Power distribution system equipment failure identification using machine learning algorithms

In this paper, an approach for identifying equipment failure faults in distribution systems is explored. This task is considered as a binary classification problem in which outages are categorized into two classes of equipment failure and non-equipment failure types. To carry out this study, actual outage data collected by Duke Energy are utilized. First, different variables that make contributions to equipment failures are described and their relationships are examined. Afterward, the presence of imbalanced classes, as a common issue in outage data set, is addressed. Then, to assure that all features are relevant, their importance is examined by employing a novel feature selection algorithm. At the end, three classification algorithms, namely decision tree, logistic regression, and naive Bayesian classifier are trained and tested and their performances are evaluated.