A Holistic Cybersecurity Framework against False Data Injection Attacks in Smart Water Distribution Systems Employing Auto-Encoders

Escalating cyber threats targeting smart water distribution systems (SWDSs) have emerged as a pressing concern. In response, this paper introduces an unsupervised cybersecurity model leveraging auto-encoders (AE) to detect the SWDSs against false data injection attacks (FDIA). The model utilized simulated clean data for learning normal system behavior and underwent validation and testing against datasets containing malicious cyberattack events. Notably, the model effectively detected all FDIA cases in both datasets. Moreover, an extensive evaluation of its performance confirms its suitability for online detection as it exhibited no delays in detection. This robustness ensures that the Supervisory Control and Data Acquisition (SCADA) center is alerted to take action against cyberattacks. The proposed detector achieved impressive classification scores including a recall of 1.00, a precision of 0.89, and an F1 score of 0.91. The proposed technique is generalizable and applicable in real-time cyber surveillance in various smart cyber-physical systems.