An Investigation of The Optimization of Potable Water Network-Libre

Fifth International Water Technology Conference IWTC 2000, Alexandria, Egypt
61
AN INVESTIGATION OF THE OPTIMIZATION OF POTABLE

WATER NETWORK
Berge Djebedjian*, Ahmed Herrick** and Magdy Abou Rayan*
* Mechanical Power Engineering Department, Faculty of Engineering,
Mansoura University, El-Mansoura, Egypt
** Ismailia Water Treatment Plant, Suez Canal Authority, Ismailia, Egypt
ABSTRACT
A potable water network hydraulic analysis is presented in this paper. A
mathematical model was developed, the model treats looped network. A computer
program is developed in order to facilitate water distribution system design which
satisfies all constraints including pipe diameter and nodal pressure.
An optimization technique is developed in order to evaluate the optimum
network configuration; the parameters are pipe diameter, flow rate, corresponding
pressure and hydraulic losses. A non-linear technique was adopted in the solution.
The hydraulic analysis is performed using the Newton-Raphson method.
The initially assumed pipe diameters are successively adjusted to suit the
existing standard commercial pipe diameters. The technique was applied on a
simple case study of gravity-fed network.
The objective of the present investigation is to present a practical tool to
help in the optimization of water distribution system, design and operation.
INTRODUCTION
The optimization of pipe networks has been studied and various researchers
have proposed the use of mathematical programming techniques in order to
identify the optimal solution for water distribution systems. The optimal solution
always means minimum cost of the network. The word "minimum" is only relative.
With different criteria or objective functions; different values for "minimum cost"
can be obtained for the same system. Depending on the situation, one objective
function may be appropriate for one system and totally inappropriate for another
62
system. The specific form of the objective function is often determined by the
operating policies of a particular company.
The optimization techniques can be categorized as follows: deterministic
optimization techniques (including linear, dynamic, and non-linear programming)
and stochastic optimization techniques (such as genetic algorithms and simulated
annealing).
Linear programming is used to optimize a design of a pipe network two
principal approaches have been developed (Alperovits and Shamir, 1977, and
Quindry et al., 1981). Alperovits and Shamirs (1977) approach has the ability to
consider various components in a distribution network: however it is severely
limited in the size of the system and the number of loads which it can handle.
Quindry et al. (1981) improved the method allowing for a larger system to be
considered, but difficulties arise when analyzing multiple loads. The limitation of
this method is that it considers only pipe portion, it does not consider any other
component as pump, reservoir, etc. Srbu and Borza (1997) proposed a model
based on the method of linear programming to treat looped networks which have
concentrated outflows or uniform outflow along the length of each pipe.
For branched pipeline networks, Karmeli et al. (1968) presented a method
for solving the equations using the theory of linear programming to minimize the
total cost by determining the optimal commercially available pipe size for each
link, and for the head at each node. Chiplunkar and Khanna (1983) presented an
optimization algorithm for design of branched rural water supply systems using the
Lagrangian multiplier technique. Hathoot (1986) presented three formulas for
designing inclined pipelines of optimum diameter with equally-spaced, similar
pumping units. Fujiwara and Dey (1988) presented a method for design branched
networks on flat terrain by using the Lagrange multipliers method to obtain
optimal pipe size, this method is limited to branched networks location flat terrain
with a single source node and equal head for each end node. For looped networks,
Featherstone and El-Jumaily (1983) presented a method to get the minimum cost of
the network by equating the first derivative of the total cost equation with zero.
The equivalent pipe diameter method for network optimization has been
developed by Deb and Sarkar (1971) using the pressure surface profile capital cost
functions for pipes, pumps, and reservoirs. Swamee and Khanna (1974) have
shown that this method has two major drawbacks: first it lacks mathematical
justification for cost equivalent pipes; and second, a hydraulic pressure surface
over the network must be artificially created.
Non-linear programming is applied to pipe network optimization problems
and many researches have been reported. Jacoby (1968) proposed a non-linear
programming method with continuous variables, thus obtaining a solution with
63
theoretical diameters, to be rounded off to commercial values. The author applied

this very complex method to a simple network with two loops and five branches,
excluding complex networks. Cenedese and Mele (1978) proposed an optimal
method assuming that the most economical distribution system is always an open
network and adding some connections, so it increases the total cost than the
optimum cost. This method is applicable only to a system with a constant input
head without pumping. In practice, this can only be the case where the supply is
from a high level water reservoir. Samani and Naeeni (1996) proposed a non-linear
optimization technique coupled with the Newton-Raphson method to minimize the
design total cost with constraints in pipe diameters, flow velocities and nodal
pressures.
In the present investigation a non-linear programming is applied for pipe
network optimization. An optimal solution for looped water distribution networks
by using computer facilities is developed. The model uses the Newton-Raphson
method for the hydraulic analysis of the network and the sequential unconstrained
minimization technique (SUMT) of Fiacco and McCormick (1964) to solve the
optimal design of network. This model is presented for new or partially extend
water distribution networks, which operate with gravity (Elevated Tanks).
OPTIMIZATION MODEL FORMULATION

