AIBRIEF
A research pipeline scores 66,935 US substations on a 30–60 minute horizon, with reported detection near 80% at the most operationally relevant thresholds.
During the May 2024 geomagnetic storm, the first G5 in two decades to push auroras to subtropical latitudes, North American grid operators had continent-scale warnings and limited time to act. A new Microsoft Research pipeline reframes the problem at substation resolution: it scores 66,935 continental U.S. substations on a 30–60 minute horizon.
The pipeline starts with L1 solar-wind measurements feeding AE (Auroral Electrojet) and Dst (Disturbance Storm Time) indices, then fuses them with local ground conductivity, geology, and grid topology using gradient boosting. The output is dB/dt, the rate of geomagnetic-field change that drives geomagnetically induced current (GIC) into transformers.
For 2020–2026 evaluation, the source reports detection rates of 76.5% at ≥10 nT/min, 81.2% at ≥20 nT/min, and 64.1% at ≥50 nT/min, with false-alarm rates rising alongside severity. Simple linear regression served as the risk-stage comparison. Inference ran in roughly 333 milliseconds across the full modeled fleet.
The blog labels its continental map a representative scenario, not a deployed product, and explicitly calls for utility and operational validation before any grid use. That changes the unit of warning: a 30–60 minute, asset-level list would let a control room pre-position crews, shift load, or take sensitive transformers offline in the highest-risk substations before the next G5.