How the Grid Works, and Why It Falls Over

Nothing is stored behind your wall socket. The electricity you are using was made in the last fraction of a second, by a machine spinning in step with every other machine on the continent, and it will be gone the instant you stop drawing it. The grid is not a reservoir. It is one rotating engine, thousands of miles across, held in balance every second of every day.

That design is why the lights are on almost all the time. It is also why, when it does go wrong, it can go wrong across eight states in under seven minutes.

Every generator on the grid turns in step

Alternating current does not flow one way. It reverses direction over and over, and the number of complete reversals per second is the grid's frequency. Every large generator on a synchronous grid turns at a speed locked to that frequency, so they are all effectively bolted to the same crankshaft.

Frequency is therefore the grid's tachometer, and it tells you instantly whether supply and demand match. Draw more than is being generated and the spinning machines are dragged down, so frequency falls. Generate more than is drawn and they speed up, so it rises. No meter reading is involved; the physics reports the imbalance directly.

Synchronous area	Normal band	Worst momentary dip allowed	Back inside the band within
Continental Europe	±50 mHz	800 mHz	15 minutes
Great Britain	±200 mHz	800 mHz	15 minutes
Ireland & Northern Ireland	±200 mHz	1,000 mHz	15 minutes
Nordic	±100 mHz	1,000 mHz	15 minutes

Annex III, Commission Regulation (EU) 2017/1485, which also sizes Continental Europe's reserves against a 3,000 MW reference incident.

Inertia is the flywheel that buys the first seconds

When a big power station drops off, nothing human reacts in time. What reacts is mass. Every turbine and generator rotor already spinning is a flywheel storing kinetic energy, and the moment demand exceeds supply that energy is dragged out of them. They slow down, which is the same thing as frequency falling, and how fast they slow depends on how much spinning mass is connected.

Engineers call that the rate of change of frequency. Heavy thermal and hydro plant gives a gentle slope. A grid leaning on inverter-connected wind, solar and imports carries far less rotating mass, so the same loss makes frequency fall much faster.

Britain has a documented case. On 9 August 2019 a lightning strike on a transmission circuit was cleared normally, but within a second of each other the Hornsea offshore wind farm dropped 737 MW and a steam turbine at Little Barford dropped 244 MW. With about 150 MW of small generation that went with them, 1,131 MW vanished at once.

National Grid ESO's technical report found that parts of the system then saw a rate of change of frequency of 0.125 Hz per second or more. That is the trip setting on protection relays fitted to small embedded generators nationwide, and an estimated 350 MW of that generation duly took itself off, deepening the hole.

Four layers of defense, in the order they act

The grid has four safety nets, and they take over from one another as the seconds pass.

Inertia acts instantly and for free, because it is just physics. Frequency containment reserve is generation held back on purpose that ramps up as frequency droops; in Continental Europe half must arrive within 15 seconds of a 200 mHz deviation and all of it within 30 seconds.

Frequency restoration reserve then replaces it and pulls frequency back inside the band within 15 minutes. Load shedding is the fourth layer, and it works by taking customers off.

Transmission and distribution do completely different jobs

People say "the grid" as if it were one thing. It is two things wearing the same coat, and the difference decides how long your outage lasts.

The lesson in those two numbers is the one most homeowners never hear. Nearly every time you lose power, the fault is a few streets away on the distribution network, and a crew fixes it in hours. The rare event that runs for days is a transmission or generation failure.

Push the domino

Transmission networks are built to a rule called N-1: whatever single component fails, the system must survive on what is left. It works, right up to the moment two things go at once, or one goes while something hidden was already broken.

The model below is a region leaning on four tie lines for part of its power. Trip what you like and watch what the survivors carry.

Take one line out and the other three go from 150 MW to 200 MW each. That is exactly their rating, and exactly the point: the system was built to survive it. Take a second and the remaining pair is asked for 300 MW apiece, protection opens them, and the region becomes an island that cannot feed itself.

This is not a thought experiment. On 14 August 2003 three 345 kV lines in northern Ohio sagged into untrimmed trees between 15:05 and 15:41, while the control room's alarm software had been dead since 14:14.

