Online Machine Learning-Based Dynamic Security Assessment with Protection Modeling Trained on RMS-EMT Co-Simulation Data

This paper presents a machine learning-based dynamic security assessment (DSA) framework that incorporates protection system behavior using hybrid RMS-EMT co-simulation. The IEEE 39-bus system was modified to include inverter-based resources (IBRs), realistic distance relay models, and a combination of electromagnetic transient (EMT) and root-mean-square (RMS) domains. A diverse set of contingencies was simulated in DIgSILENT PowerFactory to generate a comprehensive dataset reflecting various fault types and system responses. For each distance relay, local impedance trajectories were used to train two classification models based on random forest (RF) and convolutional neural networks (CNN). The RF models achieved higher accuracy, precision, recall, and F1-score, while CNN models also demonstrated competitive performance. Both models exhibited fast inference times and strong resilience to phasor measurement unit (PMU) noise, making them suitable for real-time deployment in wide-area monitoring and control systems. The results highlight the potential of combining detailed co-simulation data with machine learning to enable accurate and timely DSA in power systems with high IBR penetration.