抄録
Electric power systems in Japan suffer from major damage from typhoons almost every year. Typhoons often cause undesirable long-time power failures. To speedily restore the electric power supply, one needs to predict the amount of damage accurately for an approaching typhoon. This paper presents a method of predicting the amount of electric power damage for the Amami archipelago using Gaussian process (GP) regression. The typhoon track is evaluated by considering its wind characteristics. A nonstationary kernel is introduced as the covariance function of the GP in consideration of the nonstationarity of damage. This predictor can yield not only the predicted amounts of damage but also their confidence measures. Simulation results based on actual data are shown to illustrate the effectiveness of this approach.
