Augmenting distance protection with a data-based approach to fault classification under uncertainty

Electric Power Systems Research, vol. 262, pp. 113544

Abstract

The increasing use of grid-connected converters challenges the dependability and security of distance protection. Unlike synchronous machines, converters provide limited and control-dependent fault currents distorting apparent impedance trajectories. In combination with low short-circuit strength and varying pre-fault operating states, this increases the risk of misoperation of conventional distance relays, particularly near the Zone 1 boundary. This work applies Random Forest classification to the IEEE 14-bus system under varying system and operating conditions. The classifier reaches 93.6% accuracy for 3ph-gnd faults and reveals systematic misclassifications. SHAP analysis identifies apparent impedance, pre-fault power flow and system strength as dominant drivers, linking data-driven results to classical protection concepts. Threshold and receiver–operating–characteristic analyses show that optimal operating points deviate from fixed reach settings and shift the balance between dependability and security. A 90.08% reach performs best for 3ph-gnd faults, while 77.58% improves 1ph-gnd performance, indicating that static geometric zones cannot ensure balanced protection performance.

Authors 3

1. Sarah Anhaus corresponding Aachen

   RWTH Aachen University

   Affiliation as printed

   IAEW of RWTH Aachen University, Aachen, Germany

2. Arlind Damjani Aachen

   RWTH Aachen University

   Affiliation as printed

   IAEW of RWTH Aachen University, Aachen, Germany

3. Willem Leterme Aachen

   RWTH Aachen University

   Affiliation as printed

   IAEW of RWTH Aachen University, Aachen, Germany