Estimating Animal-Related Outages on Overhead Distribution Feeders Using Boosting

Faults on overhead distribution feeders have significant impact on the distribution reliability. Literature review on outages shows that overhead lines are highly susceptible to environmental factors such as weather, trees and animal. Historical analysis of outage data recorded by utility in Kansas showed that the occurrences of animal-caused outages are dependent on weather conditions and time of the year. This paper proposes models based on neural network combined with boosting algorithms based on AdaBoost to estimate weekly animal-related outages. Effectiveness of the proposed models is evaluated using actual data for four cities of different sizes in Kansas. Performance of the proposed models are compared with each other and with neural network without boosting, and previously implemented models. The results clearly show that boosting reduces mean square error and mean absolute error, and increases correlation between the estimated outages and observed outages. Additionally, AdaBoost+ performs better than AdaBoost.RT with lower errors and higher correlations.