Predicting Grid Faults Before They Happen: AI for Weather-Driven Grid Resilience
summary
P4R converts 50-member weather ensembles stored in Zarr with 50-member ensembles into calibrated probabilistic fault forecasts up to five days ahead, using Python feature pipelines and XGBoost quantile-regression models orchestrated on Kubernetes. Synthetic historical ensembles 2017–2024 reconstruct missing ensemble-era predictors (like CAPE) to enable consistent backtests. Deployed at SP Energy Networks and Scottish and Southern, forecasts refresh four times daily, drive amber/red business bands, and helped pre‑position crews so that during a 350,000‑customer storm 95% restored within 24 hours.
AI-written summary · about
chapters (6)
- 0:00Introduction to P4R project
- 3:51The core problem and solution
- 8:09Modeling approach and uncertainty
- 11:01Trial results and real-world impact
- 19:27Benefits and future roadmap
- 24:06Q&A session
related talks
42mPredicting Weather-Caused Rare Events: A Utility Outage Prediction Use Case
09mGen AI Workflows for Climate Risk Modeling in Food Systems
38mElectrifying Big Issues Through Analytics
40mDeep Learning and Time Series Forecasting for Smarter Energy
30mPredictive Maintenance and Anomaly Detection for Wind Energy
11mEnhancing Coastal Resilience with AI
42mRobust Automated Forecasting in Python and R
25mWeather Forecasting: An Application of Artificial Intelligence, Cloud, IOT &...