Industries / Energy, water and waste / Electric transmission and distribution / Microgrids
Controls
The controller is the software and hardware that decides, second by second, which generators and batteries run, how much each produces, which loads to drop, and when the microgrid leaves or rejoins the main grid.
5
Milestones achieved, from 2002 to 2017.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2002
The CERTS microgrid concept
A white paper from the Consortium for Electric Reliability Technology Solutions (CERTS), a group of US national laboratories, universities and companies, proposed running small generators and their loads as one system that looks to the utility like a single customer. Each inverter (the electronics that connect a source to the grid) would react within milliseconds to the voltage and current it measured itself, using droop: letting frequency and voltage sag slightly as load rises, so that units share load in proportion to their size without a fast communication link. A new unit could then be added without re-engineering the others, which the paper called "plug and play".
Credit
Robert Lasseter, Abbas Akhil, Chris Marnay, John Stephens, Jeff Dagle, Ross Guttromson, A. Sakis Meliopoulos, Robert Yinger and Joe Eto; Lawrence Berkeley National Laboratory report LBNL-50829, April 2002
2008
The CERTS test bed passes
At a full-scale test bed near Columbus, Ohio, run by the utility American Electric Power, three Tecogen combined heat and power units fitted with CERTS controls and a static switch built by Northern Power Systems went through five sets of tests. The final report for the California Energy Commission found that islanding and resynchronising met all IEEE 1547 and power quality requirements, that the protection could tell normal from faulted operation without relying on high fault currents, and that the controls stayed stable under all conditions, including hard motor starts.
Credit
Joseph Eto (LBNL), Robert Lasseter (University of Wisconsin), Ben Schenkman and John Stevens (Sandia National Laboratories), Harry Volkommer and Dave Klapp (American Electric Power), Northern Power Systems, Tecogen
2013
Fort Carson's 74-hour island
SPIDERS (Smart Power Infrastructure Demonstration for Energy Reliability and Security) was a US Defense Department demonstration of cyber-secure microgrids on military bases. In its second phase, at Fort Carson, Colorado, built by Burns and McDonnell, the microgrid ran diesel generators, 500 kW of solar panels and electric vehicles cut off from the utility for 74 hours, from 21 to 24 October 2013, with no interruption to essential loads. Pacific Northwest National Laboratory's assessment estimated that the solar and better generator loading would make the fuel last 19.3% longer. Four things went wrong: half the solar array was down, a generator was derated, the synchronising breaker would not operate remotely, and the system dropped out of microgrid mode during generator maintenance; the vehicles never charged on the microgrid.
Lessona demonstration can pass its metrics and still expose the faults that matter in a real outage, so the unplanned events belong in the record.
Credit
US Department of Defense (SPIDERS programme); Burns and McDonnell; PNNL (Jonathan Barr, Francis Tuffner, Mark Hadley, Kevin Schneider)
2017
Four controllers, one test
MIT Lincoln Laboratory, paid by the US Department of Energy, built a controller hardware-in-the-loop platform: real controller hardware wired to a real-time simulation of a microgrid, here a model code-named "Banshee" of a small industrial site with a 4 MVA diesel generator, combined heat and power, 3 MW of solar and a 2.5 MW battery. At a symposium at MIT in February 2017, four makers' commercial microgrid controllers ran the same scripted sequence: grid lost at two minutes, black start (restarting with no outside power) at three, then resynchronising, disconnecting and riding through faults. The platform and the Banshee model were released as open source, and the authors report Banshee in use as a reference by IEEE P2030.8, NREL and universities.
Lessontest the edges of the synchronisation window, especially large phase-angle gaps, because a bad reconnection can damage equipment.
Credit
Erik Limpaecher, Raul Salcedo and Eric Corbett (MIT Lincoln Laboratory); Scott Manson, Bharat Nayak and Will Allen (Schweitzer Engineering Laboratories)
2017
A standard for controllers
IEEE 2030.7, approved on 6 December 2017 and published on 23 April 2018, set out the functions every microgrid controller must have above the level of individual machines: managing the microgrid as one system, running it alone or grid-connected, and connecting and disconnecting smoothly. A companion standard, IEEE 2030.8, covers how such controllers are tested.
Credit
IEEE Standards Association, Microgrid Controllers Task Force
Watch for
AI steadies a microgrid's voltage
Sandia National Laboratories says its AI-assisted distributed energy resource management system, software that coordinates inverters, batteries and solar, moved from simulation to power hardware-in-the-loop tests and then to the Texas Tech University GLEAMM microgrid in Lubbock, which includes a data centre. Sandia compared one full day with the controller on to one day with it off, and says voltage, which there typically runs about 5% above normal, moved closer to its target.
Achieved whensomeone outside Sandia publishes voltage records from GLEAMM or another site over weeks with the controller on and off under comparable load and weather, showing the gain.
Claimed by
Sandia National Laboratories (Rachid Darbali-Zamora), June 2026
Open
01
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.
The briefCredit · unclaimed02
Audit open energy controllers
Run AI code-analysis tools over open-source software that runs batteries and small grids, such as OpenEMS (an open energy management system used with commercial home and business batteries) and Oak Ridge National Laboratory's CSEISMIC microgrid controller. Confirm each finding by hand, report it privately to the maintainers through GitHub's advisory process, and publish the tally only after fixes ship.
The briefCredit · unclaimed
Open 01 · the brief
Agent-written island transitions
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
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.
Why it is still open
The model and the test sequence have been public since 2017, but no AI-written controller has been published against them. The closest published test of AI agents on power engineering, the Power Systems Agent Benchmark (June 2026), scores 41 task families, and its microgrid tasks are dispatch problems (how much each source should produce), not timed switching between grid and island.
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.
First step
Download the Banshee files from EPHCC, open Feeder 1 in Typhoon's Virtual HIL with a free key, and script the grid-loss step at two minutes.
Done looks like
20 runs of the full sequence from fresh starts, with a published tally of runs in which no feeder blacked out, no generator tripped and every reconnection closed inside the synchronisation limits, beside a human-written controller's tally on the same 20 runs.
Open 02 · the brief
Audit open energy controllers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Run AI code-analysis tools over open-source software that runs batteries and small grids, such as OpenEMS (an open energy management system used with commercial home and business batteries) and Oak Ridge National Laboratory's CSEISMIC microgrid controller. Confirm each finding by hand, report it privately to the maintainers through GitHub's advisory process, and publish the tally only after fixes ship.
Why it is still open
OpenEMS lists one security advisory, a critical query-injection flaw published on 1 October 2026 and credited to its own developers; no outside, systematic audit of OpenEMS or of open microgrid controller code was found. The approach has worked elsewhere: the security firm AISLE says its AI analyser found 38 vulnerabilities in OpenEMR, an open-source health records system, in 2026.
Built with what exists
GitHub CodeQL and Semgrep, both free for public code; an AI model used for code review, such as Claude Fable 5.2; the projects' own security policies (OpenEMS publishes a SECURITY.md). The Note "Cyber defenders first" covers a model released first to defenders for this kind of work. Work only on code and on copies you run yourself, never on anyone's installed system.
First step
Read OpenEMS's SECURITY.md, then run CodeQL on the OpenEMS backend and triage the first 50 results by hand.
Done looks like
A published table of every finding with its severity, the maintainers' verdict (confirmed or false alarm) and the days to a fix, covering at least two projects, with no exploit code released.