An interpretable multi-dimensional feature space-based fault diagnosis method for UHVDC transmission line

Accurate and rapid fault diagnosis is critical to the safe and stable operation of ultra-high-voltage direct current transmission systems. The limited protection time window restricts fault feature extraction and may result in protection misoperations, especially under complex fault conditions. To address these challenges, a novel diagnostic approach is proposed based on multi-dimensional feature space construction guided by fault information distribution. Physically interpretable features are first derived through circuit-theoretic analysis to ensure strong physical relevance. Principal Component Analysis is then applied to capture dominant statistical characteristics and reduce feature redundancy. XGBoost is subsequently employed to model nonlinear feature interactions and improve classification accuracy. Furthermore, SHapley Additive exPlanations are integrated to quantify individual feature contributions, facilitating feature space optimization and enhancing model interpretability. Simulation results show that the proposed method achieves accurate fault identification within a 1 ms window under fault resistances up to 500 Ω and maintains robust performance across varying sampling frequencies. • A novel UHVDC fault diagnosis method using multi-dimensional feature space. • Features are extracted via circuit analysis and optimized using PCA. • XGBoost models nonlinear features; SHAP quantifies their contributions. • Achieves accurate, fast diagnosis under 500 Ω faults and varied sampling rates.