Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Water Resour Manage

https://doi.org/10.1007/s11269-018-2044-z

Improving Consumer Satisfaction in Water Distribution


Networks Through Optimal Use of Auxiliary Tanks
(A Case Study of Kashan City, Iran)

Milad Latifi 1 & Mohammad Amin Gheibi 2 &


Seyed Taghi (Omid) Naeeni 1

Received: 9 January 2018 / Accepted: 9 July 2018


# Springer Nature B.V. 2018

Abstract Temporal and spatial variations in pressure may lead to consumer dissatisfaction
and distrust of water distribution networks when it comes to reliable performance. Pressure
management is a set of programs and operations conducted in water distribution networks to
adjust the pressure. Constructing new auxiliary tanks in proper locations at the best height for
the area they serve minimizes the pressure fluctuations. Additionally, chlorine is often injected
in the reservoirs and tanks to improve the water quality. The goal of this research was to
improve the condition of the network by adding auxiliary tanks with appropriate locations,
heights and chlorine concentration. An optimization model is prepared to optimize consumer
satisfaction, water quality and the relevant costs as objective functions. The performance of the
models are evaluated by a selected case study; and the objectives are optimized in three
scenarios. Using the proposed model in a water distribution network, a trade-off diagram of
reliability and costs is obtained, that lets the decision makers select the proper options
considering the available fund. A new indicator, the consumer satisfaction index, is also
proposed as a way to evaluate the performance of water distribution networks.

Keywords Auxiliary tanks . Consumers’ satisfaction index . Optimization . Reliability . Water


distribution network

* Milad Latifi
milad.latifi@ut.ac.ir

Mohammad Amin Gheibi


m_gheibi@sbu.ac.ir
Seyed Taghi (Omid) Naeeni
stnaeeni@ut.ac.ir

1
School of Civil Engineering, College of Engineering, University of Tehran, Tehran, Iran
2
Department of Water Engineering, Shahid Beheshti University, Tehran, Iran
M. Latifi et al.

1 Introduction

In the last century, water distribution networks (WDN) have been widely developed. Consid-
ering the importance of these systems to society, it is crucial to increase their reliabilities,
improve their service levels and performances, and raise the water quality with respect to
financial constraints.
WDNs are designed to supply sufficient water with proper quality and standard. Neverthe-
less, nonuniform distribution of pressure along the network can cause many problems for
consumers and operators. High pressure results in increased leakage and damage risk for some
parts of the network. However, loss of sufficient pressure in a network can also cause problems
for those who rely on it. Improving water quality is also important, especially when consid-
ering its role in protecting people against diseases. Therefore, supplying high-quality water is a
significant investment in the future. Thus, water tank location, height and chlorine concentra-
tion must be optimized to meet consumer satisfaction. This WDN optimization model also
includes water quality and cost control or affordability. Past research on these concepts are
reviewed in the following.

1.1 Reliability

Reliability is an essential factor in designing and managing water distribution networks. The
reliability of a system is the probability of offering its expected performance without any
failure during a certain period of time under certain conditions (Bazovski 1961). In this
regard, researchers have presented different metrics for measuring the reliability of water
networks.
Gupta and Bhave (1996) defined the volume reliability factor (Rv) as the ratio of the total
available outflow volume to the desired outflow volume at a node for all states in the analysis.
Todini (2000) presented a resilience index (Ir) concept as a measure of network performance
using nodal surplus pressure head, which showed the ability of a network to overcome sudden
failures. Liserra et al. (2014) suggested a metric based on the delivered demand, which
required pressure at nodes and computed total reliability of a WDN by averaging nodal
metric values. Latifi and Naeeni (2016) proposed a model to improve the hydraulic and
quality reliability of urban water reservoirs using auxiliary reservoirs and hourly scheduling
of pumps. Creaco et al. (2016) developed the reliability index presented by Todini (2000) to
generalize resilience and failure indices.
The above metrics are mostly used to evaluate the reliability of the network according to its
mechanical or hydraulic failures. Consumer satisfaction is one of the most important factors in
network operation. So far, no reliability metric has been introduced concerning this crucial
item. In this research, a mathematical function is presented to quantify consumer satisfaction.
Pressure management is an appropriate way to satisfy consumers.

1.2 Pressure Management

Recent research in the field of pressure management has focused mostly on design and
operation of pumps and valves. Farmani et al. (2006) considered pipe rehabilitation decisions,
tank sizing, tank siting and pump operation schedules as the design variables. Their objectives
were maximizing the reliability and water quality of networks as well as minimizing total
costs. Nicolini and Zovatto (2009) used the NSGA-II algorithm to minimize the number of
Improving Consumer Satisfaction in Water Distribution Networks Through...

pressure reducing valves (PRVs) as well as leakage from pipes through controlling the PRVs
and finding their optimal locations. Costa et al. (2010) applied a hybrid genetic algorithm to
minimize total energy generated by the pumps in every operational time with constraints on
node pressures, tank water level and pump flow. Kurek and Ostfeld (2013) optimized water
quality, tank size, and pump operation costs utilizing the NSGA-II algorithm. In their work,
water quality was assessed both by disinfectant residuals and water age. Abunada et al. (2014)
optimized water networks incorporating demand balancing tanks and considering reliability
index constraints. They selected tank volume, pipe diameters and water level in tanks as the
decision variables. Campisano et al. (2016) proposed a methodology to reduce leakage through
improved pressure control in WDNs. Latifi et al. (2018) presented a model to improve the
fuzzy reliability of the WDNs by optimal using of the PRVs.

