Frequency control and system balancing

Electrical energy is produced at the instant it is consumed, so generation must match load continuously. Frequency is the indicator of that balance across a whole interconnection: when generation exceeds load, the surplus accelerates every synchronous machine and frequency rises; when load exceeds generation, frequency falls. Keeping frequency within a narrow band, typically within ±0.05 Hz of 60 Hz in normal operation, requires a layered set of controls acting from fractions of a second to hours. This lesson follows a sudden loss of generation through those layers: inertial response, primary control by governors, secondary control by automatic generation control between balancing areas, and tertiary control and reserves, and then examines the challenge of falling inertia as inverter-based resources replace synchronous machines, and the role of underfrequency load shedding as the last line of defense.

Inertial response

In the first moments after a generator trips, nothing has yet acted to replace its output. The power deficit is supplied by the kinetic energy stored in the rotating masses of all synchronous machines (and synchronous motors), which slow down together. Summing the swing equations of Lesson 8 over the whole system gives the initial rate of change of frequency (RoCoF):

where \(\Delta P\) is the lost generation, \(S_{\text{sys}}\) the total rating of online synchronous machines and \(H_{\text{sys}}\) their combined inertia constant. RoCoF matters because every subsequent control needs time to act; the faster frequency falls, the deeper it goes before those controls arrest it.

Primary control: governors and droop

Within a few seconds, turbine governors respond to the falling speed by opening valves or gates. Each governor has a droop characteristic: its output rises in proportion to the frequency drop. A droop \(R\) of 5% means that a 5% fall in frequency (3 Hz at 60 Hz) would drive the unit from zero to full output:

Droop lets many governors share a change in load stably, in proportion to their ratings, without fighting each other, which an isochronous (zero-droop) controller on every unit would do. Typical droops are 4% to 5%, with intentional deadbands of about ±0.036 Hz so that governors do not respond to normal small fluctuations.

Load also helps: motors slow down and draw less power when frequency falls, an effect called load damping, typically 1% to 2% of load per 1% change in frequency. The combined steady-state response of governors and load defines the frequency response characteristic \(\beta\) of the interconnection, in MW per 0.1 Hz:

The lowest point of the frequency excursion, the nadir, occurs a few seconds after the event, when governor response catches up with the deficit. It is deeper than the settling frequency, because turbines respond with lags of several seconds (steam reheat units are slow, gas turbines and hydro units faster but with their own dynamics). The nadir must stay above the first stage of underfrequency load shedding with margin.

Secondary control: automatic generation control

Primary control arrests the fall but leaves frequency below 60 Hz, and the replacement power comes from governors all over the interconnection, so flows between regions deviate from their schedules. Secondary control restores both. The interconnection is divided into balancing authority areas, each responsible for matching its own generation to its own load plus scheduled interchange. Each area computes its area control error (ACE) every few seconds:

where \(NI_A\) and \(NI_S\) are actual and scheduled net interchange (exports positive), \(F_A\) and \(F_S\) actual and scheduled frequency, and \(B\) the area's frequency bias in MW/0.1 Hz (negative by convention), set close to the area's own frequency response. Automatic generation control (AGC) adjusts the set points of participating units to drive ACE toward zero over several minutes. The bias term is the key: for an event outside the area, the area's governor response increases its exports, but the frequency term cancels that, so its ACE stays near zero and AGC leaves its governors' help in place; for an event inside the area, both terms point the same way and AGC raises generation to replace the loss.

Performance standards (NERC BAL-001 in North America) require each area to keep ACE small on average and to recover from large disturbances within 15 minutes. Small systems may implement AGC with a few hydro or gas units; large ones dispatch hundreds of units, and modern markets combine AGC with five-minute economic dispatch (Lesson 7) so that regulation is provided at least cost.

Tertiary control and reserves

Tertiary control restores the reserves used by primary and secondary control, by redispatching or starting units within 10 to 30 minutes, ready for the next event. Operators carry several categories of reserve:

The amount of contingency reserve is set by the largest credible loss, often the largest unit or the largest import line. With high shares of wind and solar, uncertainty in their output adds to reserve needs, and the steep evening ramp in net load (load minus solar output, the so-called duck curve) requires flexible capacity able to increase output by thousands of megawatts in a few hours.

