Industries / Energy, water and waste / Electric transmission and distribution / Microgrids
Design
Design is deciding what goes into a microgrid, how big each part should be and where its electrical boundary lies, before anything is bought.
7
Milestones achieved, from 1993 to 2024.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1993
A model for small systems
HOMER, first called the Hybrid Optimization Model for Electric Renewables, was written in 1993 for the National Renewable Energy Laboratory's Village Power Program. It simulates every hour of a year for each candidate mix of solar, wind, diesel and batteries and ranks the mixes by lifetime cost. NREL gave it away over the web, and by October 2004 it had more than 5,000 registered users in 162 countries, by NREL's count. Its original developers left NREL to run it as a company, HOMER Energy, which counted 160,000 users in 193 countries in June 2016; it is now sold by UL Solutions. The US Department of Energy calls HOMER a successful transfer of federal research to industry.
digital.library.unt.eduasiacleanenergyforum.adb.orgenergy.gov
Credit
Peter Lilienthal and NREL's Village Power Program; HOMER Energy
2000
What a site should own
Researchers at Lawrence Berkeley National Laboratory have built DER-CAM, the Distributed Energy Resources Customer Adoption Model, since 2000. It picks the mix, sizes, placement and hour-by-hour use of a building's or campus's own generators and batteries, counting savings from shifting load, trimming peak demand and selling services to the grid. The Department of Energy says the commercial design platform XENDEE grew out of DER-CAM research.
Credit
Lawrence Berkeley National Laboratory
2011
One definition for microgrids
At workshops in August 2011 and July 2012, the Department of Energy's Office of Electricity adopted the Microgrid Exchange Group's definition: "a group of interconnected loads and distributed energy resources within clearly defined electrical boundaries that acts as a single controllable entity with respect to the grid," which "can connect and disconnect from the grid" to run grid-connected or as an island. The same program set 2020 targets: systems under 10 MW that cut outage time for essential loads by more than 98%, at a cost comparable to a battery backup plus a diesel generator, while cutting emissions and raising efficiency by more than 20% each.
Credit
Microgrid Exchange Group; US DOE Office of Electricity, Smart Grid R&D Program
2017
A planning guide from IEC
The International Electrotechnical Commission published IEC TS 62898-1 on 18 May 2017, guidelines for planning and specifying alternating-current microgrids at low or medium voltage. It sorts microgrids into isolated ones, with no link to a wider grid, and non-isolated ones that run either grid-connected or in island mode, and covers load and generation forecasts, equipment planning, connection requirements and how to evaluate projects. An amendment followed in 2023.
Credit
IEC subcommittee SC 8B, Decentralized electrical energy systems
2017
Sizing on the web, free
REopt, built at NREL, is an optimization model that chooses the sizes and operation of solar, batteries, generators and other equipment that give a site the lowest lifetime cost, and can be told to carry a set share of the load through an outage of a set length. NREL says it used versions of the model at more than 10,000 sites from 2007 to 2017, leading to more than 260 MW of renewable power being built, and by September 2017 it offered REopt Lite, a free web version. By March 2020 the code was public; NREL then counted more than 1,100 web users and 325,000 calls to its programming interface. The lab, renamed the National Laboratory of the Rockies, keeps REopt.jl on GitHub under the Apache-2.0 licence.
Credit
Dylan Cutler, Kate Anderson and the REopt team at NREL
2019
IEEE's design practice
IEEE 2030.9-2019, approved on 21 March 2019 and published on 3 July 2019, sets out recommended practice for planning and designing alternating-current microgrids, grid-connected or standalone: system layout, electrical design, safety, power quality, metering and how to compare schemes.
Credit
IEEE Standards Committee 21 working group chaired by Wenpeng Luan
2024
Short tests, long promises
A June 2024 review for the California Energy Commission of seven microgrids it had funded found that those with a year of data cut energy costs by 20% to nearly 60% and demand charges by more than 40%. Most were sized for part of the facility's load, and their islanding tests "generally ended before the system would have failed in a true outage," so the run times promised in design were not shown.
Lessona design's promised hours on island are a claim until a full-length test or a real outage measures them.
Credit
Navigant Consulting (Laura Vogel, Karin Corfee, Peter Asmus and colleagues) for the California Energy Commission
Watch for
AI sizing in days
Xendee says its AI-powered software sizes a microgrid battery in a few days, a job it says "might have taken a year and $500,000" before.
