This study proposes a generalized model for predicting power outages for various types of extreme weather events. To accomplish the objective of this research, diverse features (e.g., weather data, geographical features, socio-demographic data, and infrastructure information) were leveraged as independent features, while the target variable was the number of customers who had problems with their electricity at the county level. Using the time of occurrence of extreme weather as defined by the National Weather Service, the top ten influential events were selected using the impact index based on cumulative power outages due to each type of extreme weather event. Additionally, a generalized model was created to predict power outages using weather data from one hour before the outage, outage data from one hour prior, as well as socio-demographic, geographic, and infrastructure information and this model was evaluated. The model was developed in two ways: first, as an Ensemble model trained using individual extreme weather events, and second, as a Unified model using all types of extreme weather conditions. As a result of evaluating the model using mean directional accuracy (MDA), one of the evaluation metrics, the ensemble model showed an accuracy of over 0.4 for weather event types such as Winter Storms, Cold/Wind Chill, Frost/Freeze, and Ice Storms. Although this study focused on creating a model specific to Texas, it is possible to expand the data to develop a nationwide model.

This study uses transfer learning to propose a model for predicting power outages caused by various extreme weather events that are being exacerbated due to climate change. Transfer learning is a technique that allows a model trained on a significant source domain to be fine-tuned for a target domain with limited data, enabling practical regression or classification even in scarce data scenarios. In Texas, the model was developed using data-rich events like ‘Heat’ and ‘Thunderstorm Wind’ as the source domain to predict relatively data-scarce events such as ‘Excessive Heat’, ‘Winter Storm’, and ‘Flood’ as the target domain. The results indicated significant effectiveness of transfer learning, particularly in transitions such as from ‘Heat’ to ‘Excessive Heat’, ‘Heat’ to ‘Winter Storm’, and ‘Thunderstorm Wind’ to ‘Flood’. This suggests that identifying efficient combinations of source and target weather events is crucial. Moreover, this methodology is anticipated to be applicable in Texas and other states facing similar challenges.