Perfomance comparison of Machine Learning (ML) based fault detection scheme for a compensated transmission line

Compensated transmission lines are employed in power system networks to increase energy transmission capacity, improve system stability, decrease transmission losses, and provide more flexibility in power management. However, the effectiveness of protection relays in detecting, classifying, and locating faults in such networks is significantly impacted. This is due to changes in impedance within the fault loop caused by devices used in compensated networks, which alters the apparent impedance seen by the relay. This study introduces three different fault detection and classification schemes for compensated transmission lines. Initially, an IEEE 9-bus system with series compensation at line 4–6 is utilized for data collection. Feature extraction using Principal Component Analysis (PCA) is then conducted, followed by the implementation of fine k-nearest neighbors (KNN), Medium Nuero network (MNN), and SQVM classifiers. A comprehensive performance analysis is carried out, revealing that SQVM achieves the highest overall accuracy of 99.3%, while the fine tree outperforms SQVM in single-line fault detection.