Development of a Machine-Learning based Fault Localization Method under Complex Conditions in Distribution Power Systems

With the increasing penetration of distributed generation, distribution power systems are shifting from traditional radial structures to meshed networks, which increases the complexity and uncertainty of fault detection. Conventional FIE-based methods, relying mainly on current direction, often fail under conditions such as reverse power flow and high-impedance faults. To overcome these limitations, this study proposes a machine learning-based fault localization model that utilizes multiple features, including line current, FIE status, load demand, distributed generation output, and fault resistance. A dataset generated by distribution power system analysis software reflects diverse operating conditions, and the model is trained using the XGBoost algorithm. Simulation results show that the proposed method achieves higher accuracy and stability than conventional approaches and can be effectively applied to real-time fault detection in modern distribution power systems.