Development of a Typhoon Power Outage Model in Guangdong, China

Abstract

In this study, typhoon power outage models were developed to predict the number of customer outages and damaged poles in Xuwen, Guangdong, China using data from three typhoons: Rammasun, Kalmaegi and Mujigae. Our hold-out validation demonstrates that these models can accurately estimate the number of customer outages and damaged poles from typhoons. The mean relative error is 5.1% in the customer outage model and 9.6% in the pole damage model. This study has two key improvements over previous work. First, our models provide more accurate predictions through leveraging a variety of static and dynamic variables that characterize the typhoon hazard and local environmental conditions. Maximum gust wind speed, elevation, land cover and duration of strong winds are the four most important variables for both the customer and pole models. Second, we demonstrate that it is possible to provide skillful predictions of power outages at 1-km resolution. This is a significantly higher spatial resolution than past tropical cyclone power outage models. The models will be used to support improved decision making. The operational power outage forecasts will be used by the electrical utility for pre-positioning crews and materials and to notify customers who are likely to lose power.