Discussions and Closures

made an elaborate study to address this, which is to be appreciated,

Discussion of “Optimal Operation of Water
because this kind of work is more useful to practicing engineers and
Distribution Networks under Water utility managers. The discussers found a few shortcomings in the ef-
Shortage Considering Water Quality” by ficiency criteria explained by the authors. This paper elaborates on
Mohammad Solgi, Omid Bozorg-Haddad, those shortcomings and offers alternate better methods.
Samaneh Seifollahi-Aghmiuni, In Eq. (25) of the original paper, in the numerator, a value of 1 is
Parisa Ghasemi-Abiazani, and taken when there is a failure time step following a success time
Hugo A. Loáiciga step. Success is taken when the supply for the whole network is
more than or equal to the desired demand (β × De) of the network
DOI: 10.1061/(ASCE)PS.1949-1204.0000233 in a given time step. This indicator, however, may be misleading.
Assume a case in which there can be a few nodes with excess sup-
Downloaded from ascelibrary.org by 165.50.249.73 on 11/23/17. Copyright ASCE. For personal use only; all rights reserved.

T. R. Neelakantan 1 and C. R. Suribabu 2
1
Senior Professor, Dept. of Civil Engineering, Kalasalingam Academy of
Research and Education, Kalasalingam Univ., Krishnankoil, Tamil
Nadu 626126, India; formerly, Associate Dean and Professor,
School of Civil Engineering, SASTRA Univ., Thanjavur, Tamil Nadu
613401, India (corresponding author). ORCID: https://orcid.org/
0000-0001-5721-3398. E-mail: neelakantan@klu.ac.in
2
Professor, Centre for Advanced Research in Environment, School of Civil
Engineering, SASTRA Univ., Thanjavur, Tamil Nadu 613401, India.
Managing chlorine concentration as well as pressure in water distri-
bution networks during a water shortage resulting from a poor quality
water source is not addressed much in the literature. The authors
ply (more than the desired demand, which is not preferred) and a
few other nodes with lesser supply in a given time step. Further
assume that theP total excessPsupply is more than the total deficiency.
In this case, NCi S
i¼1 i;h > NCi
i¼1 β × Dei;h , which indicates that the
network is in success state. Though there are failures at the node
level, it is considered as a success at the network level. There can be
critical nodes in which the demands are relatively large compared
to other nodes in the network. If one such node fails and another
such node gets excess supply, the number of consumers affected
will be large. To avoid such confusion, supply to a network can
be considered successful when supplies to all the nodes are success-
ful. In other words, even a single failure at any one node is
to be considered as a failure at the network level. Hence,
Eq. (25) of original paper can be modified as

(
PNh 1; if ðSi;h < β × Dei;h ; and Si;h−1 ≥ β × Dei;h−1 Þ; for any i ∈ ð1 to NCiÞ
h¼1
0; otherwise
γ β ¼ 100 × ( ð1Þ
PNh 1; if ðSi;h < β × Dei;h Þ; for any i ∈ ð1 to NCiÞ
h¼1
0; otherwise

In this revised form of the equation, the numerator (excepting Consider a network with two nodes with the data given in
100) gives the number of failure events (an event can be one or Table 1, and consider the two cases of results shown in Table 1.
more continuous time steps), and the denominator gives the total In both cases, both Nodes 1 and 2 experienced failure in five
number of time steps in which the network is in a failure state. time steps each; however, the number of failure events varied.
Hence, the mean duration of a failure event can now be written as Node 2 is associated with a larger demand compared to Node 1, and
(100=γ β ) time steps. hence, higher resilience of Node 2 is better. In Case 1, Node 2
In Eq. (26) of the original paper, the ratio of the number of times experienced failure in five time steps and in two spells, whereas in
a node changes from a success state to a failure state to the number Case 2, it experienced the same failure in one spell (continuous
of time steps in which the node is in a failure state is worked first. failures in five time steps). The resiliency indicator is meant to de-
The geometric mean of these ratios of all nodes multiplied by 100 scribe how rapidly a system is likely to recover from failure, once a
gives γ β0 . Instead of a geometric mean, why is a demand-based failure has occurred (Srinivasan et al. 1999). Hence, it is obvious
weighted mean not considered? The problem with a geometrical
mean is explained with a simple hypothetical example. Let the
inner component of Eq. (26) of the original paper for a given node Table 1. Data for Resiliency Explanation with Two Cases
i be expressed as Number of
failure events
 Number of
PNh 1; if ðSi;h < β × Dei;h ; Si;h−1 ≥ β × Dei;h−1 Þ (spells) Ai
Node Demand failures
h¼1
0; otherwise (i) (unit) (time steps) Case 1 Case 2 Case 1 Case 2
Ai ¼  ð2Þ
PNh 1; if ðSi;h < β × Dei;h Þ 1 2 5 1 2 0.2 0.4
h¼1
0; otherwise 2 12 5 2 1 0.4 0.2

