Contributions to the real-time leak management of water systems
(English) This thesis presents several contributions to the state of the art of real-time monitoring of water distribution networks. Specifically, this thesis analyses the leak management problem, one of the major challenges within the field of monitoring of water distribution networks. Its importan...
| Autor: | |
|---|---|
| Tipo de documento: | tese |
| Data de publicação: | 2025 |
| País: | España |
| Recursos: | Universitat Politècnica de Catalunya (UPC) |
| Repositório: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglês |
| OAI Identifier: | oai:dnet:upcommonspor::e2c745a80242105c8f6b3616e73a2c1e |
| Acesso em linha: | https://hdl.handle.net/2117/460936 https://dx.doi.org/10.5821/dissertation-2117-460936 |
| Access Level: | Acceso aberto |
| Palavra-chave: | 628 - Enginyeria sanitària. Aigua. Sanejament. Enginyeria de la il·luminació 004 - Informàtica Àrees temàtiques de la UPC::Enginyeria civil Àrees temàtiques de la UPC::Informàtica |
| Resumo: | (English) This thesis presents several contributions to the state of the art of real-time monitoring of water distribution networks. Specifically, this thesis analyses the leak management problem, one of the major challenges within the field of monitoring of water distribution networks. Its importance lies in the high costs associated to the water losses. The thesis proposes contributions to address two fundamental problems: the localization of leaks and the strategic placement of sensors to improve leak localization. These contributions are mainly focused on data-driven methodologies, characterized by their independence from hydraulic models, which can be hard to generate and calibrate for water utilities. About leak localization, an initial two-stage data-driven approach is proposed. First, a quadratic programming problem estimates the complete hydraulic network state, given by the hydraulic head (pressure + elevation) at the nodes. This process, denoted as Graph-based State Interpolation (GSI), only uses data from pressure sensors and the structure of the network. Then, a technique called Leak Candidate Selection Method (LCSM) compares leak and leak-free estimated states to indicate the leak location. The complete estimation-localization method is called GSI-LCSM. Throughout the thesis, several improvements are proposed for this method. Firstly, learning schemes are applied or combined with GSI-LCSM to improve node-level localization accuracy. Three different learning-based methods are presented, namely GSI-DL, which uses Dictionary Learning (DL) to learn from estimated residuals; LL-GSI-LCSM, which adds an adaptive learning layer to GSI, improving the localization during its online application; and DeepFGSI, which derives a simplified version of GSI, used to add network information to the layers of a deep learning scheme. Then, the interpolation of GSI is improved by considering the physics behind the dynamics of a water network, leading to the development of the Analytical Weights GSI (AW-GSI) strategy. This method uses the structure of the network and the nominal reconstructed state to generate a new set of graph weights, which are then used to configure a quadratic programming problem, analogue to GSI. Furthermore, additional sensor types, such as flow sensors or demand meters, are not considered by GSI or AW-GSI to improve the estimation process. Therefore, various sensor fusion strategies are conceived, leveraging the Unscented Kalman Filter algorithm, leading to the UKF-GSI estimation approach. This process is coupled with LCSM in order to improve leak localization. Finally, the sensor placement problem is also explored through a model-free perspective. This strategy uses genetic algorithms (GA) to minimize a structural metric related to node-sensor distances. The GA-based operations are customized for the problem at hand, and premature convergence countermeasures and stopping conditions are defined in order to lead to suitable sensor placement solutions. All the proposed methodologies are tested using the L-TOWN benchmark from the BattLeDIM2020 challenge, which consisted in a leak detection and localization competition with several international teams. In this thesis, an initial comparison between GSI-LCSM and a model-based method is provided, leading to conclusions about the advantages of each category of methods. Then, all the proposed improvements to GSI-LCSM are compared to the base methodology, demonstrating the suitability of the methods and analysing the advantages that each one adds to GSI-LCSM. |
|---|