Achieved whena customer or outside engineer publishes a real project in which Xendee's design time and cost are measured against a conventional study of the same site.
Claimed by
Michael Stadler, Xendee's chief technology officer, to Utility Dive, September 2025, when Eaton took a minority stake in the company
One model file, many tools
The Department of Energy says the planning tools DER-CAM, REopt and RONM have adopted MG-RAVENS, a shared data model for microgrids built on the Common Information Model (CIM), the IEC's standard way of describing grid equipment, and that a prototype workflow passed one design between tools "never designed to directly interface."
Achieved whena team outside the national laboratories moves one real site's model from one tool to another through MG-RAVENS without retyping it, and publishes the files.
Claimed by
US DOE Office of Electricity, draft strategy, March 2026
Open
01
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.
The briefCredit · unclaimed02
An agent that redoes studies
An AI agent that reads a published REopt study's description of a site, writes the REopt inputs itself, runs REopt.jl and reports the sizes it finds, scored against the sizes in the published study.
The briefCredit · unclaimed03
Site survey from photos
A phone workflow that turns photos of the nameplates on a site's existing generators, transfer switches, solar inverters and batteries into a filled-in input file for a design tool (kW, kWh, voltage, fuel, tank size), scored against a survey done by hand. Photos are taken by the site's own staff from outside guards and enclosures, never by opening equipment.
The briefCredit · unclaimed
Open 01 · the brief
Predicted against delivered
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
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.
Why it is still open
The California Energy Commission's 2024 review set islanding goals beside short tests for a handful of projects, and the Department of Energy's March 2026 strategy says planning tools still have to move to "field validated, industry-adopted platforms." Crossref searches in November 2026 for pooled predicted-against-measured microgrid comparisons found none.
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.
First step
List every CEC-500 microgrid final report since 2015 and mark which ones give both a design figure and a measured one for the same quantity.
Done looks like
30 microgrids, each with predicted and measured values side by side and linked, the median gap for each measure, and a second person rechecking 10 rows against the reports with no disagreements.
Open 02 · the brief
An agent that redoes studies
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An AI agent that reads a published REopt study's description of a site, writes the REopt inputs itself, runs REopt.jl and reports the sizes it finds, scored against the sizes in the published study.
Why it is still open
The Power Systems Agent Benchmark (June 2026) scores AI agents on microgrid dispatch, the hour-by-hour running of equipment already chosen, using generated cases, not on designing a microgrid from a written brief. Sandia researchers put a language model in the loop of an open capacity-planning tool, QuESt Planning, but had it adjust the tool's configuration "rather than direct asset siting or sizing decisions," and AgentiGrid (September 2026) has an agent run power flow studies on transmission grids. The laboratory publishes studies with exact answers to check against: its October 2026 study for a PG&E service center chose 176 kW of ground-mounted solar, 9 kW of rooftop solar, 26 kW of wind and a 103 kW, 994 kWh battery. Xendee's claim of AI design comes with no published scored test.
Built with what exists
REopt.jl (Apache-2.0) with an open solver; study reports from osti.gov; a tool-using AI model, for example an open-weights one such as Mistral Large 4 (see the Note "A trillion open parameters"), in an agent framework that can run Julia.
First step
Take five osti.gov REopt studies whose reports list their inputs, redo them by hand in REopt.jl to confirm they reproduce, then give the same reports to the agent.
Done looks like
20 published REopt studies redone by the agent, with its solar and battery sizes within 10% of the published ones in at least 15, and the full tally and agent logs published.
Open 03 · the brief
Site survey from photos
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone workflow that turns photos of the nameplates on a site's existing generators, transfer switches, solar inverters and batteries into a filled-in input file for a design tool (kW, kWh, voltage, fuel, tank size), scored against a survey done by hand. Photos are taken by the site's own staff from outside guards and enclosures, never by opening equipment.
Why it is still open
Field-service apps such as BuildOps' OpsAI already read heating and cooling nameplates into asset records, but no tool was found that writes a microgrid design tool's inputs from them, or that reports how often it misreads ratings.
Built with what exists
A phone camera; a vision-capable AI model; the REopt.jl input format or HOMER Pro's component library; a spreadsheet of true values for scoring.
First step
Collect 50 generator and inverter nameplate photos from makers' published manuals and spec sheets, record their true ratings, and measure one model's reading accuracy field by field.
Done looks like
200 nameplate photos from at least 20 sites read with accuracy reported for each field, and input files for 10 sites compared line by line with hand surveys.