Poster Abstract: Autonomous AI-Driven Grid Protection: Sub-Cycle Fault Response via NPU-Optimized Neural Networks

Abstract

Relay Protection and Automation (RPA) devices maintain powersystem stability by isolating faults using predefined or adaptivethresholds. However, even flexible configurations require manualpre-tuning, limiting their adaptability in dynamic grid environments,such as those with renewable energy integration. ArtificialIntelligence (AI) augments RPA functionality by generalizing tonew, unseen data, improving fault detection accuracy in transientor noisy scenarios. Our early studies demonstrate that embeddedAI systems outperform static thresholds while operating in realtime. These systems, implemented via neural networks on Systemon-Chip (SoC) platforms with dedicated Neural Processing Units(NPUs), avoid cloud-dependent latency.