1.3 Water Quality

Water quality is another aspect of consumer satisfaction. Among different types of disinfection
methods, injecting chlorine and its derivatives are widely used in water networks, due to the
chlorine compounds applicability as well as its abilities in killing microorganisms. If residual
chlorine concentration falls below a certain threshold, disinfection fails and microorganisms
are able to grow, which makes the water pathogenic. Besides, high residual chlorine concen-
trations may change the water taste and produce carcinogenic byproducts, such as trihalo-
methanes (THMs). Therefore, residual chlorine concentration can be used as an appropriate
index to ensure water health and measure consumer satisfaction within a network (Kang and
Lansey 2010). Minimum, optimum and maximum concentrations of residual chlorine have
been proposed in the latest version of BGuidelines for drinking-water quality^ of WHO (2011)
as 0.2, 0.5 and 1.0 (mg/L), respectively.
Tabesh and Dolatkhahi (2006) analyzed water distribution networks by the HDSM ap-
proach, considering residual chlorine, water age and leakage. Kurek and Ostfeld (2014)
presented a model to maximize the water quality and minimize the pumping costs in water
distribution networks considering storage-reliability costs. Babaei et al. (2015) used time-
dependent chlorine injection and pump speed to optimize hydraulic and quality reliabilities. In
the present research, a new water quality evaluating index is introduced and used as an
objective function.
In urban water supply systems, storage tanks and auxiliary tanks are constructed for storing
water, balancing unsteady outflow, supplying and balancing water pressure at nodes, storing
and supplying fire and emergency demand, and adjusting water quality. Hence, tanks are of
great importance in WDNs because of their capabilities in the management of pressure and
quality of water. In some cases, there are low pressure networks, due to increasing population,
demands or network extension. In these cases, using the tanks is one solution, and in some
cases, it is the only solution. In this research, auxiliary tanks were employed for pressure
management. Main objectives of the study are:

– introducing a new metric to assess the consumers’ satisfaction according to hydraulic


performance of the WDN;
– defining a new index to evaluate the quality performance of the WDN;
– constructing a model to improve the hydraulic and quality reliabilities of WDN through
optimal use of auxiliary tanks. The model is able to find the optimal locations and heights
as well as chlorine concentrations of new tanks.
M. Latifi et al.

2 Methodology

A simulation model was constructed to simulate the hydraulic and quality behavior of the
network. New indicators for determining consumer satisfaction and quality reliability of the
network are proposed. An optimization model has been developed to find the optimal location,
height and chlorine concentration of auxiliary tanks. These two models are thus carefully
linked together to meet the goals of the main model.

2.1 Simulating Water Distribution Network

Hydraulic behavior of WDNs has been analyzed through the simulation model. The simulation
model takes the characteristics of the network elements (nodes, pipes, tanks, pumps, etc.) and
calculates the pressure at nodes and velocity in pipes, using continuity equation at nodes and
energy loss equation in pipes. The gradient method developed by Todini and Pilati (1987) was
applied to structure the simulation model using MATLAB R2013 software. This method,
already used in most software, presents robust performance without needing any initial nodal
head.
Classic methods applied for analyzing water distribution networks are based on the
demand-driven simulation method (DDSM). In this approach it is assumed that the entire
demand is supplied to all nodal pressures. The head-driven simulation method (HDSM)
is suggested to simulate the network more realistic. In this approach, the delivered
discharge of each node directly depends on the nodal pressures. In deficient pressure
networks, the results of DDSM and HDSM models are considerably different. The
relationship between nodal pressure and its delivered discharge has been suggested by
Germanopoulos (1985); Wagner et al. (1988); Gupta and Bhave (1996); Tabesh et al.
(2013); and Ciaponi et al. (2015).
Several relationships have been presented for conditions with nodal pressure lower than the
desired pressure. Recently, Ciaponi et al. (2015) proposed Eq. (1a), considering indoor
demands, which is very close to the results presented by Tanyimboh and Templeman
(2010). Eqs. (1b) and (1c) (Tabesh et al. 2013) were used more widely under nodal pressure
conditions above the desired pressure. In this research, a combination of Tabesh et al. (2013)
and Ciaponi et al. (2015) were selected. Accordingly, nodal delivered demand in terms of node
pressure is expressed as follows:
8  
>
> exp −2:570 þ 7:873  H avl
i =H
des
>
> Qi ¼ Qi
avl req
  if i ≤H
H avl des
ðCiaponi et al:; 2015ÞðaÞ
>
> i =H
1 þ exp −2:570 þ 7:873  H avl des
>
>
<  avl  12
Hi
> Qavl
¼ Qa þ Q b if H des ≤H avl
i ≤H
max
ðTabesh et al:; 2013ÞðbÞ
>
>
i
H des
>
>  max  12
>
> Hi
>
: Qavl
i ¼ Qa þ Qb i >H
H avl ðTabesh et al:; 2013ÞðcÞ
max
if
H des
ð1Þ
req max
where Qavl
i = available discharge at node i; Qi = the required discharge at node i; H = the
maximum head, over which the available discharge remains constant in different pressure
des
values; H avl
i = the available pressure head at node i; and H = the minimum pressure head
req
required for supplying the entire demand (Qi ). The value of Hdes depends on the height of
Improving Consumer Satisfaction in Water Distribution Networks Through...

buildings to be served and has been mostly selected between 25 and 35 m. In this research
Hdes = 30 (m); Qa = a volumetric portion of available discharge (not depending on pressure);
and Qb = the head dependent portion of the available discharge. Qa ¼ Qb ¼ 0:5 Qreq i has been
proposed by Tabesh et al. (2013), assuming equal portions for volumetric and head dependent
discharges. Demands are divided into volumetric and head dependent portions by Eq. (1),
becoming more realistic. Figure 1 shows the schematic view of Eq. (1). In this study, the values
of H des
i and H max
i are considered as 30 and 60 m, respectively.
Extended period simulation (EPS) with 1 h time steps is carried out to consider the
variations of demand in WDN. The introduced simulation model, prepared using MATLAB
software, will consider new tanks added to the networks.

