Grid-Forming BESS Inertia: How Synthetic Inertia Restores System Strength

A 200 MW battery system built with grid-following inverters contributes approximately 0.05 to 0.18 seconds of inertia to the grid. The same hardware, re-controlled as a grid-forming (GFM) resource, can be configured to deliver 2.5 to 12 seconds of effective inertia constant — a 20- to 60-fold increase in the machine-seconds of rotational energy that holds frequency steady during a contingency. That gap is the entire commercial and engineering case for grid-forming storage, and it is almost entirely a controls difference, not a hardware difference.

The reason it matters now is arithmetic. Synchronous generation is retiring faster than the frequency-response services that replace it are being procured. Inverter-based resources (IBR) — solar, wind, and increasingly the batteries bolted to them — carry no rotating mass. When they displace synchronous machines, system inertia falls, rate of change of frequency (RoCoF) rises, and the frequency nadir after the largest credible loss gets deeper and closer to under-frequency load shedding thresholds. Engineering teams writing grid-connection studies in 2026 are discovering that the binding constraint on new renewable capacity in weak-network regions is not the energy resource and not the battery cost — it is system strength.

The Inertia Deficit: Why IBR Displacement Changes Grid Dynamics

Classical power system dynamics rest on a comfortable assumption: at any instant, thousands of gigawatt-seconds of rotational kinetic energy are coupled to the grid through synchronous machines. The swing equation for a single machine relates that stored energy to frequency deviation:

2H · (dΔf/dt) = ΔP_m − ΔP_e − D·Δf

where H is the inertia constant in seconds (stored kinetic energy at rated speed divided by the machine's MVA rating), and dΔf/dt is the initial rate of change of frequency. H is the single number that determines how fast frequency moves in the first few hundred milliseconds after a generation loss. A large thermal unit has H between 2 and 6 seconds. A hydro unit sits at 2 to 4. A modern gas turbine, with a lighter rotor, is closer to 2 to 3.

An inverter is not a rotating machine. It has no kinetic energy to release. Its power electronics switch at kilohertz rates and its response is governed by control loops with time constants measured in milliseconds, not by physics with time constants measured in seconds. Consequently, when a GFL inverter is asked to support frequency, it does so only after measuring a deviation and running it through a controller — the defining characteristic of grid-following. The result is a system where the same megawatt of capacity displaces synchronous inertia without replacing it in the sub-second window where inertia acts.

The practical consequence is that in regions with high IBR penetration, the frequency nadir after a unit trip is set not by how much fast frequency response is contracted, but by how little inertia remains. National grid operators have responded by treating inertia itself as a procured service rather than a free by-product of generation — the UK's system operator was among the first to run dedicated stability markets of this kind, and Australia's NEM has pursued similar arrangements through its market bodies as coal retirements accelerated.

Grid-Following vs Grid-Forming: Two Ways to Be a Current or Voltage Source

The distinction that matters is a voltage-source versus current-source question.

• Grid-following (GFL) inverters regulate their output current using a phase-locked loop (PLL) to track the existing grid voltage angle. They behave as controlled current sources. They cannot establish a voltage or a frequency — they can only inject current into a voltage that already exists.
• Grid-forming (GFM) inverters regulate their terminal voltage and angle, behaving as controlled voltage sources behind an impedance, in the same conceptual way a synchronous machine does. They can establish voltage and frequency without an external reference, which is why the same control architecture underpins islanded microgrids and weak-grid support.

GFM is not one algorithm. The main families in commercial equipment today are:

• Droop control — active power is a function of frequency deviation (P–f droop) and reactive power a function of voltage deviation (Q–V droop). Simple, robust, well understood, but provides no intrinsic inertia unless augmented.
• Virtual Synchronous Machine (VSM) — a swing-equation emulator in software, so the inverter's power response includes a derivative term on frequency. The virtual inertia J (or equivalent H) is a tunable constant. VSM can be layered onto GFL hardware, which is why it appears in so many retrofit proposals.
• Virtual Oscillator Control (VOC) — an oscillator-based synchronization scheme that produces near-sinusoidal voltage references with strong coupling between units; attractive for high-inverter-share islanded systems.
• Droop with derivative / fast frequency response layers — droop plus an explicit df/dt term, functionally similar to VSM but implemented as a supplementary loop rather than a full emulation.

