Application of machine learning algorithms for predicting vegetation related outages in power distribution systems

Abstract

A large number of faults in power distribution systems is caused due to vegetation growing near power lines. Therefore, to maintain high system reliability, outages should be prevented as much as possible before they occur. This paper proposes a data-driven approach to predict vegetation-related outages in power distribution systems. Three Machine Learning (ML) methods i.e., the Neural Network (NN), Decision Tree Classifier (DTC) and Random Forest Classifier (RFC) are used to predict the vegetation-related outages. Historical outage data and weather data are used as the inputs to the ML methods. Then, the ML models are trained and used to predict the probability of occurrence of an outage in the next fourteen days. A risk map is generated by incorporating the geographical location of distribution feeders based on the predicted outage probabilities. Moreover, a real-time outage prediction platform is developed to provide the utilities a better insight into vegetation-related outages. The accuracy of predicting failures is found to be 72.57%, 84.06% and 93.79% for NN, DTC and RFC, respectively.