Machine Learning Embedded in Distribution Network Relays to Classify and Locate Faults

Distribution system relays equipped with machine learning (ML) algorithms can improve protection operations by performing an automated and self-learning monitoring and decision making analysis. This work evaluates the potential for a Support Vector Machine (SVM) classifier to be embedded inside each relay and provide decision making support based on the analysis of typical electrical performance measurements. The experiment, described here, used simulated data from the IEEE 123 bus feeder to understand the SVM algorithms abilities. The results showed that the algorithm deployed inside each relay could accurately classify three fault conditions that occur anywhere on the feeder, estimate the fault's region, and define a specific action for the relay switch. This initial assessment indicates that advanced, data-driven relay analysis could provide value in a typical feeder.