The engineering difference that separates VSM-on-GFL from true GFM is what happens at very low short-circuit ratio. VSM still depends on a phase reference to operate stably in some implementation variants, and its inertia contribution shrinks as the grid weakens because its control bandwidth is pushed against a resonant, high-impedance network. A true GFM inverter does not require an external phase reference at all — it produces the angle. That is the origin of the divergence shown in the chart below.

Two features of that chart are worth pausing on. First, the ordering never reverses: GFM leads at every SCR from 1.5 to 10. Second, the VSM curve is non-monotonic — it peaks around SCR 3 to 4 and falls away on both sides, a signature of a phase-locked implementation chasing a reference in a network that is either too weak (loss of reference integrity) or too strong (controller bandwidth insufficient to emulate meaningful inertia). A GFM converter's curve is monotonic because the capability is intrinsic rather than emulated through a tracking loop.

Quantifying Synthetic Inertia: The H Constant for a Battery

Engineers are trained to read H for a rotating machine. For a battery, H is a derived quantity, and it is easy to overstate. The honest formulation treats the battery as an energy-limited inertia emulator with three constraints: an energy budget, a power budget, and a duration over which the emulation is permitted to act.

Start from the swing equation and treat the available fast-response power as the inertial injection:

H_eff = (P_response · t_hold) / (2 · S_rated)

For a battery whose emulated inertia is declared over a support window t_hold, the effective H rises linearly with the response power and with the duration of the window, and falls with plant MVA rating. This is why inertia figures quoted by vendors are meaningless without a stated t_hold. A 10 MW injection sustained for 10 seconds on a 100 MVA plant is H = 0.5 s. The same injection sustained for 30 seconds is H = 1.5 s. Neither number is wrong — they describe different services.

Worked example, using a 200 MW / 400 MWh LFP BESS at a weak connection point (SCR 1.5, a genuinely difficult network):

• Declared support window: 10 seconds (aligned with typical fast frequency response product definitions)
• Allowable inertia response power: 40 MW (20% of rated, keeping the remainder for other contracted services)
• Effective H (per the relation above): (40 MW × 10 s) / (2 × 200 MVA) = 400 MW·s / 400 MVA = 1.0 s
• Energy consumed by the profile: a triangular 40 MW ramp over 10 s ≈ 200 MW·s = 0.056 MWh
• SOC impact: 0.014% of a 400 MWh system — energetically trivial

Note also the distinction between H declared on the plant's own MVA base versus on a system base. A 200 MW plant declaring H = 1.0 s on its own base contributes 200 MW·s of inertial energy. On a 20 GW system base, that is H = 0.01 s of system inertia — real but modest. Inertia is a portfolio service; single-asset contributions only matter when aggregated across tens of sites, which is precisely why grid operators set minimum procurement volumes rather than minimum per-asset ratings.

SCR and System Strength: Where GFM Becomes Mandatory

Short-circuit ratio is the ratio of the network's short-circuit capacity at the point of connection to the rating of the connected plant. It is the standard proxy for system strength. The conventional operating bands:

• SCR above 5: strong network. Grid-following inverters are stable with conventional PLL tuning and no supplementary controls. Inertia is not the binding constraint.
• SCR 3 to 5: moderate. GFL remains viable with careful tuning, but voltage control and damping requirements rise, and some operators now require frequency-support capability as a condition of connection.
• SCR 2 to 3: weak. PLL-based synchronization becomes fragile; oscillation modes appear between inverters; GFL plants increasingly require supplementary damping. Grid-forming capability moves from optional to strongly preferred.
• SCR below 2: very weak. GFL-only connection is widely treated as infeasible without extensive network reinforcement or supplementary plant. GFM is effectively mandatory.

The regulatory direction of travel is clear. Multiple jurisdictions have moved from encouraging to requiring grid-forming behaviour for new storage connections in weak areas, and the technical standards work has followed: the IEEE 2800 series on interconnecting IBR has become the reference framework, with grid-forming performance requirements added in recent revisions. In Australia, the connection process for weak-network projects in the NEM's renewable energy zones has front-loaded system-strength studies into the application stage — a change that materially alters project timelines for anyone developing in the north-west of the network.

