Publication

Abstract : Power system faults are crucial to detect to ensure power systems' reliable and stable operation under dynamic and fault conditions. While achieving high accuracy, existing datasets and frameworks don’t allow for real-time use and don’t consider all the practical fault scenarios. To overcome these drawbacks, this work introduces a novel, integrated dataset designed explicitly for real-time fault detection and transmission line classification–a gap existing public datasets do not address. The proposed dataset includes three-phase voltage and current measurements at the critical buses of an IEEE 5-bus system. The locations are selected based on the strategic role in monitoring system stability and capturing details of faults. It covers five fault types that are simulated in seven fault locations. Unlike existing datasets, which tend to include numerous features irrelevant to fault identification, the proposed dataset is deliberately limited to only the most essential and relevant parameters. This feature selection maintains classification accuracy while reducing computational overhead. The key results demonstrate that machine learning methods such as decision trees and k-nearest neighbours achieved over 87 % fault classification accuracy. Comparative analysis confirms that the proposed dataset is highly relevant to modern machine learning applications, offering high accuracy, practical deployment potential, and minimal reliance on complex control infrastructure.