WeatherNext Outage Forecasting

Predict power outages 24-48 hours in advance using Google DeepMind's WeatherNext AI weather forecasts.

This project correlates WeatherNext AI weather forecasts with EAGLE-I power outage data on Google Cloud BigQuery to help electric utilities pre-position repair crews ahead of severe weather events.

---

What This Project Does

1. Extracts multi-ingredient weather features (wind, shear, hail proxy, precipitation, temperature) at multiple lead times (24-48h)
2. Detects outage events and evaluates which forecasts detected them
3. Scores risk per county using data-driven thresholds (p90 wind, p80 updraft)
4. Trains ML models (BQML regression) to predict outage severity
5. Pre-positions crews 24-48 hours before a storm hits

Quick Start

See the project README on GitHub for setup instructions.

Documentation

• Architecture — System design, data flow, configuration reference
• Concepts — Weather variables, outage cadence, thresholds, ML approach
• Data Sources — EAGLE-I, WeatherNext Graph, US county boundaries
• Cost Estimates — BigQuery pricing, optimization strategies
• Looker Studio — Dashboard setup and view mapping
• Vertex AI — Advanced: standalone AutoML alternative to BQML