学术文献库

Large-scale power failures are induced by nearly all natural disasters from hurricanes to wild fires. A fundamental problem is whether and how recovery guided by government policies is able to meet the challenge of a wide range of disruptions. Prior research on this problem is scant due to lack of sharing large-scale granular data at the operational energy grid, stigma of revealing limitations of services, and complex recovery coupled with policies and customers. As such, both quantification and firsthand information are lacking on capabilities and fundamental limitation of energy services in response to extreme events. Furthermore, government policies that guide recovery are often sidelined by prior study. This work studies the fundamental problem through the lens of recovery guided by two commonly adopted policies. We develop data analysis on unsupervised learning from non-stationary data. The data span failure events, from moderate to extreme, at the operational distribution grid during the past nine years in two service regions at the state of New York and Massachusetts. We show that under the prioritization policy favoring large failures, recovery exhibits a surprising scaling property which counteracts failure scaling on the infrastructure vulnerability. However, heterogeneous recovery widens with the severity of failure events: large failures that cannot be prioritized increase customer interruption time by 47 folds. And, prolonged small failures dominate the entire temporal evolution of recovery.