2.2 The Proposed Hydraulic Reliability Index

The value of the reliability index should be calculated based on the hydraulic simulation
of the network and by computing the pressure heads in nodes. As mentioned earlier,
several metrics were proposed to quantify the reliability of the water distribution
networks. This work focuses on consumer satisfaction, providing sufficient water with
proper pressure and considering minimum daily pressure fluctuations. Significant pres-
sure changes during a day may lead to dissatisfaction and distrust of consumers toward
the effective performance of the water distribution network. To satisfy consumers,
networks should meet minimum diurnal pressure changes, with an average pressure
value at or close to the desired pressure.
In this work a mathematical function is proposed to quantify the mentioned aspect of
network performance. In this function, the index reaches its maximum value when average
pressure is in the desired range and temporal pressure changes are low. This index was thus
named, the Consumer’s Satisfaction Index (CSI).
In defining CSI, pressure fluctuations as well as pressure surplus and shortages are primary
issues of consumer satisfaction. Considering the pressure fluctuations during the simulation
period, standard deviation (SD) of water distribution network pressure must be calculated and
used to evaluate a fuzzy penalty function (Eq. 3). The standard deviation of pressure in node i,
σi, during simulation is expressed as follows:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uT  2
u
u ∑ H i;t −H i
t
σi ¼ t¼1 ð2Þ
T
where Hi, t = pressure at node i and in state t; and T = total number of simulation states.
The fuzzy penalty function (Eq. 3) is used to normalize SD and determine nodal perfor-
mance as follows:
8 σi
< 1− if 0≤σi < 10
PIDi ¼ 10 ð3Þ
:
0:05 if σi ≥10
where PIDi = the SD-based component of performance index at node i, Fig. 2a.
The effects of pressure surplus or shortage on consumer satisfaction can be obtained by
calculating the average pressure during the simulation period. This component is based on the
M. Latifi et al.

Available Discharge at Node


Qreq

Hdes = 30 m H max = 60 m
Pressure Head at Node (m)
Fig. 1 Relationship between nodal pressure and the demand provided at nodes

1.00

0.75
PID

0.50

0.25

0.05
0.00
10 20

(a)

1.00

0.75

0.50
PIH

0.25

0.00
H min= 5 m Hdes = 35 m Hmax = 50 m

Pressure Head (m)


(b)
Fig. 2 Fuzzy penalty function to consider the a standard deviation of pressure; b deviation of average pressure
from desired pressure
Improving Consumer Satisfaction in Water Distribution Networks Through...

deviations between average pressure and desired pressure and is calculated using a fuzzy
penalty function (Eq. 4), as derived by Tabesh et al. (2013) as follows:

8
>
> 0 if H i ≤5
>
> !0:51
>
>
>
> H i −5
>
< if 5 < H i ≤35
30 ð4Þ
PIH i ¼
>
>
> H i −35
>
>
> 1− if 35 < H i ≤50
>
> 30
>
:
0:25 if H i > 50

where H i = average pressure in the simulation period at node i; and PIHi = the fuzzy utility
function of average pressure at node i. The schematic view of Eq. (4) is shown in Fig. 2b.
Finally, utility components for each node are geometrically combined to obtain the nodal
consumer satisfaction index.

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
CSI i ¼ PIH i  PIDi ð5Þ

where CSIi = the consumer satisfaction index of node i. The proposed indicator is a normal and
dimensionless one that gives values in the range [0, 1]. Thus, the total network CSI, is
summarized as follows:
NN
∑ Qreq
i  CSI i
i¼1
f 1 ¼ CSI ¼ NN
ð6Þ
∑ Qreq
i
i¼1

where Qreq
i = the average diurnal demand of node i; and NN = total number of nodes.

2.3 Water Quality Simulation

The discrete volume element method (DVEM) was presented by Rossman et al. (1993). This
dynamic implicit method is used for quality analysis of water networks. In DVEM, each pipe
is discrete to some volume elements; and the concentration of substances in each element is
computed with respect to the concentrations of upstream and downstream elements. Assuming
a complete mix process at upstream elements and neglecting detention time at nodes, the
concentration at node i is obtained as follows:
βk  
∑ Q jC j L j; t
j∈fk g
C i ði; jÞ ¼ C k ðt Þ ¼ βk
ð7Þ
∑ Qj
j∈fk g

where Ck = concentration at node k; {k} = all pipes connected to node k; and Qj and Lj =
discharge and length of pipe j, respectively. Finally, Cj(Lj, t) = the concentration in pipe j, at
element Lj and at time t.
M. Latifi et al.

Regarding the storage tanks with entering pipes (j), the concentration is calculated as
follows:

1 h   i
C T ðt þ τ Þ ¼ C j L j ; t Q j τ þ V T ðtÞ C T ðtÞ ð8Þ
V T ðt Þ þ Q j τ

where CT and VT = fully mixed concentration and volume of the tank, respectively; and τ =
time interval. It is assumed that an outgoing pipe (j) carries the fully mixed tank concentration,
which is expressed as follows:
C j ð0; t þ τ Þ ¼ C T ðtÞ ð9Þ

A pressure dependent quality analyzer was developed, integrating HDSM and DVEM. A
model was developed for quality analysis of the network based on the Lagrangian time-driven
method (TDM). This approach is robust and applicable for computing the changes of chlorine
concentration in quality time steps.

