AI-Based Sensor Optimization and Anomaly Detection in Water Distribution Networks

Water Distribution Systems (WDSs) are critical infrastructures that rely on advanced technologies, such as IoT devices and smart sensors, to ensure efficient water management and delivery. While these innovations enhance operational reliability, they also increase the complexity of monitoring by adding more devices, making the network more resource-intensive and susceptible to cyberattacks. Strategic sensor placement and optimization are essential to maintaining robust anomaly detection while minimizing the number of monitored devices. In this study, we propose a sensor placement optimization and anomaly detection framework for WDSs, aimed at evaluating the effectiveness of water pumps in their current positions for anomaly detection. Using the BATADAL dataset, which contains data from water pumps, junctions, and tanks, we analyse the contribution of each pump’s location to anomaly detection. Our approach employs the Gini index within a Random Forest model to rank the importance of water pumps and assess whether their placement contributes effectively to the monitoring system. Pumps that provide minimal utility for anomaly detection are flagged for exclusion, allowing for the optimization of resource allocation. Further, we integrate a Long Short-Term Memory (LSTM) model to predict anomalies using only the essential devices. The LSTM model employs a continuous learning mechanism, enabling it to adapt to dynamic changes in the WDS, such as the addition or removal of sensors. This adaptability ensures the system remains reliable in real-time operations, maintaining robustness even in evolving scenarios.