DAICS: A Deep Learning Solution for Anomaly Detection in Industrial Control Systems

Deep Learning is emerging as an effective technique to detect sophisticated cyber-attacks targeting Industrial Control Systems (ICSs). The conventional approach detection in literature learn the "normal" behaviour of system, be then able label noteworthy deviations from it anomalies. However, during operations, ICSs inevitably and continuously evolve their behaviour, due e.g., replacement devices, workflow modifications, or other reasons. As a consequence, accuracy anomaly process may dramatically affected with considerable amount false alarms being generated. This paper presents DAICS, novel deep learning framework modular design fit large ICSs. key component 2-branch neural network that learns changes ICS small number data samples few gradient updates. supported by automatic tuning mechanism threshold takes into account prediction error under normal operating conditions. In this regard, no specialised human intervention needed update parameters system. DAICS has been evaluated using publicly available datasets shows increased rate compared state art approaches, well higher robustness additive noise.