Professional Documents
Culture Documents
Abstract: Water distribution systems can be divided into district metered areas (DMAs) to improve their management. DMAs are
Downloaded from ascelibrary.org by Zhejiang University on 11/11/22. Copyright ASCE. For personal use only; all rights reserved.
individual service regions within a distribution system that have a defined set of supply sources, and hydraulic and quality parameters
can be controlled at their entrances. Designing DMA configurations to meet management goals, such as pressure or demand uni-
formity, is challenging because of the unique topological conditions of each water system, the variability in water demand required by
each geographic zone, and the high interconnection of pipes within water systems. This research couples geospatial analysis with a
hydraulic simulator to design DMAs for a highly constrained large water network. Solutions specify modifications of valve settings
and tank operations and the addition of new pipes to the network. The performance of the DMA design is evaluated based on ob-
jectives to minimize implementation cost, demand similarity, pressure uniformity, and water age and constraints related to pressure
and number of entrances per district. The methodology is applied to design DMAs for E-Town, a large water system that was pre-
sented at the Battle of Water Networks District Metered Areas. The coupled approach identifies DMA designs that satisfy constraints
with satisfactory performance for multiple objectives. DOI: 10.1061/(ASCE)WR.1943-5452.0001243. © 2020 American Society of
Civil Engineers.
Author keywords: Urban water systems; District metered areas (DMAs); Hydraulic modeling; Geospatial tools.
Introduction when defining the boundaries of the districts, due to the instal-
lation of shut-off valves that reduce redundancy in a looped
Water distribution systems (WDSs) are composed of highly inter- system (Gomes et al. 2015). Stagnation may occur through
connected loops of pipes, and the main purpose of the WDS is to the introduction of new dead ends, and water quality may be
supply water to connected customers while satisfying hydraulic impacted (Armand et al. 2018). The implementation of DMAs
and water quality parameters. With the development of new urban may also be constrained by economics because retrofitting a
areas, water systems have been expanded to provide water to new WDS into DMAs requires the installation of flow meters and
settlements, and the difficulty of controlling the performance of a in some cases pressure-reducing valves (PRVs) to isolate subsec-
system increases with its size. To improve control of large net- tions (Galdiero et al. 2015). On the other hand, the implemen-
works, utilities define sectors or zones (Di Nardo et al. 2013) tation of DMAs can allow a utility to perform water balances and
and aggregate the analysis of small networks to gain better insight address leakage problems. Contaminants that are introduced to
and control over the performance of the network as a whole. The the WDS can be identified more quickly if the WDS is sectorized
concept of a district metered area (DMA) was introduced as a de- using DMAs, leading to a more secure system (Grayman et al.
limited zone within a water distribution system to address leakage
2009; Di Nardo et al. 2014).
problems (Water Research Centre 1980), where boundaries are de-
Several methodologies have been developed to design DMAs
fined and hydraulic parameters can be controlled at the entrance of
for water distribution systems. DMAs were used initially for leakage
each DMA.
management applications, and utilities performed water balance
Dividing a WDS into DMAs is a nontrivial process, due to
audits to identify nonrevenue water. Exhaustive search procedures
the unique characteristic of water distribution networks. One of
were improved by incorporating engineering judgment to identify
the most important features of a water network is reliability, or
zones with similar geographical and topographic characteristics
the ability to provide noninterrupted service. Therefore, the de-
(MacDonald and Yates 2005; Morrison et al. 2007). Graph theory
sign of DMAs is constrained by a potential decrease in reliability
and partitioning techniques coupled with optimization algorithms
1
Ph.D. Student, Dept. of Civil, Construction and Environmental Engi- have been widely applied to automate the design process (Herrera
neering, North Carolina State Univ., Raleigh 27695, NC (corresponding 2011; Perelman and Ostfeld 2011; Diao et al. 2013; Di Nardo et al.
author). ORCID: https://orcid.org/0000-0002-1537-6006. Email: jpesant@ 2016). Researchers have also developed social network analyses,
ncsu.edu community structure, and multiobjective optimization within multi-
2
Professor, Dept. of Civil, Construction and Environmental Engineer- stage procedures (De Paola et al. 2014; Scarpa et al. 2016; Brentan
ing, North Carolina State Univ., Raleigh 27695, NC. et al. 2017; Liu and Han 2018). These methodologies consider de-
3
Professor, Dept. of Civil, Construction and Environmental Engineering, sign criteria based on performance indices that include minimizing
North Carolina State Univ., Raleigh 27695, NC.