© ASCE 07017001-1 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2018, 9(1): 07017001

 that Case 1 is preferred over Case 2 as far as resilience is concerned; Eqs. (25)–(27) may or may not affect the case study results pre-
however, for both cases, as per Eq. (26) of the original paper, γ β0 sented by the authors; however, when such equations are adopted
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi elsewhere, care needs to be taken. The authors of the original paper
works to 100 × 0.2 × 0.4 or 100 × 0.4 × 0.2 ¼ 28.28, and γ β0
may share their views if they have different opinions that will be
does not differentiate the cases. Moreover, for Case 1, the demand-
useful for future research and applications.
based weighted mean works to 100 × ð2 × 0.2 þ 12 × 0.4Þ=14 ¼
37.14, and for Case 2, it works to 100 × ð2 × 0.4 þ 12 × 0.2Þ=14 ¼
22.86. This describes that the estimation of resilience through
demand-based weighted means works better than the geometric
References
mean method. Srinivasan, K., Neelakantan, T. R., Narayan, P. S., and Nagarajukumar, C.
The discussion of Eq. (26) is also extendable to Eq. (27). That is, (1999). “Mixed-integer programming model for reservoir performance
in Eq. (27) as well, instead of a geometric mean, a demand-based optimization.” J. Water Resour. Plann. Manage., 10.1061/(ASCE)
weighted mean can be preferred. Application of modified forms of 0733-9496(1999)125:5(298), 298–301.
Downloaded from ascelibrary.org by 165.50.249.73 on 11/23/17. Copyright ASCE. For personal use only; all rights reserved.

© ASCE 07017001-2 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract., 2018, 9(1): 07017001

