Inverter-Dependent Transmission System Protection Using a Joint Random-Forest Classifier: Preprint

This paper presents a data-driven transmission system protection scheme based on a random forest classifier. Using voltage and current phasors and their sequence components, the classifier jointly outputs the fault status (fault/non-fault), fault type, and fault location. A moving-window architecture and a fault-confirmation logic were employed for real-time implementation and provide a tunable trade-off between data efficiency, detection speed, and security. Evaluation on 871 PSCAD simulated fault cases shows 100\% accuracy for fault detection, 97\% accuracy for fault-type identification, and 94.9\% accuracy for fault-location classification, with decision time in less than 2.5 cycles.