Researchers at Microsoft Research have developed a machine learning system designed to forecast space-weather risks across 66,935 substations in the continental United States. The system provides location-specific risk estimates 30 to 60 minutes ahead of potential impacts, helping grid operators prepare for geomagnetic storms.
Microsoft Research Develops ML System to Forecast Space-Weather Risks for US Power Grid Substations
The forecasting pipeline operates in three stages. First, it uses solar-wind measurements from the L1 Lagrange point to forecast Auroral Electrojet (AE) and Disturbance Storm Time (Dst) indices. Second, a gradient-boosting model combines these forecasts with local geological conductivity and grid-infrastructure data to estimate the rate of magnetic-field change (dB/dt) associated with geomagnetically induced current (GIC) risk. Finally, these predictions are aggregated into location-specific risk estimates for each substation.
In evaluations conducted for the 2020–2026 period, the model outperformed empirical approaches. The GIC risk stage achieved detection rates of 76.5% for major events (≥10 nT/min) and 81.2% for severe events (≥20 nT/min). The system's ability to incorporate local geological factors allows it to distinguish between low-risk areas and those where resistive bedrock may amplify ground-level effects.
During testing, the pipeline produced risk estimates for all 66,935 substations in approximately 333 milliseconds. While the system demonstrates how physics-grounded machine learning can support timely assessments, the researchers noted that further validation with utilities and operational data is required before the system can be used in actual grid operations.
Sources
- Forecasting space weather risks on power grids （Microsoft Research, 2026-09-30）