2.4 The Proposed Quality Reliability Index

In this research, quality reliability index is derived from Coelho (1996), which is based on the
residual chlorine in consumption nodes. Coelho (1996) considered five levels of serviceability
to evaluate the quality performance of water networks: 1) no service; 2) unacceptable service;
3) acceptable service; 4) good service; and 5) excellent service, with quality index values of 0,
0.25, 0.5, 0.75 and 1.0, respectively.
Considering the utility index, proposed for estimating the quality reliability, if the values of
residual chlorine are: below 0.1 (mg/L), 0.1–0.5 (mg/L), 0.5–0.8 (mg/L) and 0.8–1.1 (mg/L),
the water distribution network has no service, excellent service, acceptable service or unac-
ceptable service statuses, respectively, and the relevant index values are 0, 1, 0.5 and 0.25,
respectively. The increase of residual chlorine may result in generating harmful byproducts,
which can consequently decrease the reliability index. The quality level is in unacceptable
status when chlorine concentrations are over 1.1 (mg/L). Figure 3 shows the utility function of
nodal quality reliability.
The quality performance index of different nodes is generalized to the entire network as
follows (Coelho, 1996):
T NN
∑ ∑ Qreq
i;t :QRI i;t
t¼1 i¼1
f 2 ¼ NQRI ¼ T NN
ð10Þ
∑ ∑ Qreq
i;t
t¼1 i¼1

where NQRI = the network quality reliability Index; QRIi, t = the quality reliability index of
node i and time t; NN = the total number of nodes; and T = total number of simulation time
steps.

2.5 Tank Construction Costs

The third objective of this research is to minimize the construction costs of auxiliary tanks.
These costs depend on issues such as material, shape and volume of tank, type of bottom soil
Improving Consumer Satisfaction in Water Distribution Networks Through...

1.00

0.75

QRI
0.50

0.25

0.00
0.2 0.4 0.6 0.8 1.0 1.2
C
Fig. 3 Utility function of nodal quality reliability

supporting the tank, utility and installation costs, etc. Eq. (11) estimates costs based on
investigations and experiences. It also provides some simplifications, and is defined as follows:
Nres  
f 3 ¼ Cost ¼ ∑ a  V r þ b  H Res;r 2 þ e ð11Þ
r¼1

where Cost = the objective function for estimating the costs of auxiliary tanks construction; Vr
= the volume of tank r (m3); HRes, r = operational height of tank r (m); a = the total costs
proportional to 1 (m3) increase in tank volume; b = total costs proportional to increase in tank
height; and e = the miscellaneous costs of constructing new tanks.
In the case study conducted in this research, a, b and e are assumed as 1 million Rials per
cubic meter, 5 million Rials per square meter and zero, respectively.

2.6 Optimization Method

An optimization model was prepared to find the best location and height of auxiliary tanks.
Researchers have presented different methods for optimizing water distribution networks.
Considering the complexities and nonlinearity of the problem as well as the simultaneous
presence of integer and real decision variables in optimization, a genetic algorithm (GA) was
selected for optimizing the objective functions. A nondominated sorting genetic algorithm
(NSGA-II), presented by Deb et al. (2002), was applied for a two-objective optimization of the
water distribution network. The optimization model was also prepared using MATLAB to
meet the optimal solution faster.
Overall, the models used in the study are:

– Simulation model; consists of hydraulic and quality simulation subroutines,


– Optimization model; consists of selection, crossover and mutation subroutines.

Together, the simulation and optimization models yielded the optimum location and
sufficient operational height of the auxiliary tanks as well as the optimal chlorine concentration
and include the consumer’s satisfaction index (CSI), the network quality reliability index
(NQRI) and tank construction costs. Figure 4 presents a schematic view of constructed models
and their connections.
M. Latifi et al.

Input network data:


Start
Location of nodes, pipes, pumps,
reservoirs, PRVs, ...
Optimization model Length, diameter and roughness of pipes
Nodal demands
Generate random initial population Hourly demand coefficients
(Decision variables)
Simulation model

Evaluate the fitness of population Apply the new reservoirs


using simulation model

Generate new population Hydraulic simulation of the network


by Gradient method

Apply the operators of GA: Correct available nodal discharges


Selection, Crossover, Mutation based on nodal heads (HDSM)

No Compute the chlorine concentration


in each node and tank (DVEM)
Yes Optimization
Stop criteria met?
No Termination
criteria met?

Yes

Calculate the objective functions:


(1) CSI ; (2) NQRI ; and (3) Costs

Fig. 4 Schematic views of simulation and optimization models

3 Case Study

To evaluate the performance of the model, a water distribution network was selected in the
central zone of Kashan City, Iran, covering an area of 1950 (ha). According to the water
consumption evaluated in this city, a maximum discharge of water consumption is 931 l per
second. The considered network has 47.4 km (29.5 miles) of main pipeline made of steel,
polyethylene and asbestos-cement. The network was modeled using 34 nodes and 49 pipes
(with 63–400 mm diameters) presented schematically in Fig. 5. A preliminary analysis showed
that the initial water level had an insignificant effect on the results, so the tanks were assumed
half-full at the start of the simulations.
The network has low average pressure and does not supply total consumer demand. The
goal of this research was to improve the condition of the network by adding auxiliary tanks
with appropriate locations and heights, considering the construction costs and supplying
consumer demands.

4 Optimization Scenarios and Results

The authors selected three different scenarios to evaluate the performance of added auxiliary
tanks and their effects on research objectives.
Improving Consumer Satisfaction in Water Distribution Networks Through...