energy consumption (Di Nardo et al. 2013), implementation costs
Note. This manuscript was submitted on September 25, 2019; approved
on February 7, 2020; published online on May 11, 2020. Discussion period (Galdiero et al. 2015), and the variability of water demand among
open until October 11, 2020; separate discussions must be submitted for districts (Pesantez et al. 2019).
individual papers. This technical note is part of the Journal of Water Re- To assess different approaches to design DMAs, the Water
sources Planning and Management, © ASCE, ISSN 0733-9496. Distribution Systems Analysis Conference (WDSA) launched the
practical to implement, due to the cost of implementation. Finally, a two, only a handful of solutions were generated as feasible solu-
multiobjective optimization approach was applied for Network2 tions. Closing pipes to determine the DMA entrances affects the
(Zhang et al. 2017) that extended the definition of modularity to hydraulic performance of the model; therefore, an iterative pro-
include similarity among pressures. Trade-offs were explored for cedure was applied to obtain feasible DMA configurations.
three objectives: the number of boundary pipes, pressure uniform- The next step of the procedure evaluated the objectives of each
ity, and water age uniformity. feasible DMA configuration. The number of DMAs, demand sim-
The application of graph partitioning as a multistage procedure ilarity, investment cost, pressure uniformity, and water age were
to cluster and isolate districts, as proposed by Di Nardo et al. (2013), evaluated. Because there are no specific preferences for the objec-
was applied to a large network of 19,402 nodes in Singapore (Sela tives, and in accordance with previous published works (Rahman
Perelman et al. 2015). and Wu 2018; Salomons et al. 2017), we also used demand sim-
One study beyond those reported by the BWNDMA developed ilarity as the criterion to select the final solution.
a multistage procedure to design DMAs for the large E-Town With the solution selected in the rainy season, operational
water network (Pesantez et al. 2019). The approach relaxed the changes were needed for the dry season. Pressure-reducing valves
problem statement developed for the BWNDMA to focus on a were placed to limit high pressures at nodes located close to the
single-objective problem and maximize demand similarity. The pumping stations. PRVs were located at the entrance of each dis-
framework coupled a weighted clustering approach to initialize trict, and additional PRVs were included to control pressures at no-
DMAs, an optimization method to minimize variability in water des as needed. The final solution evaluated with the operational
demand among DMAs, and a heuristic method to fine-tune DMA
changes proposed for the dry season was selected based on the min-
boundaries.
imum investment cost. Fig. 1 shows the flowchart of this iterative
Most of the design approaches that were developed and applied
procedure.
for large water distribution networks used multistage procedures, in
which clustering algorithms were applied first, followed by optimi-
zation techniques and engineering judgment. The computational
complexity of automated approaches may become impractical for
large water networks with multiple objective functions, and divid-
ing procedures into sequential steps is necessary to reduce compu-
tational complexity. Studies by Gilbert et al. (2017) and Salomons
et al. (2017) concluded that care should be used when replacing
engineering judgement with optimization and software tools for
large real-world highly constrained networks, such as E-Town. This
research demonstrates a joint cognitive approach, in which engi-
neering judgement, geospatial analysis, and hydraulic simulation
are coupled within a multistage procedure.
Methodology
Overview
The methodology presented in this manuscript aims to obtain a fea-
sible solution to the complex problem posed by the BWNDMA.
Similar to approaches used by other BWNDMA participants
(Salomons et al. 2017), engineering judgement was used to cor-
rect the hydraulic issues of the original network as part of a pre-
liminary analysis.
A widely used geospatial tool (ArcGIS version 10.4) was
applied to perform geospatial analysis and to select the main
pipes of the system and the zones connected to them to obtain a
layout of the system’s functioning. This analysis was also included
Fig. 1. Flowchart of the proposed methodology.
as a step in the approaches developed by Grayman et al. (2009),
the network and the average and hourly demand at the system eters were stored as possible entrances, which would be evaluated
level. In this step, we identified issues related to water shortages by the hydraulic simulation. An iterative approach based on engi-
or operational problems, which can be addressed in subsequent neering judgement was used to test the hydraulic feasibility of
steps of the analysis. For E-Town, the two drinking water treat- solutions and reassess the entrances to the DMAs using geospatial
ment plants supply a total of 2,850 L/s during the rainy season and analysis, which is similar to the iterative and engineering judgement
2,080 L/s during the dry season. The average systemwide demand approach to obtain a feasible solution taken by Salomons et al.
