Two-Stage Diagnosis Method for Single-Phase Conductor-Break Faults in Neutral-Point Ungrounded Distribution Networks Based on Feature Selection and BO-XGBoost

To address the weak current signatures of single-phase conductor-break faults and the difficulty in distinguishing different grounding configurations at the break point in neutral-point ungrounded distribution networks, a two-stage diagnosis method based on physics-based feature fusion, minimum redundancy maximum relevance (mRMR), and Bayesian optimization-extreme gradient boosting (BO-XGBoost) is proposed. A candidate feature set is constructed by combining sequence components with the voltages at the sending and receiving ends of the line. mRMR feature selection and XGBoost parameter optimization are then performed separately for Stage I conductor-break detection and Stage II fault-configuration identification. Simulation results show that the proposed method can distinguish conductor-break states from non-break states and identify the three conductor-break configurations. It reduces the feature dimensionality while maintaining satisfactory cascaded diagnostic performance, providing a reference for single-phase conductor-break fault diagnosis.