A Machine Learning Approach for Fault Detection in Series-Compensated Three-Phase Transmission Line System Using DWT-Based Feature Extraction

Abstract

This paper presents a machine learning-based approach for fault detection in Series-Compensated Three-Phase Transmission Line Systems using Discrete Wavelet Transform (DWT)-based feature extraction. Fault detection plays a critical role in ensuring the stability and reliability of power systems by enabling early identification of abnormal operating conditions. In the proposed method, current signals are sampled at 20 kHz to accurately capture transient fault signatures. These signals are then decomposed using the Symlet 4 (Sym4) wavelet up to Level 8, where significant fault-related features are embedded in the detail coefficients. The Median Absolute Deviation (MedAD) of the Level 8 detail coefficients (D8) is calculated and used as the key feature input to an Ensemble Subspace K-Nearest Neighbors (KNN) classifier. The experimental results demonstrate the effectiveness of the proposed approach, achieving 100% detection accuracy, a prediction speed of 1000 observations per second, a training time of 1.6102 seconds, and a detection time of 17.7 ms. These outcomes confirm the suitability of combining DWT and machine learning for accurate and real-time fault detection in series-compensated transmission systems.