The evaluation of the optimal number of sensors, together with their positioning, is a crucial choice for a simultaneously efficient and economically sustainable monitoring system. Most studies have focused on approaches based on simulation algorithms, which explicitly consider water flow within networks. These methodologies require well-calibrated hydraulic models of high quality and calculation capabilities not possessed by all water companies. For this purpose, this paper develops a study based on the generation of clusters of topologically uniform distribution networks to identify the best-performing positioning strategies. Twelve real case studies were divided into clusters through the implementation of the K-Means algorithm, combined with the use of methods concerning multivariate statistics. Each network belonging to the relevant cluster was studied using the Graph Theory approach, a computationally inexpensive method that only requires topological information on the distribution system by mapping the network as a graph consisting of nodes and vertices. The various sensor layouts studied were obtained through the adoption of four different types of graphs, the use of five clustering algorithms and the use of eight centrality measures to identify the most central nodes as potential sensor locations. The results of this approach were evaluated through the use of hydraulic simulations used as a one-off validation tool, with the aim of identifying the types of graphs, algorithms and measures that provide the best performing sensor layouts. By aggregating and normalizing the overall results of each network topologically associated with a cluster, the best-performing average layouts relative to the cluster itself were identified, allowing the direct estimation of placements using Graph Theory. These results were applied and validated through the study of a thirteenth real network, highlighting how this generalized approach is a promising result for its future application, without the need for any hydraulic simulation.
La valutazione del numero ottimale di sensori, unita al loro posizionamento, costituisce una scelta cruciale per un sistema di monitoraggio contemporaneamente efficiente ed economicamente sostenibile. Gran parte degli studi si è concentrata su approcci basati su algoritmi di simulazione, che considerano esplicitamente il flusso d'acqua all'interno delle reti. Tali metodologie richiedono modelli idraulici ben calibrati di alta qualità e capacità di calcolo non possedute da tutte le aziende idriche. A tale scopo, il presente lavoro sviluppa uno studio basato sulla generazione di cluster di reti di distribuzione topologicamente uniformi che consenta di identificare le strategie di posizionamento più performanti. Dodici casi studio reali sono stati suddivisi in cluster tramite l’implementazione dell’algoritmo K-Means, unito all’utilizzo di metodi concernenti la statistica multivariata. Ogni rete appartenente al relativo cluster è stata studiata tramite l’approccio basato sulla Teoria dei Grafi, metodo computazionalmente poco oneroso, che richiede solo informazioni topologiche sul sistema di distribuzione mappando la rete come un grafo costituito da nodi e vertici. I vari layout di sensori studiati sono stati ottenuti tramite l’adozione di quattro diversi tipi di grafi, l'uso di cinque algoritmi di clustering e l’uso di otto misure di centralità per identificare i nodi più centrali come potenziali posizionamenti di sensori. I risultati di tale approccio sono stati valutati tramite l’utilizzo di simulazioni idrauliche usate come strumento di validazione una tantum, con l'obiettivo di identificare i tipi di grafi, gli algoritmi e le misure che forniscono i layout dei sensori più performanti. Aggregando e normalizzando i risultati complessivi di ogni rete associata topologicamente ad un cluster, sono stati individuati i layout mediamente più performanti relativi al cluster stesso, permettendo la diretta stima dei posizionamenti tramite la Teoria dei Grafi. Tali risultati sono stati applicati e validati tramite lo studio di una tredicesima rete reale, evidenziando come tale approccio generalizzato costituisca un risultato promettente per la sua applicazione futura, senza la necessità di alcuna simulazione idraulica.
A generalized topological based approach for water quality sensor placement in drinking water distribution networks
Di Domenico, Laura
2021/2022
Abstract
The evaluation of the optimal number of sensors, together with their positioning, is a crucial choice for a simultaneously efficient and economically sustainable monitoring system. Most studies have focused on approaches based on simulation algorithms, which explicitly consider water flow within networks. These methodologies require well-calibrated hydraulic models of high quality and calculation capabilities not possessed by all water companies. For this purpose, this paper develops a study based on the generation of clusters of topologically uniform distribution networks to identify the best-performing positioning strategies. Twelve real case studies were divided into clusters through the implementation of the K-Means algorithm, combined with the use of methods concerning multivariate statistics. Each network belonging to the relevant cluster was studied using the Graph Theory approach, a computationally inexpensive method that only requires topological information on the distribution system by mapping the network as a graph consisting of nodes and vertices. The various sensor layouts studied were obtained through the adoption of four different types of graphs, the use of five clustering algorithms and the use of eight centrality measures to identify the most central nodes as potential sensor locations. The results of this approach were evaluated through the use of hydraulic simulations used as a one-off validation tool, with the aim of identifying the types of graphs, algorithms and measures that provide the best performing sensor layouts. By aggregating and normalizing the overall results of each network topologically associated with a cluster, the best-performing average layouts relative to the cluster itself were identified, allowing the direct estimation of placements using Graph Theory. These results were applied and validated through the study of a thirteenth real network, highlighting how this generalized approach is a promising result for its future application, without the need for any hydraulic simulation.File | Dimensione | Formato | |
---|---|---|---|
2022_07_DiDomenico.pdf
Open Access dal 03/07/2023
Descrizione: Testo della Tesi
Dimensione
6.04 MB
Formato
Adobe PDF
|
6.04 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/190277