Detection of Faulty Regions in Power Distribution Grid with Limited Measurement

Restoration of power supply after occurrence of faults in distribution lines is utmost challenging task for operators due to lack of sufficient real time information in the distribution system territory. Recently initiatives are taken by distribution companies to install micro phasor measurement units ($\mu$PMUs) and sensors to monitor and recognize the injurious area in the system. However, implementation of the projects are still challenging due to high investment cost. The present study tries to highlight a machine learning based approach to identify faulty region in the distribution system with limited installation of $\mu$PMU. A random forest (RF) classifier is utilized to assess the faulty region in the distribution grid with low state information supplied by $\mu$PMUs. The method endows the training with a large simulated data obtained from installed $\mu$PMUs for different types and location of faults in three phases of the system. Proposed methodology has been tested on IEEE 13 bus unbalanced distribution system. The efficacy of the proposed approach is established with the test results. The investigation also depicts the efficiency of RF algorithm in fault detection in comparison with other classifier such as support vector machine (SVM) and artificial neural network (ANN).