Improving the predictive accuracy of hurricane power outage forecasts using generalized additive models

Area of Science:

• Environmental science
• Electrical engineering
• Disaster management

Background:

• Electric power is critical infrastructure, necessitating rapid restoration after hurricanes to minimize economic and societal losses.
• Effective resource allocation (crews, materials) for power restoration hinges on accurate storm severity and outage risk predictions.
• Existing statistical regression models for pre-hurricane outage estimation have limitations in applicability and predictive accuracy.

Purpose of the Study:

• To evaluate the effectiveness of generalized additive models (GAMs) for predicting hurricane-induced power outages.
• To compare the predictive performance of GAMs against traditional generalized linear models (GLMs).
• To develop and validate a GAM for improved power outage forecasting in the Gulf Coast region.

Main Methods:

• Development and validation of a generalized additive model (GAM) using historical hurricane power outage data from the Gulf Coast.
• Comparative analysis of the GAM's predictive accuracy against established generalized linear models (GLMs).
• Statistical regression techniques applied to forecast power outage severity and geographic risk.

Main Results:

• The generalized additive model (GAM) demonstrated higher predictive accuracy than previously used generalized linear models (GLMs).
• The validated GAM provides more reliable estimates of power outages.
• Improved forecasting enables better pre-storm resource prepositioning and post-storm restoration planning.

Conclusions:

• Generalized additive models (GAMs) offer a more accurate approach to forecasting hurricane-related power outages compared to traditional methods.
• Enhanced predictive accuracy facilitates more efficient restoration of electric power services.
• This research supports improved disaster preparedness and resilience in hurricane-prone areas.