The important engineering subtlety is that SCR is not a fixed property of the site. Adding synchronous condensers, reconfiguring the network, or connecting other GFM resources raises effective system strength and can move a site out of the weak category. This makes SCR a design variable in the optimization, not a fixed input — and it is the reason system-strength studies and battery control architecture should be co-optimized rather than sequentially specified.

Energy and Headroom Cost of Inertia Services

A BESS delivering inertia is a BESS not doing something else. Three resources are consumed simultaneously:

• Power headroom — the instantaneous MW reserved for inertial response. Reserved MW cannot be sold into energy markets as firm capacity.
• SOC headroom — inertia and fast frequency response are bidirectional, so the battery must remain inside an operating band that permits both charge and discharge. A common requirement is to hold SOC between 40% and 60% when inertia services are being offered, which directly reduces the usable energy for arbitrage.
• Cycle throughput — each inertia event consumes a small amount of energy but contributes to cycle count, and therefore to degradation through the year × C-rate × cycles-per-day relationship that governs LFP capacity fade.

Against those costs sit the available revenues. Inertia and fast frequency response markets have moved from demonstration to routine procurement in several systems, with prices that vary by an order of magnitude between markets and between scarcity events. Stability services are frequently procured on multi-year contracts rather than short-run auctions precisely because the operator needs certainty of response — which makes them attractive to financiers even when the headline price is modest, because they add contracted cashflow to a merchant-exposed asset.

The correct modelling approach is to co-optimize: dispatch the battery for energy arbitrage and ancillary services in the same optimization, subject to the SOC band, power headroom, and a minimum availability commitment for each service. Sequential optimization — sizing first for arbitrage, then adding services — systematically overstates the achievable combined revenue because it ignores the opportunity cost of reserved headroom.

Modelling Inertia Services in Energy Optima

Grid-forming storage projects need three things modelled together, and they are usually modelled apart: the dispatch economics, the battery's degradation trajectory under the contracted service duty cycle, and the technical feasibility of the connection at the relevant SCR.

In Energy Optima, a practical workflow:

1. Define the BESS in Battery-PCS → Operating Limits with the parameters that govern the inertia service: rated power, rated energy, minimum and maximum SOC for service availability, and the maximum power step the PCS can deliver within one grid cycle.
2. Set the service commitments in the EMS Configurator → SOC Thresholds and reserve the corresponding power headroom. The EMS will not dispatch reserved MW into energy markets, which correctly captures the opportunity cost in the optimization.
3. Run the dispatch with ECONOMIC_DISPATCH or MILP_HYBRID over an 8760-hour horizon with the ancillary services priced as additional revenue streams. The MILP formulation handles the headroom constraints and the simultaneous-service conflicts natively; rule-based dispatch is adequate for a first-pass sensitivity but will overstate availability double-counting.
4. Verify degradation with the battery SOH/RTE model. Inertia services add cycle throughput that looks trivial per event and meaningful per year. The platform's degradation model interpolates across 16,000+ SOH/RTE data points organized by year, C-rate, and cycles per day, so a duty cycle with many shallow partial cycles is evaluated on its real trajectory rather than an averaged C-rate.
5. Check the financial case in the 25-year projection with the stability service contracted at its actual tenor, layered over merchant or PPA energy revenue. Augmentation planning triggered on an SOH threshold will show whether the added cycling moves the first augmentation earlier.

The result that most often surprises engineers running this for the first time is that the GFM premium is not primarily a CAPEX premium. In many cases the incremental hardware cost over GFL is modest — a controls upgrade, sometimes additional PCS capability, occasionally a larger DC/DC stage. The dominant cost is the revenue foregone from reserved headroom, and the dominant value is the connection that would otherwise not be granted at all. That is a very different optimization problem from the one most storage models were built to solve, and it is the reason grid-forming projects need to be evaluated on system-strength-adjusted economics rather than raw levelized cost.

Dispatch strategy choice interacts strongly with all of this. A rule-based controller with a fixed SOC band cannot express an inertia availability commitment separately from an energy arbitrage band, so it will either over-deliver inertia (accumulating unnecessary cycle cost) or under-deliver (breaching the contracted availability and incurring penalties). An economic dispatch controller with explicit service reservations, solved over a full 8760-hour profile, prices the trade-off correctly and produces a defensible number for both the offtaker and the lender.