Fig. 5 The schematic view of a part of water distribution network of Kashan city, Iran (courtesy of Google
Earth)

4.1 Scenario 1

The objective functions, considered in scenario 1 were maximizing the consumer satisfaction
index (CSI) and minimizing the construction costs of the auxiliary tanks (Cost). The optimi-
zation constraints are: governing hydraulic equations (continuity and energy equations, im-
plicitly included in the simulation model) and the limitations in adding auxiliary tanks to the
network (maximum number of tanks and minimum and maximum height of tanks). The
decision variables are number, locations and heights of auxiliary tanks. The optimization
problem can be written as follows:
 
Maximize f 1 Pi;t ¼ CSI  i ¼ 1; …; NN t ¼ 1; …; 24
Minimize f 3 H Res;r ; V r ¼ Cost
ð12Þ
Subject to : H min < H Res;r < H max
NTank ¼ f 1; …; NTank max g
where Pi, t = pressure head at node i and at time t; Hmin and Hmax = minimum and maximum
possible heights of tank, respectively, here assumed as 10 and 50 m, respectively; HRes, r and Vr
= operational height and volume of tank r, respectively; NTank = number of added tanks in the
optimization; and NTankmax = maximum number of augmentable tanks, here assumed as five.
The considered values of Hmin, Hmax and NTankmax are assumed according to the authors
experiences. Adding a new tank with a height lower than 10 m will not impose sufficient
pressure in the adjacent nodes. Constructing a tank with a height higher than 50 m made it
more expensive and not affordable.
In this research, the variables of a genetic algorithm (GA) were selected considering the size
of the problem as well as the results of previous experiences. The population size represented
one generation and was equal to 300. This GA model was developed to allow a multi-objective
optimization model to use integer and real types of variables simultaneously, considering the
M. Latifi et al.

existence of integer decision variables. Crossover probability and mutation rate were 0.8 and
0.05, respectively. The tournament method was used to select the parents. The scenarios were
run in a core i7 computer with 6 GB RAM to gain optimum responses after about 40′000
fitness evaluations in about 24 h.
Several solutions were obtained by executing the program. Considering the multi-objective
optimization of the problem, a trade-off (Pareto) diagram was plotted to present the results, as
shown in Fig. 6, which shows that the cost of constructing new tanks increases with an
increasing consumer satisfaction index. Four solutions were selected and used for assessing the
model performance. These responses are shown in Fig. 6 by A, B, C and D in the Pareto curve.
The characteristics of these options are presented in Table 1.
In option A, the network (without any auxiliary tanks) bears the highest fluctuation of
pressure, falling to 10 m in some hours, and offers minimum value of CSI (0.568). The
pressure oscillation is reduced in option B by adding a tank at the height of 20.8 m at
node 18. However, the network average pressure increases only by about 1 m and
reaches to 17.8 m, much lower than the desired pressure. In this option CSI = 0.80.
The pressure fluctuation becomes more reduced in option C by inserting two tanks at
nodes 17 and 19. In this case, the average network pressure is closer to the desired
pressure and CSI is about 0.9. In option D, five tanks are added to nodes 9, 15, 19, 23
and 25, and the fluctuation of pressure is minimized by spending maximum costs;
moreover, the average pressure of the network is slightly increased. In option D CSI =
0.945. Figure 7a presents the temporal variations of average pressure in the network
designed for A, B, C and D options.
Figure 7b illustrates the hourly variations of pressure at node 28. Thus, the pressure
variation is high in option A; and the average pressure is low, with a nodal CSI = 0.60. In
option B, the fluctuation of pressure is limited by adding a tank; however, the average pressure
is still low, leading to a nodal CSI = 0.88. In option C, no significant change is observed in the
fluctuation of pressure by adding three tanks; while average pressure increases considerably
(CSI = 0.87). In option D, the pressure fluctuation is controlled and average pressure aligns
more closely with the desired pressure; thus, the nodal CSI = 0.95.

0.95

0.9
D
0.85 C

0.8
CSI

B
0.75

0.7

0.65

0.6
A
0.55
0 5000 10000 15000 20000 25000 30000
Cost (Million Rials)
Fig. 6 Trade-off diagram for two objectives, maximizing CSI and minimizing Costs
Improving Consumer Satisfaction in Water Distribution Networks Through...

Table 1 The characteristics of selected options in Scenario 1

Option Locations of Height of Volumes of Total costs CSI


auxiliary tanks (nodes) auxiliary tanks (m) auxiliary tanks (m3) (Million Rials)

B 18 20.8 2300.4 4463.6 0.800


C 17 31.4 479.1 10867.7 0.900
19 26.4 1974.0
D 9 29.2 2417.8 25077.7 0.945
15 30.9 329.1
19 28.2 412.2
23 29.0 420.0
25 24.2 1352.0

30

25

20
Pressure Head (m)

15

10
Option (D) - CSI=0.945
Option (C) - CSI=0.900
5
Option (B) - CSI=0.800
Option (A) - CSI=0.568
0
0 6 12 18 24
Time (hr)
(a)
45

40

35

30
Pressure Head (m)

25

20

15 Option (A) - CSI=0.60


Option (B) - CSI=0.88
10 Option (C) - CSI=0.87
Option (D) - CSI=0.95
5

0
0 6 12 18 24
Time (hr)
(b)
Fig. 7 a Average pressure of the network; b Pressure variations at node 28; in options A, B, C and D
M. Latifi et al.

