Date of Completion


Embargo Period



Outage Prediction Model; Machine Learning; Storms; Numerical Weather Prediction

Major Advisor

Emmanouil Anagnostou

Associate Advisor

Marina Astitha

Associate Advisor

Malaquias Pena

Associate Advisor

Guiling Wang

Associate Advisor

David Wanik

Field of Study

Environmental Engineering


Doctor of Philosophy

Open Access

Open Access


Storms are the primary cause of extensive power outages in electric distribution networks. Storm power outage prediction is therefore fundamental for an efficient and effective response to weather-based power outages. In this work, we address the problem of storm outage prediction by improving an existing outage prediction model (OPM) originally developed for thunderstorm and extratropical storms, and extending its structure for predicting outages associated with mixed phase precipitation (snow, ice, freezing rain) events. The herein developed OPMs use either regression trees or statistical models fed by numerical weather prediction outputs, leaf area index (LAI), infrastructure, land cover, soil type, elevation, and historical outage data to forecast number and spatial distribution of power outages during storms. OPM improvements for thunderstorms and extratropical storms consist in the introduction of new modules: a storm classifier, a multimodel optimization, and a module for implementing LAI. Cross-validation results show that the median absolute percentage error of the new OPM decreases from 130 percent to 59 percent for outage predictions at the service territory level, and that the OPM skills for operational forecasts are consistent with the skills based on historical storm analyses. Beyond thunderstorms and extratropical storms OPMs, two models are developed for predicting power outages during snow and ice storms: a machine learning (ML) based model predicting outages on a regular 4-km grid, and a generalized linear model (GLM) for power outage prediction at the town level. The most important variables for both models are assets, leaves on trees, snow density, and - for the ice model only - freezing rain and gusts. Coss-validation results show that the GLM has more skills than ML models for extreme events prediction, while ML models have better performance for lower impact events and present lower errors in the spatial distribution. The creation of reliable OPMs also allowed us to quantify the effects of a resiliency improvement on the power grid (the Enhanced Tree Trimming, ETT), using two independent methodologies. The first approach is a statistical study of the change of frequency of outage-free locations. The second approach uses the OPM as a vulnerability assessment tool to evaluate the change in the number of outages before and after ETT. From the statistical approach we show a 49% to 65% reduction of the outage-free grid cells during thunderstorms and extratropical storms coming from tree trimming. The OPM based analysis indicated that storm outages were reduced between 16% and 48% after performing ETT.