Robust Centralized Protection Scheme With AI-Based Fault Diagnosis Capabilities for Graph-Structured AC Microgrids

This paper presents graph neural networks (GNNs)-based fault diagnostic framework (GFDF) with cyber-attack detection capabilities for ac microgrids (MGs). GFDF employs GNNs on graphical representation of MGs, augmented with a multi-head attention mechanism, to accurately assimilate dynamics associated with fault events by learning node embeddings. This approach effectively assigns weights to the neighboring nodes based on their contributions, ensuring resilience to abnormal data and adaptability to changing operating conditions. GFDF uses current measurement of single end of each line and line parameters as graph node and link attributes, respectively. Additionally, this paper proposes a robust intelligence-based centralized protection scheme (ICPS), intended to address the failure of legacy protection infrastructure in MGs caused by various logical and physical reasons. It utilizes decisions made by GFDF with accelerated computation throughput using dedicated hardware (GPU-NVIDIA GeForce GTX 1650) to meet stringent protection time requirements. A comparative assessment of GFDF with the existing techniques, and the implementation of ICPS on medium voltage CIGRE MGs through hardware-in-the-loop (HIL) experimentation, leveraging real-time digital simulator (RTDS) setup, and commercial SEL relays to emulate realistic operational environments, validates the practicality of the work.