428 - A Holistic Cybersercurity Framework Against False Data Injection Attacks in Smart Water Distribution Systems Employing Auto-Encoders

Escalating cyber threats to smart critical infrastructures, such as smart water distribution systems (SWDSs), have emerged as a pressing concern. Malicious faulty data injections to the sensors, control systems, and data repositories of the SWDSs misguide the decision-making of the operator resulting in undesirable consequences. To eradicate these attack consequences, this paper proposes an unsupervised, deep-learning-based, holistic cybersecurity model leveraging auto-encoders to detect, mitigate, and protect the SWDSs against targeted false data injection attacks (FDIA). The proposed model is trained offline on the clean, simulated temporal dataset spanning one year, and evaluated online on the 3 months-long hourly sampled dataset with 5 targeted FDIA events spanning 4 hours each. The model has successfully detected all attacks present in the validation dataset alongside providing healthy, reconstructed measurements to replace the compromised sensor signals. The results revealed that the reconstructed measurements precisely follow the healthy-conditioned system’s behavior, which will help protect the SWDSs against the FDIA by replacing the tempered measurements with reconstructed, clean data. Moreover, an extensive evaluation of the model's performance substantiates its ability to send early alarming signals to the Supervisory Control and Data Acquisition (SCADA) center to undertake proactive measures against potential cyberattacks. The proposed model resulted in impressive classification scores including a recall of 1.00, a precision of 0.75, and an F1 score of 0.86. The proposed technique is generalizable and can be employed in the real-time monitoring and control systems in various smart cyber-physical systems.