Dominion’s grid archive becomes a foundation for new AI work

PingThings and Dominion Energy Virginia announced a five-year alliance on 28 October, expanding their PredictiveGrid work. The programme aims to apply AI to seven years of full-resolution measurements collected since 2019.[1]

PingThings says it holds more than five petabytes of Dominion sensor data. The archive includes the transmission system serving Northern Virginia’s Data Center Alley, where electrical behaviour must be understood across very different timescales.[1]

The disturbance and its lead-up

A 2025 study by Dominion and Rensselaer Polytechnic Institute researchers illustrates why that history matters. They investigated oscillations around 14.7 hertz from a data-centre region, including one event that lasted two hours. Examining several months of measurements revealed earlier dynamics around 10–11 hertz that lost stability before the larger oscillations.[2]

The researchers linked the instability to an interaction between grid conditions and the data centre’s uninterruptible power supplies; the owner confirmed the equipment involved. This was a measurement-based engineering investigation. Its useful evidence extended well beyond the period when the disturbance was most obvious.[2]

Keeping detail accessible

PingThings identifies BTrDB as its storage engine. The original 2016 database paper describes a tree with individual readings at its leaves and statistical summaries above them. A query covering a long period can use those summaries; zooming into an incident retrieves finer detail while retaining the original measurements.[3][4]

The current platform also records context: sensor identity, units, sampling rate and the associated substation or asset. Analysts can work in JupyterLab with Python libraries, then schedule an analysis as a continuing service. Model outputs can be stored alongside the original signals.[3]

These capabilities connect three views of the same measurement: its place in a long trend, its behaviour during a short incident and the equipment that produced it. The archive can support both a retrospective investigation and an analysis that keeps running as new readings arrive.[3]

The partners plan technical publications and case studies, with Dominion acting as a reference utility so others can learn from the work.[1]

Report an error or suggest a correction