Author(s): Carlos Penas; Roberto Minguez
Linked Author(s): Roberto Mínguez
Keywords: Intelligent water network Water distribution networks State estimation Parameter calibration Network prognosis EPANET
Abstract: This project presents the development the software package SENSEI (acronym for Smart Water Networks Using Artificial Intelligence) which has been designed to enhance state and losses estimation, and parameter calibration for water distribution networks, leveraging the EPANET API. The package employs a pressure-driven approach, which is widely recognized for providing a more realistic representation of network behavior under varying pressure conditions compared to traditional models. This approach is particularly suitable for addressing challenges such as leak detection and demand variability, which together form part of the application techniques of artificial intelligence for smarter water management systems, as highlighted in recent studies. The mathematical and computational procedures behind the package are based on previous research that has developed pressure-driven state estimation algorithms, which involve solving a non-linear weighted least squares problem. This project also improves these methods in a context of a multi-period analysis, allowing dynamic calibration of pipe roughness coefficients over time. Unlike conventional single-period methods, this multi-period approach captures variations in pipeline conditions due to aging or changes in operating environments, providing more accurate and robust estimations. By incorporating data at multiple time steps, the algorithm improves the accuracy of condition estimation, helping utilities monitor changes to infrastructure and detect issues such as leaks more effectively. Furthermore, the package is capable of efficiently solving nonlinear weighted least squares problems, which is the mathematical basis for solving state estimation. The SENSEI package is a pioneering tool for the intelligent management of water distribution networks. It is a versatile solution that supports real-time monitoring, and hence allows, among others, the performance of tasks such as prognosis of potential problems (like breakages in real time), long term estimation of leakages and roughness coefficient estimation. All of the above forms a set of powerful tools to help in decision making and, ultimately, to address unaccounted-for-water. The package integrates seamlessly with EPANET; and as a novelty, in addition to supporting networks that have elements that the aforementioned software usually works with, it also supports the use of sensor data (such as pressure measurements in junctions, head levels in tanks and reservoirs, and flows in pipes) and the use of pseudo-measurements, to fill gaps when real data is incomplete, making them adaptable to various operational scenarios. Furthermore, from a mathematical point of view, the methodology is designed to consider the uncertainties inherent to including information from sensors and pseudo-measurements. The methodology has been tested on a real water distribution network with a total of more than 2000 kilometers of pipes, demonstrating its effectiveness in providing valuable information for system diagnosis, leak detection and operational management on realistic networks. In summary, by bridging the gap between traditional hydraulic modeling and modern data-driven approaches, SENSEI offers water utilities and practitioners a practical and efficient solution for optimizing network performance and planning. As a pioneering tool, it contributes significantly to the reliability, efficiency, and sustainability of water distribution systems, supporting smarter and more adaptive management practices.
Year: 2025