Frequency regulation

What it is (precise)

Grid frequency reflects the instantaneous balance of supply and demand. When load briefly exceeds generation, frequency sags; when generation leads, it rises. Frequency regulation closes that gap continuously.

The system operator broadcasts a regulation signal — in many markets a normalized command updated every 2 to 4 seconds from the automatic generation control (AGC) system — and each enrolled resource follows it proportionally to its enrolled capacity, injecting power (discharge) or absorbing it (charge) to nudge frequency back toward target. In the classic control hierarchy this is secondary frequency control; Europe's FCR sits at the primary layer but is likewise procured as a market product.

For a stationary BESS this is executed at the plant controller, which translates the dispatch command into per-PCS setpoints across the racks and containers and meters performance at the point of interconnection (POI).

It differs from primary frequency response or fast frequency response, which are autonomous, droop-based reactions to a locally measured frequency deviation rather than the following of a centrally issued signal. Many projects run both: an autonomous droop response for large events and market regulation for normal minute-to-minute operation, with the plant controller arbitrating between them so neither overwrites the other.

Why it matters in a real grid-scale project

Frequency regulation rewards what a battery does well — fast, accurate, bidirectional power — so it has often delivered strong dollar-per-MW returns, which is why early merchant storage was frequently underwritten on regulation revenue: PJM's RegD fleet in the mid-2010s, the GB Enhanced Frequency Response tender of 2016 (around 200 MW procured), and the 100 MW / 129 MWh Hornsdale plant earning outsized Australian FCAS revenue from 2017.

The energy throughput per cycle is shallow because the signal averages near zero, so the binding design drivers are power rating, round-trip efficiency, and controller accuracy rather than deep energy capacity.

The commercial catch is saturation and performance scoring. Regulation markets are shallow — typically a few hundred MW to roughly 1 GW of requirement per ISO (~1% of peak load) — and clear quickly once batteries enter, so prices can fall sharply; PJM regulation, GB dynamic services, and Australian FCAS have all softened materially as BESS capacity arrived.

Many markets pay a mileage or performance component that penalizes slow or inaccurate response, so plant-controller accuracy and availability feed straight into settlement. For project finance this makes regulation income real but volatile, pushing most modern projects toward revenue stacking across regulation, energy arbitrage, and a capacity market position — the core of the merchant vs. contracted decision.

Typical values and standards

Target frequency is 60 Hz in North America and 50 Hz in most of Europe and Asia, held in normal operation within tens of millihertz. Regulation dispatch typically updates every 2 to 4 seconds (PJM RegD ~2 s; CAISO/ERCOT AGC ~4 s); assets are commonly sized at a 15-minute to 1-hour energy-to-power ratio, set by market rules, not chemistry.

Continental European FCR requires full symmetric activation within 30 seconds at ±200 mHz, procured in 4-hour blocks with a 1 MW minimum; aFRR full activation is on the order of 5 minutes. GB Dynamic Containment demands sub-second response (initiate within 0.5 s, full delivery within 1 s) with full output at ±0.5 Hz; the slower Dynamic Moderation and Dynamic Regulation products saturate near ±0.2 Hz.

The rulebook roles matter. In the US, FERC Order 755 (2011) mandated pay-for-performance regulation compensation — the mileage and accuracy payments that made fast storage economic — and Order 841 (2018) opened wholesale markets to storage; NERC BAL standards define the control-performance metrics — chiefly CPS1, a statistic of how tightly a balancing authority keeps frequency error near zero — that set how much regulation each one procures.

ERCOT's fast frequency response product is a useful contrast: full response within ~15 cycles once frequency hits 59.85 Hz, sustained 15 minutes — a triggered event service, not signal-following. The safety envelope is unchanged: UL 9540 certification, UL 9540A fire-propagation test data, and NFPA 855 installation requirements apply regardless of service.

How it shows up in specs, studies and contracts

On the equipment side, look for the response chain: a modern PCS can move to a new active-power setpoint in tens of milliseconds, but what the market sees is plant-level response at the POI including plant-controller latency, telemetry scan time, and communication hops — typically a few seconds end-to-end, so verify the budget is comfortably inside the ISO / RTO qualification test for the specific product.

Enrollment normally requires passing an operator-issued test signal and then maintaining a performance score; in PJM, resources falling below a composite score threshold on the order of 0.75 face disqualification. Interconnection studies should confirm the plant can sustain the full regulation range at the POI net of transformer, PCS, and auxiliary losses.

On the contract side, regulation quietly consumes battery warranty. A plant following a fast regulation signal accumulates hundreds of shallow equivalent full cycles per year, so check whether the supply agreement counts warranty in energy throughput, cycle count, or both, and how shallow cycles are weighted.

Ask four questions early: what is the measured signal-to-POI latency; how is state of charge managed and who pays for the makeup energy when the signal is not perfectly neutral; how does regulation duty interact with the augmentation plan; and what auxiliary consumption results from continuous cycling of the thermal management system. These answers move real money at settlement and at year-ten capacity tests.

Common pitfalls

Do not conflate regulation with inertia or primary response. Regulation is signal-following on a seconds-to-minutes timescale; inertial response and droop-based primary response act in the first milliseconds to seconds after an event, and a plant paid for regulation is not automatically providing them. Likewise, an energy-neutral signal is only approximately neutral: round-trip losses mean roughly 10 percent of throughput is consumed at a 90 percent round-trip efficiency, and most markets settle that net energy at prevailing prices — a cost line that disappears from naive revenue models.

For the financial model, the consequence of the saturation described above is simple: never extrapolate today's clearing prices across a 20-year horizon; treat regulation as a high-value but decaying early revenue layer inside a broader stack with energy arbitrage, not as the anchor. Operationally, watch for state-of-charge conflicts between a regulation obligation and a day-ahead energy position: the plant controller must reserve headroom in both directions, and a scheduling desk that ignores this will fail performance scoring exactly when prices are highest.