Simulating a frequency event

import numpy as np

f0 = 60.0
def event(H, loss=0.02, R=0.05, D=1.0, Tg=1.5, share=0.4, agc=False, ffr=0.0, t_end=300.0, dt=0.01):
    """Per unit on system base: lose a fraction loss of generation at t = 0. Returns RoCoF, nadir, settling frequency."""
    df, Pg, Pffr, integ = 0.0, 0.0, 0.0, 0.0
    nadir, rocof = 0.0, None
    for step in range(int(t_end / dt)):
        Pload_change = D * df                           # load falls as frequency falls (D in pu MW per pu Hz)
        acc = (Pg + Pffr - loss - Pload_change) / (2 * H)   # d(df)/dt in pu Hz per second
        if rocof is None: rocof = acc * f0
        target = -share * df / R - (0.5 * integ if agc else 0.0)   # droop on the responsive share of capacity, plus AGC
        Pg += (target - Pg) * dt / Tg                   # turbine-governor lag
        Pffr = ffr if df * f0 < -0.15 else Pffr          # fast frequency response triggers below 59.85 Hz
        integ += df * dt
        df += acc * dt
        nadir = min(nadir, df)
    return rocof, f0 * (1 + nadir), f0 * (1 + df)

for label, kw in (("H = 5 s", dict(H=5.0)), ("H = 2.5 s", dict(H=2.5)), ("H = 2.5 s + 1% fast response", dict(H=2.5, ffr=0.01)),
                  ("H = 5 s with AGC", dict(H=5.0, agc=True))):
    rocof, nad, fin = event(**kw)
    print(f"{label:30s}: initial RoCoF {rocof:+.3f} Hz/s, nadir {nad:.3f} Hz, after 300 s {fin:.3f} Hz")
beta = 0.4 / 0.05 + 1.0
print(f"steady-state deviation without AGC: {-0.02 / beta * f0:.3f} Hz (from the frequency response characteristic)")

H = 5 s                       : initial RoCoF -0.120 Hz/s, nadir 59.819 Hz, after 300 s 59.867 Hz
H = 2.5 s                     : initial RoCoF -0.240 Hz/s, nadir 59.770 Hz, after 300 s 59.867 Hz
H = 2.5 s + 1% fast response  : initial RoCoF -0.240 Hz/s, nadir 59.843 Hz, after 300 s 59.933 Hz
H = 5 s with AGC              : initial RoCoF -0.120 Hz/s, nadir 59.824 Hz, after 300 s 60.000 Hz
steady-state deviation without AGC: -0.133 Hz (from the frequency response characteristic)

Halving inertia doubles the initial RoCoF and deepens the nadir, while leaving the settling frequency unchanged, because that depends only on droop and load damping. Fast frequency response from batteries, triggered when frequency falls 0.15 Hz, more than makes up for the lost inertia. AGC returns frequency to 60 Hz over a few minutes.

Low inertia and fast frequency response

As inverter-based wind, solar and batteries displace synchronous plants, system inertia falls, especially at times of low load and high renewable output. Operators in systems such as ERCOT, Great Britain, Ireland and Australia monitor inertia in real time and set minimum inertia levels, constraining how much synchronous generation can be turned off. Responses include:

• Fast frequency response: batteries and other inverter-based resources inject power within a few hundred milliseconds of detecting a frequency drop, much faster than any governor. ERCOT's market includes a fast frequency response product, and Great Britain's dynamic containment service procures similar response.
• Synthetic inertia: inverters, including wind turbines drawing on rotor kinetic energy, respond in proportion to RoCoF.
• Grid-forming inverters: behaving as voltage sources, they respond inherently to frequency changes like synchronous machines.
• Synchronous condensers, some with flywheels, add real inertia (Lesson 3).
• Raising RoCoF withstand capabilities of generators and their protection, so that units do not trip during fast transients.
• Demand response: interruptible industrial loads and, increasingly, data centers that can reduce consumption quickly on frequency signals.

Underfrequency load shedding

If a disturbance exceeds the reserves, for example when an interconnection separates into islands with a large generation deficit, frequency continues to fall. Below about 57 to 58.5 Hz, steam turbine blades risk damaging resonance and generator protection begins to trip units, which would accelerate the collapse. Underfrequency load shedding (UFLS) relays automatically disconnect blocks of load in stages, for example 5% to 10% of load at each of several thresholds starting near 59.3 Hz in many North American regions, to restore balance. The stages are coordinated with generator underfrequency protection, so that load is shed before generators trip, and they are designed for credible islanding scenarios with deficits of up to 25% to 30%.

Restoring the shed load must be done gradually as generation is brought up, coordinating with the system operator. Electric clocks once kept time by counting cycles, and accumulated time error is still corrected by small offsets in scheduled frequency, though its practical importance has diminished.