Identification of Line Faults in Power Distribution Networks Supported by Micro Phasor Measurement Units

Abstract

Power distribution systems are becoming more complex and vulnerable day by day due to the increasing penetration of non-linear loads, renewable sources, and critical expansions of the overhead supply networks. Thus, the chances of distribution line failures have increased. On this occasion, effective fault detection aids in developing predictive maintenance strategies, helping utilities reduce downtime and utilize resources more effectively. In this paper, a machine learning (ML)-based methodology for fault detection on overhead distribution lines has been presented. The study has demonstrated the application of the two typical tree-based ML classifiers, i.e., random forest (RF) and extreme gradient boosting (XGBoost), to diagnose distribution line faults with the help of a micro phasor measurement unit (μPMU) supported data set. The study reveals a μPMU allocation strategy and data collection method to aid in the preparation of the data set. The classification models are structured to classify the location of the line faults, followed by the faults' type, based on the prepared data set for a test network. The efficiency of the RF and XGBoost has been verified through a quantitative comparison with the existing approaches.