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

Abstract

The digital transformation of water distribution systems has streamlined monitoring and control through the integration of smart devices such as pressure sensors, smart meters, and level switches, all communicating with supervisory control and data acquisition systems. However, this connectivity introduces cyber vulnerabilities, endangering system security and economic stability. Recent cyberattacks on critical infrastructures emphasize the urgent need for sophisticated security measures. This study proposes a novel comprehensive cybersecurity framework for cyberattack detection, localization, post-processing, and impact assessment through a severity index. The framework includes two reconstruction-based optimal cyberattack detectors: (i) autoencoder, and (ii) one-dimensional convolutional neural network, both optimized using Bayesian optimization method. A Savitzky–Golay filtering technique is employed in post-processing to reduce false alarms while preserving timely attack detection. The presented approach successfully detected all cyberattacks in the BATADAL benchmark, outperforming existing models with minimal detection delays, achieving $S_{TTD}>98\text{\%}$. It ranks first among machine learning solutions, with a combined detection accuracy exceeding 95% for both models. Additionally, an attack localization framework is developed to identify the most affected regions of the water network, and an attack severity index is formulated for resource planning and decision-making, evaluated on “C-Town” benchmark, a commonly used water network for cybersecurity studies.