Gradient Boosting Feature Selection for Integrated Fault Diagnosis in Series-Compensated Transmission Lines

Series-compensated power transmission lines are integral to modern power grids and enhance the system reliability and stability. However, they introduce challenges such as voltage inversion, harmonic distortion, and nonlinear dynamics, complicating fault diagnosis in contemporary power systems. This study introduces an innovative method to analyze the waveforms of fault signals, utilizing advances in Internet and sensor technologies that provide extensive voltage and current data in a time-series form. By optimizing each phase of data processing from feature extraction to model learning, the proposed system effectively addresses fault detection, classification, and localization as a multi-classification problem. The integration of feature extraction with efficient gradient-boosting feature selection ensures high accuracy, speed, and computational efficiency, thus outperforming techniques that require heavy preprocessing. The methodology is implemented using four ensemble classifiers: Adaptive Boosting (AB), Light Gradient Boosting Machine (LGBM), Random Forest (RF), and Extreme Gradient Boosting (XGB), and is rigorously evaluated on various standard models of differing complexity and system configurations under dynamic conditions. This approach achieves high accuracy, adaptability, and robustness, overcoming practical constraints, such as high noise immunity, resource efficiency, and scalability for real-world applications. The results showed that RF achieved the highest accuracy of 99.94%, whereas XGB exhibited the fastest computational time of 0.2790 s for binary classification. RF achieved perfect accuracy (100%) for binary fault classification, whereas XGB achieved perfect accuracy for tri-class fault localization. RF achieved 99.89% accuracy in the five-class classification. LGBM demonstrates the highest noise immunity with 99% accuracy at 10 dB in ten-class multi-classification, whereas RF maintains 94.52% accuracy under extreme noise conditions (5 dB) and unforeseen signal distortions. These findings underscore the effectiveness of the proposed method, particularly the use of gradient boosting classifiers, and validate its potential for practical applications in modern power systems.