Microsoft Research Forecasts Space Weather Risks for U.S. Power Substations
Smart operations · MEP
Microsoft Research developed an end-to-end machine learning pipeline that forecasts space weather risks for 66,935 continental U.S. substations. By integrating solar-wind data, geomagnetic indices, and local geological conductivity, it produces location-specific risk estimates 30 to 60 minutes ahead of potential impacts.
During a summer internship, Microsoft Research developed an end-to-end machine learning pipeline designed to assess space weather risks across 66,935 substations in the continental United States. The system predicts geomagnetically induced current (GIC) threats by combining solar-wind data from the L1 Lagrange point with forecasts of Auroral Electrojet (AE) and Disturbance Storm Time (Dst) indices, alongside local geological conductivity and infrastructure characteristics. It aims to provide location-specific alerts 30 to 60 minutes before impacts occur.
Key takeaways
- The multi-stage pipeline integrates L1 solar-wind data, AE/Dst forecasts, and substation geology to predict magnetic-field change rates (dB/dt) via gradient boosting.
- During evaluation across 2020–2026 data, the system detected 76.5% of major events (≥10 nT/min), 81.2% of severe events (≥20 nT/min), and 64.1% of extreme events (≥50 nT/min).
- Inference across all 66,935 substations completed in roughly 333 milliseconds, enabling rapid scenario modeling for grid planning and resilience analysis.
- Microsoft Research noted that further validation with electric utilities and operational data is required before the system can be deployed in live grid operations.
Source：Microsoft Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.