is 1,925 L/s but the peak hourly demand of the system exceeds the (2017). The hydraulic simulation is described in the following
capacity at 2,953 L/s at 7 a.m. The storage capacity and configu- section.
ration of the existing tanks were evaluated to ensure that adequate
pressures were provided in the system. Next, the topography of
the network was analyzed to identify zones of considerable ele- Hydraulic Simulation
vation differences, which can produce undesirable variable pres- The list of indices corresponding to possible entrances for each sub-
sure distribution and require the installation of pressure-reducing zone was passed to EPANET. Using the MATLAB EPANET tool-
valves to mitigate high pressures. This analysis identified one zone kit, entrances were closed to isolate each subzone and convert it
in the network where elevation ranges from 1.0 to 63.0 m. Meeting into a DMA. EPANET was used to simulate the pipe network once
minimum pressures of 15 m at high elevations and maximum DMAs were determined and pipes were set as closed or opened in
pressures of 60 m at low elevations is difficult in this area of the input file. The toolkit provided an automatic approach to edit
the network, reducing the number of feasible solutions for DMA pipe status, controls for tank levels, and pump settings and to
configurations. execute hundreds of simulations. After each simulation, the con-
Based on the analysis of network operations, components were straints were evaluated, and solutions were discarded if they did
updated as shown in Table 1 to improve the hydraulic performance not meet constraints on water age (less than 60 h) or the number
of the model. Preliminary analysis identified that the hydraulic of nodes that were isolated and not included in any DMA (less than
model convergence was slowed due to several tanks located close 55). Few solutions were feasible. Solutions that violated only pres-
together that emptied and filled at every second during the 168-h sure or tank water level constraints were kept for further analysis.
period of simulation. The model input file was modified to bypass For these solutions, the controls and settings of the valves were
these tanks, and a set of critical pipes and valves surrounding the edited to meet the constraints and improve the performance of
tanks were opened. The original hydraulic model also reported neg- infeasible and feasible solutions. The controls were edited mainly
ative pressures at several nonzero demand nodes at the peak hourly to prevent abrupt fluctuations of the water level of the tanks,
demand. Preliminary analysis identified that closed valves caused whereas existing and new valves (especially PRVs) were manipu-
negative pressures, and valves at Tank 1 and Tank 5 were opened lated to reduce high pressures in some DMAs.
to mitigate these negative pressures. The set of changes shown in An iterative procedure was applied to explore solutions using
Table 1 highlights the importance of applying engineering judgement the partitioned graph as an input to the hydraulic simulator and
to analyze the system and improve operational performance. evaluating each solution using the MATLAB EPANET toolkit until
10 alternative solutions were identified. Finally, the best solution in
terms of demand similarity (DS) was selected and further fine-
Geospatial Analysis
tuned by opening valves and adding more storage to the system.
Geospatial analysis was applied to split the network into clusters As described by Saldarriaga et al. (2019), the DMA configura-
(possible DMAs) using ArcGIS. First, the input file was used to tion designed for the rainy season should also work during the dry
map the graph representing the network into ArcMap, which is the season with a minimum number of operational changes. During
main component of ArcGIS that performs geospatial analysis. The the dry season, there is a substantial decrease of inflow from the
main pipes were identified as the links of the graph with a diameter three water treatment plants, and to cope with this issue, E-Town
of at least 300 mm, which is the same criteria used by Grayman adds two pump stations, as described by Saldarriaga et al. (2016).
et al. (2009), Diao et al. (2013), and Ferrari et al. (2014). Mapping During the rainy season, the DMA configuration is identified based
the system using ArcMap revealed that a few terminal nodes were on the assumption that the pump stations do not operate. During the
connected directly to the main pipes. These nodes remained outside dry season, the pump stations are operational, and the water net-
of the DMA configurations until a set of feasible solutions could be work requires modifications to maintain the pressure of nonzero
achieved. Due to the topography and topology of the system, some demand nodes within the required range of 15–60 m. Due to the
clusters were already isolated and are kept as obligatory DMAs. new flow direction of pipes close to the pump stations, pressure-
Whereas other data could be used to identify DMAs based on reducing valves were placed to limit high pressures at nodes
existing infrastructure (e.g., highways, railroads), pressure zone that violated the pressure constraints. With new water sources,
the required number of entrances for these districts (6, 9, 12, 13, 17).