The Sammis–Star line tripped at 16:05:57 on a relay reading overload, and in the four and a half minutes that followed, thirteen more 345 kV and 138 kV lines opened on the same kind of relay, northern Ohio separated from Pennsylvania, and 61,800 MW of load serving an estimated 50 million people went dark across eight states and Ontario.

Figures from the US–Canada Power System Outage Task Force's final report.

Load shedding is the grid saving itself by dropping you

When frequency keeps falling despite every reserve, the last defense is automatic and merciless. Relays across the distribution network watch frequency and disconnect whole feeders in pre-arranged blocks, on their own, in a fraction of a second. Nobody phones anybody. The point is to shed load faster than the shortfall can drag generators down to the speed where they trip themselves and everything goes black.

It works. When frequency in Britain fell through 48.8 Hz at 16:53:49 on 9 August 2019, the first block fired and disconnected 931 MW. The block is meant to be 5% of national demand; the relays actually reached 3.2%, and it was enough. Frequency stopped falling and was back at 50 Hz by 16:57:15.

The network operators counted 1,152,878 customers off, most of them back inside 45 minutes.

If you were one of the 1.1 million, your street was not unlucky. It was chosen in advance, by a relay setting, as the price of everyone else staying up.

Black start is the part that takes days

Load shedding buys the grid its life. When it fails, restarting is a different order of problem, because almost no power station can start without power. Boiler feed pumps, control systems, cooling and gas compressors all need electricity before they can make any.

So restoration begins with a few designated black start units — usually hydro, or gas turbines with their own diesel starting sets — that come up unaided. They energize a path to a larger station, which starts and energizes more path. Each step must match load to generation closely enough that the fragile island does not trip again, so customers return in blocks.

In 2003 that took four days in parts of the United States, and parts of Ontario ran rolling blackouts for over a week.

Texas in February 2021 shows the near miss: ERCOT's frequency bottomed at 59.302 Hz and sat below 59.4 Hz for four minutes and 23 seconds against a nine-minute limit, which the University of Texas review describes as leaving the grid within minutes of a complete blackout.

Load shed orders reached 20,000 MW. Had it gone black, restoration would have been measured in days, in a deep freeze.

Minutes, hours, or days: what this means for your house

It all collapses into one question: which failure are you preparing for, and how long does it last?

What failed	How often	Typical duration	What carries you through
Local distribution fault: tree limb, blown fuse, animal, car into a pole	Roughly once or twice a year for the average customer	Minutes to a few hours	A battery or power station on a few circuits; you barely notice
Automatic under-frequency load shedding	Rare, and never announced	15 minutes to a few hours	Anything with instant transfer; the grid comes back on its own
Rolling blackouts during a supply shortfall	Regional, in extreme heat or cold	Hours, repeated over days	Storage that recharges between rotations, or a generator
Storm damage to the distribution network	The main driver of the long tail	1 to 7 days	Fuel or sun; a battery alone will not span it
Cascading failure and black start	Once in decades, per region	Several days, restored in blocks	Multi-day autonomy, water, heat, and a plan

The common mistake is sizing for the common failure and then meeting the rare one. A single battery covers your fridge through a Tuesday feeder fault and runs flat halfway through a February ice storm. A generator with no fuel plan is a heavy ornament by day three.

Work it the other way round. Decide what must stay running, add up what it draws in a day, then ask how many days you want that to last. Our backup power calculator does that arithmetic against your own appliance list and gives the battery or generator size it implies.

Then read the risk where you live: outage risk by state for the United States, or the European outage and blackout atlas, which show how fat the multi-day tail is in your region.

And if you have never thought past the first few hours, what actually breaks in a week without the grid is the sobering version of this page.

Common questions

Sources: US–Canada Power System Outage Task Force, Final Report on the August 14, 2003 Blackout (April 2004); Commission Regulation (EU) 2017/1485, Annex III; Commission Regulation (EU) 2016/1388, Article 19; University of Texas at Austin Energy Institute, Timeline and Events of the February 2021 Texas Electric Grid Blackouts (July 2021); National Grid ESO, Technical Report on the events of 9 August 2019, and Ofgem's investigation report (January 2020); US Energy Information Administration, Electric Power Annual table 11.1 (2024 data).