Optimal cybersecurity framework for smart water system: Detection, localization and severity assessment

Area of Science:

• Cybersecurity
• Water Systems Engineering
• Machine Learning

Background:

• Digital transformation of water distribution systems increases cyber vulnerabilities.
• Connectivity of smart devices (sensors, meters) creates security risks.
• Need for advanced security measures against cyberattacks on critical infrastructure.

Purpose of the Study:

• Propose a comprehensive cybersecurity framework for water distribution systems.
• Develop methods for cyberattack detection, localization, and impact assessment.
• Enhance the security and resilience of smart water networks.

Main Methods:

• Implemented reconstruction-based optimal cyberattack detectors: autoencoder and 1D CNN.
• Optimized detectors using Bayesian optimization.
• Employed Savitzky-Golay filtering for post-processing to reduce false alarms.
• Developed an attack localization framework and a severity index.

Main Results:

• Successfully detected all cyberattacks in the BATADAL benchmark with >98% S_TTD.
• Achieved >95% combined detection accuracy for both autoencoder and CNN models.
• Outperformed existing models in detection accuracy and minimized detection delays.
• Demonstrated effective attack localization and severity assessment on the C-Town benchmark.

Conclusions:

• The proposed framework offers a robust solution for securing smart water distribution systems.
• The approach significantly improves cyberattack detection and localization capabilities.
• The developed severity index aids in effective resource planning and decision-making for water network security.