BRENNEN

Industries / Energy, water and waste / Electric transmission and distribution

Microgrids

Microgrids are small local power systems, from a single building to a town or an island, with their own generators and storage, that can run connected to the main grid or on their own.

2002

In April 2002 a white paper from the Consortium for Electric Reliability Technology Solutions (CERTS), a group of US national laboratories, universities and companies led on this work by Robert Lasseter, proposed organising small generators, storage and loads into "MicroGrids" that would "seamlessly separate or island from the grid" when trouble came and reconnect once it was resolved, at a time when the draft US standard required small generators to shut down.

osti.gov

37

Milestones achieved across 6 domains.

10

To watch for: announced, not yet confirmed outside.

13

Open deliverables. Nobody has shipped them.

In one minute

Where it started

Islands and remote towns have long run their own small grids, usually on diesel. In 2002 the CERTS group of US laboratories proposed a different kind: a local grid that normally runs connected to the utility but can cut loose and carry itself when the main grid fails, then reconnect. A full-scale test bed in Ohio showed in 2008 that it worked.

Where it stands

The US Department of Energy counted 461 operating microgrids in the United States in 2021, with 3.1 GW of generation, and the World Bank's energy programme counted 21,500 mini-grids serving 48 million people worldwide in 2022. Microgrids carried two university campuses through Hurricane Sandy in 2012 and a North Carolina town through Hurricane Helene in 2024, and Kodiak, Alaska, gets 99% of its power from hydro and wind. Standards now cover planning, controllers and planned islands, but most performance figures still come from the companies and utilities that built the systems.

What is open

No one keeps a public record across sites of how microgrids perform in outages, what they cost, or how their measured results compare with what their designs promised. Open tools such as NREL's REopt.jl and MIT Lincoln Laboratory's Banshee test model, used with AI agents that can read reports and run simulations, put those records and new controller tests within reach of a small team.

Domains

AchievedWatch forOpen

Advances to build with

5 research notes

Open in microgrids

3 of 13 open deliverables

Open · Design

Predicted against delivered

A public table that pairs, for at least 30 built microgrids, what the design study predicted (bill savings, renewable share, hours the essential load can run on island) with what was later measured, each figure linked to its report.

Built with what exists

The Energy Commission's EPIC final reports (the CEC-500 series), the National Laboratory of the Rockies' REopt study reports on osti.gov, utility filings, pdftotext, and a document-reading model such as Meta's Muse Ember (see the Note "The jobs agents repeat") to pull figures, each checked by a person; Python with pandas.

Credit · unclaimed

Open · Generation and storage

Planning defaults against meters

For the engine-driven CHP systems in NYSERDA's public performance database, a table of measured electric efficiency at each level of load, hours spent shut down, and the lowest load each unit actually ran at, set beside the defaults that NREL's REopt planning model assumes for the same kind and size of engine: full-load electric efficiency, minimum turndown (the lowest share of full power it may run at) and unavailable periods.

Built with what exists

CSV downloads from the NYSERDA DER Integrated Data System; REopt.jl's chp_defaults.json; Python with pandas. Data only; nobody visits or touches a plant.

Credit · unclaimed

Open · Controls

Agent-written island transitions

Give an AI coding agent the Banshee microgrid model and Lincoln Laboratory's 2017 test sequence (grid lost, black start, resynchronise, disconnect, hold a set power flow across the connection point, ride through faults). Have it write the controller logic for each transition in readable form, run it in simulation, read the results and revise until it passes, then compare it with a controller written by a person.

Built with what exists

The Banshee model files in the EPHCC repository on GitHub (the full model for MATLAB/Simulink and RTDS, Feeder 1 for Typhoon HIL); Typhoon HIL Control Center with its free Virtual HIL emulator; an AI coding agent such as Claude Fable 5.2 (see the Note "Three working devices") or the open-weight Mistral Large 4. The Note "Readable robot programs" describes the same write, simulate and revise loop for robot arms. Simulation only; no real switchgear.

Credit · unclaimed

Where it sits