4.2 Scenario 2

In this scenario, the objective functions are maximizing the network quality reliability index
(NQRI) and minimizing the construction costs of auxiliary tanks (Cost). The optimization
constraints are: governing hydraulic equations (like scenario 1), limitations in adding the tanks
to the network (like scenario 1) and limitations on chlorine concentration in the tanks
(minimum and maximum chlorine concentrations). The decision variables are chlorine con-
centrations, as well as number, locations and height of new tanks. The optimization problem
can be written as follows:
Maximize f 2ðC i Þ ¼ NQRI
 i ¼ 1; …; NN
Minimize f 3 H Res;r ; V r ¼ Cost
ð13Þ
Subject to : H min < H Res;r < H max
NTank ¼ f 1; …; NTank max g
C min < C Res;r < C max

where Ci = chlorine concentration at node i; CRes, r = chlorine concentration at the new tank r;
Cmin and Cmax = minimum and maximum allowable chlorine concentrations, assumed as 0.1
and 1.4 (mg/L), respectively.
In this scenario, the optimization results are presented by a Pareto diagram, as shown in
Fig. 8. According to this figure, the cost of tank construction increases when the quality
reliability index increases. Four optimization solutions were selected out of the obtained
responses and used to evaluate the performance of model, shown as A, B, C and D in Fig.
8. The characteristics of the selected options are presented in Table 2.
In option A, no tank is added to the network; the quality reliability index is low (NQRI =
0.577). This value increases to 0.68 by adding the first tank at minimum cost. Appending more
tanks result in higher quality reliability. In option B, NQRI increases to 0.8 by inserting one
new tank, spending 3566 million Rials ($95,093.33). In option C, NQRI increases to almost

0.95

0.9
D
0.85 C
NQRI

0.8

0.75 B

0.7

0.65

0.6
A
0.55
0 2000 4000 6000 8000 10000 12000

Cost (Million Rials)


Fig. 8 Trade-off diagram for the objective functions, maximizing NQRI and minimizing Costs
Improving Consumer Satisfaction in Water Distribution Networks Through...

Table 2 The characteristics of selected options in Scenario 2

Option Locations Height of Volumes of Concentration Total costs NQRI


of auxiliary auxiliary auxiliary of chlorine in (Million Rials)
tanks (nodes) tanks (m) tanks (m3) auxiliary tanks
(mg/L)

B 12 13.8 2614.1 1.023 3566.3 0.802


C 2 14.8 715.7 0.851 6501.4 0.893
12 13.7 1345.0 0.688
29 14.7 1326.6 0.714
D 3 19.0 489.3 0.775 11343.7 0.931
12 19.1 1104.8 0.753
20 16.2 726.3 0.718
28 19.2 2238.9 0.620

0.9 by adding 3 new tanks and spending 6501 million Rials ($173,360). In option D, NQRI
increases to 0.93 by adding of 4 tanks in nodes 3, 12, 20 and 28, and spending 11,343 million
Rials ($302,480).
Figure 9 presents the temporal variations of average chlorine concentrations in options
A, B, C and D. Option A with only one tank cannot supply sufficient chlorine for the
network. In this option, the chlorine concentration does not exceed 0.34 (mg/L). Besides,
the concentration of chlorine is low in the nodes at early hours of the day due to low
consumption as well as low flow velocity. In options B and C, one and three tanks are added,
respectively. In these options, the chlorine injection dosages in tanks are set so that the
residual chlorine concentrations in nodes are higher than the desired values. Although, the
quality reliability increases, a penalty is applied to the reliability because of excessive
concentrations of chlorine in the nodes. In option D, NQRI is maximized by adding 4 new
tanks and optimizing the chlorine dosage in tanks. In this option the average chlorine
concentration reaches the desired value (0.5 mg/L).

0.8

0.7
Chlorine Concentration (mg/L)

0.6

0.5

0.4

0.3

Option (D) - NQRI=0.931


0.2
Option (C) - NQRI=0.893
0.1 Option (B) - NQRI=0.802
Option (A) - NQRI=0.577
0
0 6 12 18 24
Time (hr)
Fig. 9 Average chlorine concentration in options A, B, C and D
M. Latifi et al.

1.00

0.95

0.90

0.85 C
D
0.80 B

0.75

0.70

0.65

0.60
A
0.55

0.50
0 5000 10000 15000 20000 25000
Cost (Million Rials)
Fig. 10 Trade-off between integrated reliability and total costs

4.3 Scenario 3

In this scenario, the integration of both the consumer satisfaction index (CSI) and network
quality reliability index (NQRI) is used as the first objective function to consider both
hydraulic and quality aspects of WDN performance. The second objective is to minimize total
construction costs of the new tanks. Optimization constraints and decision variables are like
those of scenario 2. The optimization model can be written as follows:
  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Maximize f 4 Pi;t ; C i ¼ IR ¼ CSI  NQRI i ¼ 1; …; NN t ¼ 1; …; 24
Minimize f 3 H Res;r ; V r ¼ Cost
ð14Þ
Subject to : H min < H Res;r < H max
NTank ¼ f 1; …; NTank max g
C min < C Res;r < C max

where IR is the integrated reliability of the network. The results of this scenario are
presented in Fig. 10 in the form of a Pareto diagram. Characteristics of the four selected

Table 3 The characteristics of selected options in Scenario 3

Option Locations Height of Volumes Chlorine Total Integrated


of auxiliary auxiliary of auxiliary concentration costs reliability
tanks (nodes) tanks tanks (m3) in auxiliary (Million
tanks (mg/L) Rials)

B 9 17.1 913.2 0.856 5967.20 0.857


17 23.9 735.9 0.793
C 9 18.4 810.3 0.877 11744.2 0.888
20 22.7 2583.0 0.790
23 24.9 981.6 0.713
D 14 29.3 770.9 0.774 23236.4 0.900
18 31.5 1146.0 0.823
20 23.8 2329.2 0.794
23 29.6 2523.6 0.810
Table 4 Summary of the scenarios

Scenario Objective functions Subject to Decision variables Importance of the scenario


number
 
1 Maximize f 1 Pi;t ¼ CSI i ¼ 1; …; NN t ¼ 1; …; H min < H Res;r < H max NTank Number, locations and heights Only optimizing the hydraulic
 