 Discussions and Closures
Closure to “Optimal Operation of Water
Distribution Networks under Water
Shortage Considering Water Quality” by
Mohammad Solgi, Omid Bozorg-Haddad,
Samaneh Seifollahi-Aghmiuni,
Parisa Ghasemi-Abiazani, and
Hugo A. Loáiciga
DOI: 10.1061/(ASCE)PS.1949-1204.0000233
Downloaded from ascelibrary.org by 165.50.249.73 on 11/23/17. Copyright ASCE. For personal use only; all rights reserved.
Mohammad Solgi, A.M.ASCE 1; Omid Bozorg-Haddad 2;
and Hugo A. Loáiciga, F.ASCE 3
1 cation of a demand-based weighted mean is justified because dur-
M.Sc. Graduate, Faculty of Agricultural Engineering and Technology,
Dept. of Irrigation and Reclamation Engineering, College of
Agriculture and Natural Resources, Univ. of Tehran, Karaj,
3158777871 Tehran, Iran. E-mail: Solgi_Mohammad@ut.ac.ir
2 is such that during a failure period, customers do not have any ac-
Professor, Dept. of Irrigation and Reclamation, Faculty of Agricultural
Engineering and Technology, College of Agriculture and Natural Re-
sources, Univ. of Tehran, Karaj, 3158777871 Tehran, Iran (correspond-
ing author). E-mail: OBhaddad@ut.ac.ir
3
Professor, Dept. of Geography, Univ. of California, Santa Barbara,
CA 93106-4060. E-mail: Hugo.Loaiciga@ucsb.edu
The authors thank the discusser for the comments about the original
paper. Responses to each point raised in the discussion paper
follow.
The discussion questions Eq. (25) of the original paper, arguing
that the supply may exceed demand (Si;h > Dei;h ); however, this
argument is erroneous. Eq. (25) and other efficiency criteria pre-
sented in the original paper were used to evaluate the performance
of the developed optimization model. The optimization model pre-
sented in the original paper prescribes that the amount of supply is
always less than or equal to the demand. Therefore, the condition in
which the supply exceeds the demand does not occur in any nodes
of the network at any time. Si;h is a state variable of the optimization
model. In other words, Si;h derives from the demand supply index
(αi;h ) so that Si;h ¼ αi;h × Dei;h . The value of αi;h is related to the
decision variables of the optimization model and is equal to zero or
one. When αi;h ¼ 1, the demand of node i at hydraulic time step h
is fully supplied (Si;h ¼ 1 × Dei;h ¼ Dei;h ); when αi;h ¼ 0, the de-
mand of node i at hydraulic time step h is not supplied (Si;h ¼
0 × Dei;h ¼ 0). Consequently, the only possible values for Si;h are
Dei;h or zero, and Si;h does not exceed Dei;h [see Eqs. (5)–(10) of
the original paper].
The discussion proposed a demand-based weighted mean for
evaluating resiliency instead of Eq. (26) of the original paper, which
evaluates resiliency with the geometric mean; however, a demand-
based weighted mean is not appropriate for evaluating resiliency
of a water distribution network (WDN) under water shortage.
The demand-based weighted mean presented in the discussion con-
siders nodes with high demand to be more important than nodes
with low demand. This consideration is not fair under the water
shortage situation because the individuals who are connected to
a low-demand node have the same right as those individuals who
are connected to a high-demand node. Rather, using a geometric
mean that emphasizes equality among nodes is a fairer proposition.
Solgi et al. (2015) discussed equanimity and justice principles in
water distribution networks under water shortage and presented
© ASCE 07017002-1
J. Pipeline Syst. Eng. Pract., 2018, 9(1): 07017002
efficiency criteria including resiliency and reliability based on the
geometric mean. Other studies applied the geometric mean for
evaluating the resiliency of WDNs under intermittent water supply
(Soltanjalili et al. 2013; Bozorg-Haddad et al. 2016b). A geometric
mean reflects a better value for a condition in which all nodes have
the same resiliency, in comparison to a condition in which some
nodes have a good resiliency but others do not. Implementation
of a demand-based weighted mean is reasonable in some cases.
For example, in the initial design of WDNs, using a demand-based
weighted mean is prevalent (e.g., Wang et al. 2014; Bozorg-Haddad
et al. 2016a, 2017). The purpose of an initial design is satisfying
demands with a desirable pressure. Whenever a demand is satisfied
with an undesirable pressure, it is considered as a failure. Appli-
ing a failure, the demand of customers is fully satisfied, but with a
small violation of the desired pressure; however, the operation of
WDNs under water shortage as it is proposed in the original paper
cess to water even with a low, undesirable pressure. This means that
operators and customers face a serious water supply interruption.
In this case, it is not fair that individuals connected to low-demand
nodes endure long periods without water, whereas individuals con-
nected to high-demand nodes are well supplied. For these reasons,
the demand-based weighted mean proposed by the discusser has
serious shortcomings. A demand-based weighted mean would be
a viable choice for a condition in which high-demand nodes have
good resiliency and low-demand nodes have poor resiliency in
comparison to a condition in which all nodes have the same resil-
iency. For example, consider a network with two nodes, whose
resiliency in two different cases, A and B, are listed in Table 1.
Nodes 1 and 2 have low and high demand, respectively. Case A
imposes a much longer period of failure on Node 1 in comparison
to Node 2; however, Case B is a condition in which both Nodes 1
and 2 have the same resiliency. Case A represents an unfair con-
dition, whereas Case B represents a fair condition for all customers,
regardless of the type of node to which they are connected. Thus,
