Comparison and validation of statistical methods for predicting power outage durations in the event of hurricanes.

摘要: This article compares statistical methods for modeling power outage durations during hurricanes and examines the predictive accuracy of these methods. Being able to make accurate predictions is valuable because information can be used by utility companies plan their restoration efforts more efficiently. also help inform customers public agencies expected times, enabling better collective response planning, coordination other critical infrastructures that depend on electricity. In long run, duration estimates future storm scenarios may utilities allocate risk management resources balance disruption from with cost hardening systems. We compare out-of-sample five distinct models estimating times caused Hurricane Ivan in 2004. The compared include both regression (accelerated failure time (AFT) Cox proportional hazard (Cox PH)) data mining techniques (regression trees, Bayesian additive trees (BART), multivariate splines). then validate our against two hurricanes. Our results indicate BART yields best prediction it possible predict reasonable accuracy.