24 Minimize f 3 H Res;r ; V r ¼ Cost ¼ f 1; …; NTank max g of auxiliary tanks performance of the network
and costs of new tanks
2 Maximize f 2 ðCi Þ ¼ NQRI i ¼ 1; …; NN H min < H Res;r < H max NTank Chlorine concentrations as well Only optimizing the quality
Minimize f 3 H Res;r ; V r ¼ Cost ¼ f 1; …; NTank max g C min as number, locations and performance of the network
< C Res;r < C max heights of auxiliary tanks and costs of new tanks
  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 Maximize f 4 Pi;t ; C i ¼ IR ¼ CSI  NQRI i ¼ H min < H Res;r < H max NTank Chlorine concentrations as well Simultaneously optimizing the
1; …; NN t ¼ 1; …; 24 Minimize f 3 ¼ f 1; …; NTank max g C min as number, locations and hydraulic and quality
  < C Res;r < C max heights of auxiliary tanks performance of the network
H Res;r ; V r ¼ Cost
Improving Consumer Satisfaction in Water Distribution Networks Through...

as well as costs of new tanks


M. Latifi et al.

options are presented in Table 3. Option A is related to the initial status of the network with
only one existing tank. In this option, no new tank is added and the minimum integrated
reliability is obtained. In options B and C, two and three new tanks are added, respectively.
In option B, 5967.2 million Rials ($159,125.33) are spent and IR = 0.857; in option C, 11744
million Rials ($313,173.33) are spent and IR = 0.888. In option D, four new tanks are added
in nodes 14, 18, 20 and 23, and 23,236 million Rials ($619,626) are spent while IR = 0.90. It
can be concluded that in high IRs, the integrated reliability is slightly improved despite rapid
increase of costs.
In this final scenario, all three objectives of the research were considered. Accordingly, the
results obtained from performing the model in the third scenario are appropriate considering
the consumers’ and quality reliabilities. However, the costs of installing the tanks are minimum
for meeting the mentioned reliabilities. Table 4 shows the summary of the three built scenarios.

5 Conclusions

Providing desired pressure at nodes is crucial for supplying all water distribution network
demands. These networks usually provide adequate pressure. However, pressure deficiency
may occur throughout all or in parts of the network due to the network extension or a sudden
increase of water consumption. An approach to managing the pressure in such networks is to
design and construct the auxiliary tanks. Besides providing sufficient pressure, adding new
tanks can also balance the network pressure. Temporal and spatial variations in pressure may
lead to consumer dissatisfaction and distrust about the appropriate performance of a network.
Installing new tanks in proper locations and at optimal heights provides the opportunity to
minimize pressure fluctuations. On the other hand, chlorine is often injected into reservoirs and
tanks. Although it is costly, installing new tanks is effective in providing the desired chlorine
concentration in nodes and upgrading the network quality. Therefore, it is crucial to optimize
the locations, heights and chlorine dosages of auxiliary tanks with respect to the relevant tank
construction costs.
In this research, a hydraulic and quality water network simulation model was connected to
an optimization model. The simulation model was used to analyze the network and compute
the values of objective functions. A new indicator (CSI) is also proposed to evaluate the
consumer satisfaction. Obtaining lower diurnal pressure oscillations and closer average to
desired pressure readings cause a higher consumer satisfaction index. Increasing this indicator
in a WDN ensures pressure balance and reduction in pressure fluctuations making it a proper
metric for evaluating the consumers’ satisfaction. A function is also defined for quantifying
quality reliability based on chlorine residual concentrations in the nodes, NQRI. A high level of
this indicator guarantees the water safety and meeting of health standards, due to existing
sufficient residual chlorine. The objective function of minimizing the costs of tanks installation
has been presented based on the available experiences. The location and height of tanks as well
as the concentration of chlorine in tanks are selected in such a way to optimize the objective
functions.
A case study network has been selected to evaluate the performance of models. Accord-
ingly, the objective functions have been optimized in 3 scenarios. In scenario 1, the reliability
index is optimized according to the consumer’s satisfaction and costs, as the objective
functions, suggesting several options. By spending sufficient costs and installing 1–4 tanks
in the network, daily pressure fluctuations are significantly reduced; and the average pressure
Improving Consumer Satisfaction in Water Distribution Networks Through...

becomes close to the desired pressure. This fact has been assessed in a sample node as well.
Scenario 1 is applicable if the consumer’s satisfaction is the only objective. In scenario 2 the
locations, heights and chlorine dosages of new tanks are selected in such a way that quality
reliability index is maximized and total costs are minimized. In this scenario several options
have been presented in the output of model. Based on the solutions, higher quality reliability is
obtained with spending more costs. When the quality reliability of the network is the only
objective, Scenario 2 is proper. In scenario 3, two indices, CSI and network quality reliability
index (NQRI), were integrated into a new index and optimized simultaneously with the Cost
objective function. Scenario 3 can be applied if both CSI and NQRI are important. In this study,
the proposed model was applied for assessing the considered water distribution network. Then,
the trade-off diagrams of reliability and costs were plotted according to the obtained results.
Moreover, optimal conditions cannot be achieved unless the decision makers select the
appropriate options with respect to the needed reliability and available budget. This study
indicated that inserting auxiliary tanks in proper locations and heights can substantially
improve the hydraulic reliability in pressure-deficient WDNs. Adjusting the chlorine concen-
tration in new tanks increases the quality in the WDN. Also, the proposed indicators are able to
reflect the hydraulic and quality performance of the water distribution networks.

Compliance with Ethical Standards

Conflict of Interest No conflict of interest.