Case B is preferable over Case A. However, applying the demand-
based weighted mean, the average resiliencies of Cases A and B,
respectively, are ð2 × 0.056 þ 12 × 0.333Þ=14 ¼ 0.293 and ð2 ×
0.167 þ 12 × 0.167Þ=14 ¼ 0.167, which shows that Case A is
superior to Case B. In contrast, the geometric mean applied by the
original paper correctly shows that Case B is superior to Case A.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
The average resiliencies of Cases A and B are ð0.056 × 0.333Þ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0.137 and ð0.167 × 0.167Þ ¼ 0.167, which shows that the aver-
age resiliency of Case B is better (larger) than that of Case A. Using
a demand-based weighted mean may be misleading for evaluating
the performance of a WDN under water scarcity. A geometric mean
that emphasizes equality among nodes, on the other hand, is pref-
erable over a demand-based weighted mean. By implementing a
Table 1. Data for a Water Distribution Network with Two Nodes under
Two Different Conditions
Number of Number of
failures (time failure events
steps) (spells) Ai
Demand
Node (unit) Case A Case B Case A Case B Case A Case B
1 2 18 6 1 1 0.056 0.167
2 12 3 6 1 1 0.333 0.167
J. Pipeline Syst. Eng. Pract.
 demand-based operation, some operators of WDNs may provide a
good service to many customers at the expense of a small number
of customers. From the perspective of a single customer, the
WDN fails if the network cannot satisfy its demand, regard-
less of whether the customer is connected to a high-demand or
a low-demand node. Consequently, in the original paper, the same
importance is assigned to all nodes of the network by applying a
geometric mean.
The discussion proposed applying a demand-based weighted
mean for Eq. (27) of the original paper, which evaluates water-
quality resiliency with a geometric mean; however, using the
applied geometric mean is more appropriate than the proposed
demand-based weighted mean for evaluating water-quality resil-
iency. This is so because the applied geometric mean is more
conservative. By using the geometric mean, the occurrence of
Downloaded from ascelibrary.org by 165.50.249.73 on 11/23/17. Copyright ASCE. For personal use only; all rights reserved.
low water-quality resiliency at only one node of the network con-
siderably affects the mean resiliency of the network. The proposed
demand-based weighted mean, on the other hand, is such that good
resiliency at high-demand nodes may hide poor resiliency at low-
demand nodes. Given the importance attached to water quality and
its effects on human health in the original paper, the geometric
mean is more conservative and therefore preferable over the
demand-based approach insinuated by the discussers.
Several studies have considered water quality in the normal op-
eration of WDNs (e.g., Sakarya and Mays 2000; Biscos et al. 2003;
Kang and Lansey 2010; Kurek and Ostfeld 2014). The original
paper recently considered water quality in optimal operation of
WDNs under water shortage, which is a topic that had previously
received minimal attention. Yet, there are possibilities for improve-
ment that shall be covered in future studies by the authors of the
discussed paper.
References
Biscos, C., Mulholland, M., Le Lann, M. V., Buckley, C. A., and
Brouckaet, C. J. (2003). “Optimal operation of water distribution
© ASCE 07017002-2
J. Pipeline Syst. Eng. Pract., 2018, 9(1): 07017002
networks by predictive control using MINLP.” Water SA, 29(4),
393–404.
Bozorg-Haddad, O., Ghajarnia, N., Solgi, M., Loáiciga, H. A., and Mariño,
M. A. (2016a). “A DSS-based honeybee mating optimization (HBMO)
algorithm for single- and multi-objective design of water distribution
networks.” Metaheuristics and optimization in civil engineering, Vol. 7,
Springer, New York, 199–233.
Bozorg-Haddad, O., Ghajarnia, N., Solgi, M., Loáiciga, H. A., and Mariño,
M. A. (2017). “Multi-objective design of water distribution systems
based on the fuzzy reliability index.” J. Water Supply Res. Technol.
AQUA, 66(1), 36–48.
Bozorg-Haddad, O., Hoseini-Ghafari, S., Solgi, M., and Loáiciga, H. A.
(2016b). “Intermittent urban water supply with protection of con-
sumers’ welfare.” J. Pipeline Syst. Eng. Pract., 10.1061/(ASCE)PS
.1949-1204.0000231, 04016002.
Kang, D., and Lansey, K. (2010). “Real-time optimal valve operation and
booster disinfection for water quality in water distribution systems.”
J. Water Resour. Plann. Manage., 10.1061/(ASCE)WR.1943-5452
.0000056, 463–473.
Kurek, W., and Ostfeld, A. (2014). “Multiobjective water distribution
systems control of pumping cost, water quality, and storage-reliability
constraints.” J. Water Resour. Plann. Manage., 10.1061/(ASCE)WR
.1943-5452.0000309, 184–193.
Sakarya, A., and Mays, L. (2000). “Optimal operation of water distribution
pumps considering water quality.” J. Water Resour. Plann. Manage.,
10.1061/(ASCE)0733-9496(2000)126:4(210), 210–220.
Solgi, M., Bozorg-Haddad, O., Seifollahi-Aghmiuni, S., and Loáiciga,
H. A. (2015). “Intermittent operation of water distribution networks
considering equanimity and justice principles.” J. Pipeline Syst. Eng.
Pract., 10.1061/(ASCE)PS.1949-1204.0000198, 04015004.
Soltanjalili, M. J., Bozorg-Haddad, O., and Mariño, M. A. (2013). “Oper-
ating water distribution networks during water shortage conditions us-
ing hedging and intermittent water supply concepts.” J. Water Resour.
Plann. Manage., 10.1061/(ASCE)WR.1943-5452.0000315, 644–659.
Wang, Q., Guidolin, M., Savic, D., and Kapelan, Z. (2014). “Two-
objective design of benchmark problems of a water distribution
system via MOEAs: Towards the best-known approximation of
the true Pareto front.” J. Water Resour. Plann. Manage., 10.1061
/(ASCE)WR.1943-5452.0000460, 04014060.
J. Pipeline Syst. Eng. Pract.