Advanced Adaptive Protection Scheme For Modern Power Systems: A Comprehensive Machine Learning Framework

This paper presents a real-time Adaptive Protection Scheme (APS) for Medium-Voltage (MV) distribution networks with high penetration of inverter-based Distributed Energy Resources (DER). The framework combines offline and online operational phases for improving overcurrent protection coordination and decision stability under realistic operating conditions. This paper proposes a robust real-time selection algorithm for the protection groups, embedding explicit anomaly detection and anti-flapping mechanisms on top of a previously validated adaptive engine. The online algorithm includes a variational autoencoder (VAE)-based current measurement filter and outlier detector, and a support vector machine (SVM) for mapping filtered real-time current measurements to the predefined protection settings groups. A multi-criteria anti-flapping logic is then applied to enforce membership-based confidence, temporal dwell times, and M-of-N persistence before issuing setting group changes. The adaptive protection scheme is IEC 61850-compliant and can be demonstrated in real-time simulation and hardware-in-the-loop (HIL) setups. Extensive short circuit studies on the IEEE 33 bus test feeder across multiple distributed energy resources (DER) penetration levels and two fault resistance regimes (solid, 0 Ω; and 30 Ω) show a 99.9% successful fault clearing rate, with improved selectivity, sensitivity, and overall robustness of the protection scheme while significantly reducing spurious setting oscillations under DER ramps, noisy measurements, and sensor dropouts.