Predicting Weather-Caused Rare Events: A Utility Outage Prediction Use Case
summary
Claims a deployed ML pipeline that predicts weather-caused utility outages hourly for 24–48 hour horizons and lets operations mobilize crews 12 hours ahead to avoid penalties and reduce millions in restoration costs. Uses PySpark for large-scale feature aggregation, trains ensemble models (random forest regressors) with automated selection and random forest + SHAP and tree-based diagnostics to explain predictions, removes anomalous long-tail tickets and groups sparse operation centers; reported production use and improved proactive mobilization.
AI-written summary · about
chapters (8)
- 0:00Weather outage prediction intro
- 3:35Why outages matter
- 8:22Operational challenge
- 12:59Data challenges in rare events
- 18:55Modeling rare outages
- 23:29Explaining predictions
- 31:10Automating training
- 34:29Q&A
related talks
32mPredicting Grid Faults Before They Happen: AI for Weather-Driven Grid Resilience
26mProactive Failure Prediction Through Anomaly Detection
43mPredictive Maintenance: From POC to Production With Spark
42mRobust Automated Forecasting in Python and R
29mProbabilistic demand forecasting at scale
37mForecasting using machine learning methods
40mDeep Learning and Time Series Forecasting for Smarter Energy
01h 16mPredicting Weather Patterns using Time Series ML