References

Abunada M, Trifunovic N, Kennedy M, Babel M (2014) Optimization and reliability assessment of water
distribution networks incorporating demand balancing tanks. Proc., 12th Int. Conf. Computing and Control
for the Water Industry, CCWI2013, Perugia, Italy, Procedia Engineering, 70:4–13. https://doi.org/10.1016/j.
proeng.2014.02.002
Babaei N, Tabesh M, Nazif S (2015) Optimum reliable operation of water distribution networks by minimising
energy cost and chlorine dosage. Water SA 41(1):149–156. https://doi.org/10.4314/wsa.v41i1.18
Bazovski I (1961) Reliability theory and practice. Prentice-Hall Inc., Englewood Cliffs
Campisano A, Modica C, Reitano S, Ugarelli R, Bagherian S (2016) Field-oriented methodology for real-time
pressure control to reduce leakage in water distribution networks. J Water Resour Plan Manag. https://doi.
org/10.1061/(ASCE)WR.1943-5452.0000697
Ciaponi C, Franchioli L, Murari E, Papiri S (2015) Procedure for defining a pressure-outflow relationship
regarding indoor demands in pressure-driven analysis of water distribution networks. Water Resour Manag
29(3):817–832. https://doi.org/10.1007/s11269-014-0845-2
Coelho ST (1996) Performance assessment in water supply and distribution. PhD Thesis, Civil & Offshore
Engineering Department, Heriot-Watt University, Edinburg, UK, Cited in: Tabesh and Dolatkhahi (2006)
Costa L, Ramos H, Castro M (2010) Hybrid genetic algorithms with advanced search for optimized operation of
water supply networks. Proc., 10th Int. Conf. Computing and Control in the Water Industry, CCWI2009, p
621–626
Creaco E, Franchini M, Todini E (2016) Generalized resilience and failure indices for use with pressure-driven
modeling and leakage. J Water Resour Plan Manag. https://doi.org/10.1061/(ASCE)WR.1943-
5452.0000656
Deb K, Pratap A, Agrawal S, Meyarivan T (2002) A fast and elitist multiobjective genetic algorithm: NSGA-II.
IEEE Trans Evol Comput 6(2):182–197. https://doi.org/10.1109/4235.996017
Farmani R, Walters G, Savic D (2006) Evolutionary multi-objective optimization of the design and operation of
water distribution network: Total cost vs. reliability vs. water quality. J Hydroinf 8(3):165–179. https://doi.
org/10.2166/hydro.2006.019
M. Latifi et al.

Germanopoulos G (1985) A technical note on the inclusion of pressure dependent demand and leakage terms in
water supply network models. Civ Eng Syst 2(3):171–179. https://doi.org/10.1080/02630258508970401
Gupta R, Bhave PR (1996) Reliability-based design of water-distribution systems. J Environ Eng 1(51):51–54.
https://doi.org/10.1061/(ASCE)0733-9372(1996)122
Kang D, Lansey K (2010) Real-time optimal valve operation and booster disinfection for water quality in water
distribution systems. J Water Resour Plan Manage:463–473. https://doi.org/10.1061/(ASCE)WR.1943-
5452.0000056
Kurek W, Ostfeld A (2013) Multi-objective optimization of water quality, pumps operation, and storage sizing of
water distribution systems. J Environ Manag 115:189–197. https://doi.org/10.1016/j.jenvman.2012.11.030
Kurek W, Ostfeld A (2014) Multi-objective water distribution systems control of pumping cost, water quality,
and storage-reliability constraints. J Water Resour Plan Manag. https://doi.org/10.1061/(ASCE)WR.1943-
5452.0000309
Latifi M, Naeeni STO (2016) Improvement of the hydraulic and quality reliability of urban water reservoirs using
auxiliary reservoirs and hourly scheduling of pumps. Proc., 14th Conf. on Computing and Control for the
Water Industry (CCWI2016), Amsterdam, Netherlands
Latifi M, Naeeni STO, Gheibi MA (2018) Upgrading the reliability of water distribution networks through
optimal use of pressure-reducing valves. J Water Resour Plan Manag. https://doi.org/10.1061
/(ASCE)WR.1943-5452.0000866
Liserra T, Maglionico M, Ciriello V, Di Federico V (2014) Evaluation of reliability indicators for WDNs with
demand-driven and pressure-driven models. J Water Resour Manag 28(5):1201–1217. https://doi.
org/10.1007/s11269-014-0522-5
Nicolini M, Zovatto L (2009) Optimal location and control of pressure reducing valves in water networks. J
Water Resour Plan Manag 3(178):178–187. https://doi.org/10.1061/(ASCE)0733-9496(2009)135
Rossman LA, Boulos PF, Altman T (1993) Discrete volume element method for network water-quality models. J
Water Resour Plan Manag 5(505):505–517. https://doi.org/10.1061/(ASCE)0733-9496(1993)119
Tabesh M, Dolatkhahi A (2006) Effects of pressure dependent analysis on quality performance assessment of
water distribution networks. Iranian J Sci Technol, Trans B, Engineering 30(1):119–128
Tabesh M, Shirzad A, Arefkhani V, Mani A (2013) A comparative study between the modified and available
demand driven based models for head driven analysis of water distribution networks. Urban Water J 11(3):
221–230. https://doi.org/10.1080/1573062X.2013.783084
Tanyimboh TT, Templeman AB (2010) Seamless pressure-deficient water distribution system model. J Water
Manage, ICE 163(8):389–396. https://doi.org/10.1680/wama.900013
Todini E (2000) Looped water distribution networks design using a resilience index based heuristic approach.
Urban Water J 2(2):115–122. https://doi.org/10.1016/S1462-0758(00)00049-2
Todini E, Pilati S (1987) A gradient algorithm for the analysis of pipe network. Proc., Int. Conf. computer
applications for water supply and distribution, Leicester Polytechnic, UK
Wagner JM, Shamir U, Marks DH (1988) Water distribution reliability: simulation methods. J Water Resour Plan
Manag 3(276):276–294. https://doi.org/10.1061/(ASCE)0733-9496(1988)114
World Health Organization (WHO) (2011) Guidelines for drinking-water quality. 4th edn

You might also like