability | Good | Very good | Very good |
| Multi-unit load sharing | Frequency-based, accurate | Frequency-based, accurate | Frequency-based, accurate |
| Commercial BESS prevalence | Common | Common (dominant for utility) | Rare |
Sizing Inertia Emulation: The Arithmetic
The question "how much synthetic inertia does the plant need to provide?" is answered by working backwards from the acceptable frequency nadir. For a system with total inertia constant Hsys (MW·s/MVA on the system MVA base), a loss of ΔP MW causes an initial rate of change of frequency:
df/dt = -f0 × ΔP / (2 × Hsys × Ssys)
Take a worked example. A 60 Hz islanded network with Ssys = 500 MVA and Hsys = 2.0 s loses 60 MW of generation (a unit trip). The initial RoCoF is:
- df/dt = -60 × 60 / (2 × 2.0 × 500) = -1.8 Hz/s
If the largest loss event is 60 MW and the frequency must stay above the IEEE 1547 under-frequency trip threshold of 59.3 Hz — that is only 0.7 Hz of headroom — the system survives for roughly 0.39 seconds before reaching the trip point, assuming no governor response. Under-frequency load shedding typically arrests the decline by 300 ms, but the margin is thin. Note carefully: this is a 60 Hz network, so the 59.3 Hz threshold applies. For a 50 Hz system the equivalent safe band is 49-51 Hz and the analysis is repeated with f0 = 50.
Now add 75 MW of grid-forming BESS with H = 6 s on its own 75 MVA base. Its inertia contribution to the system base is:
Hcontribution = Hbess × Sbess / Ssys = 6 × 75 / 500 = 0.9 s
System inertia rises from 2.0 s to 2.9 s, and the initial RoCoF improves from -1.8 Hz/s to -1.24 Hz/s — a 31% reduction in the rate of frequency decline. The headroom to the trip threshold extends from 0.39 s to 0.56 s. For a 75 MW converter to deliver this, it must be able to swing its power output by roughly 2 × H × Prated / t during the transient window, which for a 200 ms nadir means power excursions well above its steady-state rating.
PCS Overplanting and Energy Headroom
This is where most front-end engineering designs go wrong. A grid-forming converter must simultaneously serve three power demands that all draw from the same rating:
- Steady-state energy dispatch — the contracted MW it is delivering to the grid, whether from a PPA schedule or an economic dispatch signal.
- Transient inertial response — the power excursion required to arrest frequency decline, which can be 1.5-2.0× the steady-state injection for 100-300 ms.
- Reactive current for voltage support — during a fault or voltage dip, the converter must inject reactive current, and the combined apparent power cannot exceed PCS rating.
A PCS rated exactly at the energy rating — that is, C-rate 1.0 with a PCS/energy ratio of 0.25 for a 4-hour system — has no headroom for items (2) and (3). In practice this means either the inertial response is clipped (defeating the purpose) or the steady-state dispatch must be curtailed during frequency events (a contract violation in most markets).
Three remedies, in increasing cost:
| Remedy | Effect on PCS Rating | CAPEX Impact | Constraint |
|---|---|---|---|
| PCS overplant (DC/AC ratio < 1.0) | 125-130 MW for 100 MW energy | +4-7% | PCS thermal rating must be justified for the transient duty cycle |
| Reserve energy headroom in SOC band | Unchanged | +2-5% (larger battery) | Requires SOC reservation logic in the EMS; reduces usable energy |
| Power-limiting dispatch schedule | Unchanged | None | Forfeits revenue — usually the least attractive option |
The PCS overplant is normally the right answer, and it has a second benefit: a lower DC/AC ratio improves the converter's efficiency at partial load and reduces I2R losses on the AC side during the majority of operating hours when it is well below rating.
Simulating GFM Behaviour in Energy Optima
Representing grid-forming behaviour correctly requires four inputs to be present and consistent in the model. A simulated plant that produces a frequency trace without them is producing a plausible-looking number with no engineering basis.
- Inertia constant and droop settings per unit. In Energy Optima these live under the BESS-PCS configuration — the operating limits section exposes frequency-watt and volt-watt curves alongside the inertia constant. Set H to the value the PCS vendor has actually commissioned, not the marketing maximum.
- PCS rating separate from energy rating. The platform models PCS rating independently of nameplate energy capacity. Enter both — this is what allows the DC/AC ratio to be a design variable rather than a hardcoded assumption.
- Frequency response function expressed as a curve, not a single point. A grid code requirement that specifies a deadband, a droop slope, and a maximum power offset is a piecewise function. Model it as one.
- Concurrent reactive duty. The volt-watt and volt-var settings must be active during the frequency event, because the PCS must satisfy both simultaneously.
The platform's dispatch strategy selection matters here. RULE_BASED dispatch will honour a frequency-droop curve but does not co-optimise it against energy value. ECONOMIC_DISPATCH will price the opportunity cost of holding headroom. For a GFM plant whose primary revenue is capacity and ancillary services, MILP_HYBRID is usually the correct choice — it can co-optimise energy arbitrage against the reserve requirement within a single 24-hour horizon while respecting the physical PCS limit as a hard constraint.
Battery degradation interlinks with this. Every transient inertial response is a high C-rate pulse, and high C-rate cycling accelerates capacity fade. Energy Optima models SOH as a three-dimensional function of calendar year, C-rate, and cycles per day, using trilinear interpolation across more than 16,000 data points derived from real cell test data. A plant that provides aggressive synthetic inertia will see a measurably different degradation trajectory than one that does not — and the 25-year financial projection will reflect it.
Grid Code Compliance Checkpoints
Before the GFM architecture is frozen, verify these against the actual connecting market's code:
- Frequency band and trip thresholds. 50 Hz markets: normal band 49-51 Hz, with defined ride-through to lower values for limited duration. 60 Hz markets: normal band 59.3-60.5 Hz. Never mix the two in a sizing calculation.
- Required RoCoF withstand. Some codes specify a maximum df/dt the plant must ride through without tripping — 1 Hz/s or 2 Hz/s are common values. This sets a floor on the PLL and grid-forming loop bandwidth.
- Inertia or fast frequency response obligation. Markets in Australia, the UK, and several EU member states now procure this explicitly. The obligation translates directly into the H and kp settings above.
- Voltage support during faults. Reactive current injection requirements during voltage dips are specified separately from steady-state voltage regulation and must be modelled concurrently with the frequency response.
- Black-start capability, if claimed. Where the plant is contracted for black start, the GFM converter must form a voltage reference into a de-energised network — a different and more demanding test than parallel operation.
The engineering conclusion is that grid-forming specification is an integrated exercise. Inertia constant, droop slope, PCS rating, SOC reservation, reactive duty, and degradation all interact, and optimising them sequentially rather than simultaneously produces a design that passes each individual check and fails as a system. Model them together.