Analysis of Outage Frequency and Duration in Distribution Systems using Machine Learning

Outage duration plays an important role in assessing the impacts of distribution system outages. Moreover, whenever an outage occurs, customers are most concerned about when power will be restored, i.e. the duration of the outage. Hence, this paper presents an analysis of the frequency and duration of outages using outage data from a distribution system network. In addition, this study performs a feature importance analysis by using random forests and gradient boosting regressors to determine which features in the outage dataset are most important in predicting the duration of an outage. The results show that climatic description, failed equipment and wind speed are the most important predictors of outage duration in the dataset used in this analysis.

I. Introduction

Outage frequency and duration impact system reliability and customer satisfaction. With regards to outages, customers are most concerned about the duration of outages [1]. A major priority of utilities is to reduce the amount of time that customers are left without power. The impact of outages on customers can range from inconvenience and stress (for residential customers) to loss of revenue and man-hours (for commercial and industrial customers). The study in [2], which was conducted in 2015, estimates that the average cost per event for a momentary outage in the United States ranges from around $3.9 for residential customers to as high as$12,952 for medium and large commercial and industrial customers. Results from the same study show that the average cost per event rises as the outage duration increases. Hence, there is a need to study factors that significantly impact the duration of an outage.