Early Detection and Classification of Power Systems Faults in Medium Voltage Distribution Lines Based on Sparse Representation

Abstract

The increasing demand for reliable and resilient electrical power necessitates advanced fault detection strategies in Medium Voltage Distribution Lines (MVDLs), which account for over 80% of power outages. This study proposes a novel approach for the early detection and classification of Short Circuit Faults (SCFs) in MVDLs using a Sparse Representation (SR) framework combined with Mel-Frequency Cepstral Coefficients (MFCCs). The method was validated through simulations on an IEEE 5-Bus System, where Line-to-Ground (L-G), Line-to-Line (L-L), and Double Line-to-Ground (L-L-G) faults were induced under both noise-free and noisy (25 dB SNR) conditions. Feature vectors derived from voltage and current signals were transformed into MFCCs, and SR was applied via dictionary learning and Orthogonal Matching Pursuit to extract fault-relevant signatures. Classification was performed using Support Vector Machines trained on sparse codes. Two models were developed: one for fault type classification and another for combined fault-type and localization using zonal segmentation. Results show that the SR-based classifier significantly outperformed conventional models (ANN, SVM) in both accuracy and noise robustness. The SR model achieved 98.7% accuracy in noise-free and 96.8% in noisy settings, with F1-scores exceeding 96%. Localization accuracy ranged from 97.5% (2-zones) to 90.5% (4-zones). The proposed SR-based classifier demonstrated high computational efficiency and scalability, enabling real-time applicability for medium-voltage grid monitoring. This research demonstrates the efficacy of integrating SR and MFCCs for robust MVDL fault diagnostics, with potential applications in enhancing grid stability, enabling predictive maintenance, and minimizing service disruptions.