Artificial Neural Network SoftSensor for Water Distribution Networks

Water distribution networks (WDNs) are critical infrastructure, delivering a vital resource yet facing challenges in monitoring and maintaining optimal performance. This paper proposes a new methodology to develop a softsensor (SS) for WDNs that leverages the power of Artificial Neural Networks (ANNs). The proposed softsensor integrates the digital twin of the WDN, a virtual replica that mirrors the physical network. This approach minimizes the physical sensors needed for data acquisition, reducing hardware costs and maintenance complexities. The ANN is trained using a Computational Fluid Dynamics (CFD) dataset, capturing the intricate hydraulic behavior of the WDN. This ANN-based softsensor offers unique advantages. Unlike traditional sensors that provide data for specific locations, the ANN can estimate parameters at different network points based on a single unknown value queried. Our evaluation reveals a remarkable accuracy of 99%, signifying a significant advancement in predictive capabilities for WDNs. Real-time monitoring, anomaly detection, and proactive maintenance strategies can benefit from the proposed methodology.