| Ensuring the reliability of the U.S. electric power infrastructure systems is of utmost importance. Due to the complex interdependencies that exist between the electric infrastructure and all other critical lifelines in the U.S., disruption in the electric sector can adversely affect our national security, digital economy, public health, and the environment, and have debilitating socio-economic impacts on our society. Hurricanes regularly cause widespread and prolonged power outages in the U.S. Having accurate, pre-landfall estimates of the degree of hurricane impacts can help the managers of other critical utilities, emergency response personnel, and other critical service providers and government officials throughout the impacted area best prepare for and respond to the hurricane.;The overarching goal of this dissertation is to develop a series of reliability models for coastal power distribution systems that are prone to hurricane impacts. Chapter 1 presents the scope of my thesis, followed by an introduction to the statistical learning methods explored in this dissertation. Chapter 2 presents a seasonal forecast model that estimates the annual number of Atlantic hurricanes. The proposed model achieves higher predictive accuracy than the current leading hurricane forecast models.;Chapters 3, 4 and 5 present the development of different types of power outage forecast models that are essential in painting an informative picture of the state of the reliability of coastal power systems that are prone to hurricane impacts. More specifically, Chapter 3 present two classes of predictive models: a utility-specific and a generalized forecasting model that can estimate the duration of power outages caused by hurricanes. Chapter 4 presents an outage duration model that can estimate the number, and geographic location of hurricane-induced power outages with great accuracy. Chapter 5 presents a geographically generalized outage model that estimates the fractions of customers without power prior to a hurricane landfall. All the proposed models achieve higher predictive accuracy that the current operational models. Chapter 6 closes with a summary of the research contributions of this dissertations, and concludes with outlining a few possible future research directions that could build upon and improve the work presented in this thesis. |
