DEEP LEARNING–BASED ANOMALY DETECTION IN LARGE-SCALE CYBER-PHYSICAL SYSTEMS

• Dr. Michael R. Johnson

  Department of Computer Engineering, Norway

  Author

Cyber-Physical Systems (CPS) operate through tight integration of computation, networking, and physical processes, making them vulnerable to anomalous behaviors caused by cyberattacks, sensor faults, and system malfunctions. With increasing system complexity and data volume, conventional statistical methods often fail to capture non-linear patterns in large-scale CPS environments. This study investigates a deep learning–based anomaly detection framework employing LSTM networks, Autoencoders, and Temporal Convolutional Networks (TCNs). Experimental evaluations performed using the SWaT and WADI CPS datasets demonstrate that the proposed model achieves high accuracy (98.3%), low false-positive rates (1.8%), and strong generalization to unseen disturbances. The results confirm the feasibility of deep learning as a robust anomaly detection solution for modern CPS infrastructures.

• pdf

This work is licensed under a Creative Commons Attribution 4.0 International License.