Abstract
Modern distribution networks increasingly incorporate intelligent automation schemes to enhance resilience and reduce service interruptions following faults. To support these strategies, this paper investigates the use of machine learning models for fault location, aiming to quickly identify the faulted area and support safe service restoration of non-faulted areas. A comparative study is conducted using three supervised learning methods: Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB), applied to fault location in a distribution test system adapted to include Distributed Energy Resources (DER). Using steady-state current measurements generated from probabilistic fault scenarios based on historical data, each model is evaluated in terms of classification accuracy and computational feasibility. Results indicate that the models demonstrated high classification accuracy and efficient execution time, confirming the viability of machine learning (ML)-based approaches as effective decision-support tools for intelligent fault isolation and service restoration.