ASTRO: Adaptive Spatio-Temporal Reinforcement Optimization for GNN Powered Anomly Detection in Cyber Physical Systems
Researchers have developed ASTRO, a new anomaly detection framework for cyber-physical systems that utilizes reinforcement learning and Graph Neural Networks. ASTRO dynamically optimizes decision boundaries by integrating a Deep Q-Network with GNNs, temporal modeling, and an attention mechanism to capture spatial and temporal dependencies. The framework demonstrated strong performance on real-world industrial datasets, achieving an F1 score of 0.990 on the SWaT dataset and outperforming existing methods by nearly 14% on the WADI dataset. AI
IMPACT This framework could enhance the security and reliability of industrial control systems by providing more accurate and adaptive anomaly detection.