Fault Location Estimation through Tree-Based Prediction Model for Multiple Estimation Elimination

Abstract

For the reliable operation and integrity of the power distribution system, recovery is crucial in the event of a fault. Power distribution systems supply electricity to multiple consumers, making them susceptible to faults caused by natural disasters, equipment malfunctions, and various other reasons. Therefore, accurate identification of the fault location is essential to minimize damages resulting from power outages. In this context, the study of fault location in distribution systems is actively researched to enable precise and rapid recovery. Among various methods, impedance-based single-ended techniques utilizing fundamental wave components are commonly employed for fault location in distribution systems. The majority of methodologies calculate the equivalent distance to the fault point based on the substation as a reference. At the substation, voltage and current data before and after the fault occurrence are obtained. Subsequently, the fault segment is estimated through iterative comparison between line impedance and the estimated equivalent impedance. However, impedance-based methods face the challenge of multiple estimation problems, where several fault location candidates satisfying the same impedance conditions may exist. In this paper, a supervised learning-based approach is proposed to estimate fault points by training fault paths and segments. The proposed method explores fault path candidates based on data from the substation, eliminating multiple estimations and calculating the fault distance using the estimated fault data. The effectiveness of the proposed method is validated through verification in various fault conditions in an IEEE 123 test system using OPEN DSS and Python.