Additionally, in southern E-Town, pressures were low, and solutions
were explored to maintain the pressure above the minimum (15 m of
water) within the two identified districts (DMAs 1 and 16).
pressure nonuniformity value. Pressure uniformity was one of 2% of nonzero demand nodes. Most solutions reported by the
the objective functions of the BWDNDMA problem, though the BWNDMA also reported low water age at most nodes, with only
approach focused primarily on demand similarity to represent a few nodes as outliers (Saldarriaga et al. 2019).
DMA size.
Operational Changes
Results of the Objective Functions Few modifications were required to use the DMA configuration in
the dry season. The settings of 15 PRVs were modified to decrease
Number of DMAs pressure to a range of 15–60 m of head of water. Ten pipes were
changed from closed to open, and these changes did not count
The objective function values of the solution reported here are pre- as operational changes as specified by the BWNDMA. A few con-
sented in Table 3. Though the number of DMAs defined by the trols were applied for a set of pumps that changed the flow direction
utility as manageable and convenient was 15, our solution used an in the southern part of the system. These operations can be auto-
18-DMA configuration, which produced the lowest value of DS matically defined by the utility to optimize its operations. Please
among districts. The DS value could not be further decreased due refer to Tables 5 and 6 presented in the Appendix section, where
to geographic characteristics that constrained the expansion of cer- we show in detail the required operational changes between
tain districts, and the number of DMAs could not be further seasons. One of the main shortcomings of the BWNDMA, as dis-
reduced. The main objective of the coupled procedure was to obtain cussed by others (Rahman and Wu 2018; Saldarriaga et al. 2019),
a feasible general solution for this complex problem; therefore, was that the solution cannot specify different DMA configurations
single objectives such as DS can be improved at the expense of between seasons. Changes in configurations could benefit both the
other conflicting objectives. Further work explores algorithmic ap- stated goal of the solution (demand similarity) and other metrics,
proaches to obtain a configuration with similar number of users in such as pressure uniformity.
each DMA, based on total water demand (Pesantez et al. 2019). These results are part of a study that was reported specifically for
The results obtained by this procedure are compared with the special edition of the Battle of Water Networks District Metered
the results reported by Saldarriaga et al. (2019). Table 3 shows that Areas in which the authors participated. In the summary reported by
for some objective functions, this solution outperformed some Saldarriaga et al. (2019), this solution is not cited, and the report
solutions published by other participants, specifically in terms of herein completes the set of seven submitted solutions. Similar to
number of required DMAs, pressure uniformity, and investment other approaches completed for the BWNDMA, we did not imple-
cost. A sensitivity analysis of our approach was evaluated based on ment an optimization algorithm and used primarily engineering
the outcomes of different numbers of DMAs (from 15 to 25, as judgement and trial-and-error procedures (Salomons et al. 2017).
explained previously), where the 18-district configuration provided The results reported by this work compare favorably with other so-
the best solution in terms of DS and cost of investment. lutions reported by Saldarriaga et al. (2019), and our approach out-
performs several works that are based on multistage optimization
Size of the Districts and Cost of Intervention techniques. Due to the complexity of solving a real-world highly
constrained problem and the difficulty of obtaining an initial feasible
The size of districts evaluated in terms of demand similarity was solution, this work that was presented for the BWNDMA offers a
below the average of the DS values reported in the BWNDMA valid alternative. For some optimization algorithms, identifying a
overview paper. The investment cost for the DMA configuration feasible solution can be a critical first step, and we include the details
reported here was around US $440,000, or approximately US and the input file of the solution reported here to allow other re-
$40 per node. The cost is related to the number of PRVs needed searchers to build on this work. Please see the supplemental infor-
to identify the reported DMA configuration and to keep the pres- mation for the EPANET input files for rainy and dry seasons.
sure of nonzero demand nodes within the requested range. The cost
reported by this study ranked second compared to the solutions
submitted to the BWNDMA (Saldarriaga et al. 2019). A compre- Conclusions
hensive analysis is required to evaluate the cost-to-benefit ratio of
the solution and is part of ongoing research. The research presented in this paper developed a coupled approach
using engineering judgement to perform geospatial analysis and
hydraulic simulations to design DMAs for a large real-world water
Pressure Uniformity
network. When working with large networks and multiple objec-
Pressure uniformity values were substantially affected by the 55 non- tives, the procedure identified a feasible solution, which was a
zero demand nodes located outside the DMA configuration. Because major challenge posed by the BWNDMA, as highlighted by other