When dealing with leaky Water Distribution Networks (WDNs), the need of a proper rehabilitation plan is becoming more and more urgent due to recent climate changes and serious water scarcity. Often, a WDN manager is faced to several decision variables to schedule the interventions on the network but it is not always straightforward to assess the optimal intervention order. In this work, several methodologies to define intervention schedules are proposed on the basis of engineering knowledge and expertise, a-priori considerations led by experience, and ex-ante known variables. These schedules are referred to as Ex-ante, or Engineering-Expertise, or Experience-led Schedules (ESs). Then the study is focused on the development of a Genetic Algorithm (GA) able to find an optimal intervention schedule for a given network. This schedule is then compared to ESs in the analysis of the performance of both the algorithm and ESs, also considering the computational effort required for the elaboration. Several water distribution networks are considered to study the behaviour of the methodologies applied to different networks. In particular, four WDNs with different size and properties are analysed: from little networks of 22 nodes to large networks of 2859 nodes. Several objectives can be of interest for the optimization of the interventions scheduling; the proposed approach allows the network manager to customize the optimization problem following its necessities. Different objectives are studied and two fitness functions are proposed. The first fitness function tries to minimize the Water Loss Volume (WLV) throughout the entire network during the whole intervention period. The second fitness function brings the problem back to an economic point of view: it is a multi-objective (MO) function that tries to minimize the Intervention and Management Cost (IMC). Finally, various consideration and conclusion could be appreciated: the solutions from the genetic algorithm had marginal improvements in comparison with ESs, but it allowed the analysis of the fitness functions in relation to different ESs and to different networks.
Quando si ha a che fare con reti di distribuzione idriche degradate, la necessità di un adeguato piano di riabilitazione diventa sempre più urgente a causa dei recenti cambi climatici e conseguente scarsità d'acqua. Un gestore idrico si trova spesso ad affrontare diverse variabili di decisione e trovare l'ordine ottimale degli interventi non è banale. In questo lavoro sono proposti diversi piani di intervento sulla base di conoscenze ingegneristiche, considerazioni preventive e guidate dall'esperienza e variabili conosciute a priori. Ci si riferisce a queste soluzioni come soluzioni ES. In seguito, lo studio è focalizzato sullo sviluppo di un algoritmo genetico (GA) in grado di trovare una soluzione ottima data una rete. Questa soluzione è quindi comparata alle ES per analizzare la performance di entrambe le soluzioni delle ES e la soluzione dell'algoritmo, considerando anche il costo computazionale richiesto dall'algoritmo. Molteplici reti idriche sono state considerate per studiare il comportamento delle metodologie applicate ai diversi tipi di reti. In particolare, quattro reti di diverse dimensioni e caratteristiche sono state analizzate: da piccole reti di 22 nodi fino a reti di 2859 nodi. Nell'ottimizzazione, numerosi obiettivi possono essere di interesse per l'ottimizzazione della pianificazione degli interventi; ciò permette al gestore idrico di personalizzare il problema di ottimizzazione a seconda delle sue necessità. Differenti obiettivi sono stati studiati e due funzioni fitness sono proposte in questo lavoro. La prima tenta di minimizzare il volume di acqua perso (WLV) attraverso l'intera rete per tutta la durata dell'intervento. La seconda funzione fitness riporta il problema in termini economici: è una funzione multi-obiettivo che tenta di minimizzare il costo di intervento e gestione (IMC). Infine, varie considerazioni e conclusioni possono essere dedotte: le soluzioni dell'algoritmo genetico hanno miglioramenti marginali se comparati con le ES, tuttavia l'algoritmo ha permesso l'analisi delle funzioni fitness in relazione alle differenti ES e alle differenti reti.
Effectiveness of a genetic algorithm to prioritize the rehabilitation plan of leaky water distribution networks
PREITE, LUCA
2021/2022
Abstract
When dealing with leaky Water Distribution Networks (WDNs), the need of a proper rehabilitation plan is becoming more and more urgent due to recent climate changes and serious water scarcity. Often, a WDN manager is faced to several decision variables to schedule the interventions on the network but it is not always straightforward to assess the optimal intervention order. In this work, several methodologies to define intervention schedules are proposed on the basis of engineering knowledge and expertise, a-priori considerations led by experience, and ex-ante known variables. These schedules are referred to as Ex-ante, or Engineering-Expertise, or Experience-led Schedules (ESs). Then the study is focused on the development of a Genetic Algorithm (GA) able to find an optimal intervention schedule for a given network. This schedule is then compared to ESs in the analysis of the performance of both the algorithm and ESs, also considering the computational effort required for the elaboration. Several water distribution networks are considered to study the behaviour of the methodologies applied to different networks. In particular, four WDNs with different size and properties are analysed: from little networks of 22 nodes to large networks of 2859 nodes. Several objectives can be of interest for the optimization of the interventions scheduling; the proposed approach allows the network manager to customize the optimization problem following its necessities. Different objectives are studied and two fitness functions are proposed. The first fitness function tries to minimize the Water Loss Volume (WLV) throughout the entire network during the whole intervention period. The second fitness function brings the problem back to an economic point of view: it is a multi-objective (MO) function that tries to minimize the Intervention and Management Cost (IMC). Finally, various consideration and conclusion could be appreciated: the solutions from the genetic algorithm had marginal improvements in comparison with ESs, but it allowed the analysis of the fitness functions in relation to different ESs and to different networks.| File | Dimensione | Formato | |
|---|---|---|---|
|
2023-04-16 Thesis_Luca_Preite.pdf
accessibile in internet per tutti
Dimensione
8.88 MB
Formato
Adobe PDF
|
8.88 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/10589/208855