The optimization of the network design problem is the identification of the
commercial pipe size diameters combination that give the minimum cost under
certain conditions such as the specified demands and prescribed range of pressures
at given nodes.
The performance of the proposed approach is tested on a simple network
which does not contain pumps. The existence of a reservoir is taken as a water
source node of fixed head.
The minimization of cost for a gravity-fed network is expressed by the
objective function which is assumed to be a function of pipe diameters and lengths,
(Savic and Walters, 1997):
f (D1 ,..., D N ) =
c ( Di , Li )
(1)
i =1
where c( Di , Li ) is the cost of the pipe i with the diameter Di and the length Li , and
N is the total number of pipes in the network.
The minimization of cost for a gravity-fed network is characterized by the
following conservation laws and constraints:
(1) Mass conservation at each junction node:

Qin
Qout = Q e
64
(2)
where Qin and Qout are the flow into and out of the junction, respectively, and
Qe is the demand at the junction node.
(2) Energy conservation in each loop for a gravity-fed network can be written as:
hf = 0
(3)
The head loss h f in the pipe is expressed by the Hazen-Williams or DarcyWeisbach formula. The Hazen-Williams formula is selected to represent the
head losses in the pipes:
1.852
10.675 Li Qi
h f = 1.852
(4)
4.8704
C
Di
where Qi is the pipe flow and C is the Hazen-Williams coefficient.
For more than one source node (reservoir) available in the system, additional
energy equations are written for paths between any two of the nodes.
(3) Minimum pressure head requirements at each node in the network is given in
the form:
H j H j , min
j = 1,..., M
(5)
where H j is the head at node j, H j , min is the minimum required head at the
same node and M is the total number of nodes in the network.
(4) Minimum diameter requirements are defined by:
Di Dmin
i = 1,..., N
(6)
where Dmin is the minimum diameter. The diameter of each pipe is chosen
from a specified set of commercial pipes.
The SUMT Method

The Sequential Unconstrained Minimization Technique (SUMT) was first
suggested by Carroll (1961) and thoroughly investigated by Fiacco and
McCormick (1964). The formulation of the constrained minimization problem is in
the form:
minimize z = f (x)
(7)
(8)
subject to
c j ( x) 0; j = 1,2,,m
where c j ( x) presents the constraints and m is the number of constraints. Fiacco
and McCormick (1964) used the following formulation to generate a sequence of
65
feasible vectors to the original problem for a strictly decreasing sequence of r

values tending to zero:
m
1
(9)
minimize L ( x, r ) = f ( x) + r
c
(
x
)
j =1 j
Hence, the generalized objective function for the cost can be introduced as;
L(Di , H j , r ) = f (Di ) + r
1
+
D
D
i =1 i
min
1
j =1 H j H j , min
(10)
In the above expression, when the diameters and pressure heads are in the
allowable ranges, r should be considered equal to zero which means it does not
affect the real cost. The objective function given by Eq. (10) is minimized by the
SUMT method to obtain the minimum cost.
Computational Analysis
The coupled hydraulic and optimization analysis of network can be
summarized as follows:
1) Assume the diameters of the pipes.
2) Solve the non-linear system of equations for the hydraulic analysis via the
Newton-Raphson method to obtain the pressure heads at nodes, discharges and
head-losses of all pipes.
3) Compute the objective function of Eq. (10).
4) Use the SUMT method to minimize the cost objective function. If the objective
function is not minimum, pipes diameters should be changed. Then, repeat the
cycle from stage (2).
Case Study
The performance of the Newton-Raphson method for the hydraulic analysis
with the SUMT method is tested on a simple two-loop network which does not
contain pumps or valves, Fig.1. This network is originally presented by Alperovits
and Shamir (1977) and taken as a model network by many investigators (Quindry
et al., 1981; Goulter et al., 1986; Fujiwara et al., 1987; Kessler and Shamir, 1989;
Bhave and Sonak, 1992; Sonak and Bhave, 1993; Eiger et al., 1994; Savic and
Walters, 1997; and Cunha and Sousa, 1999). It is fed by gravity from a constant
head reservoir (210 m). There are eight pipes, 1000 m long each. The demands
given in cubic meters per hour and the ground level (m) at each node are shown in
Fig. 1.
66
210 m
3
100 m /h
100 m /h
(150 m)
L = 1000 m
(160 m)
L = 1000 m
L = 1000 m
270 m3/h
L = 1000 m
120 m3/h
L = 1000 m
(150 m)
(155 m)
L = 1000 m
3
L = 1000 m
3
200 m /h
330 m /h
7
(160 m)
6
L = 1000 m
6
(165 m)
Fig. 1 Test problem network (Alperovits and Shamir, 1977)
The minimum acceptable pressure requirements for nodes 2 to 7 are defined

as 30 m above ground level. The Hazen-Williams coefficient C for all pipes of the
network is equal to 130.
The cost in arbitrary units per one meter of pipe length for the 14
commercially available diameters in inches are presented in Table 1. The mixed
units (SI and foot-pound-second) are used in this study since they were applied by
Alperovits and Shamir (1977) and other researchers.
67
Table 1. Commercially available pipe sizes and cost per meter, Alperovits and
Shamir (1977)
Diameter
(in.)
Diameter
(mm)
Cost
(units)
1
2
3
4
6
8
10
12
14
16
18
20
22
24
25.4
50.8
76.2
101.6
152.4
203.2
254.0
304.8
355.6
406.4
457.2
508.0
558.8
609.6
2
5
8
11
16
23
32
50
60
90
130
170
300
550
COMPUTATIONAL RESULTS AND DISCUSSION

The previous optimization model is applied to the case study. The computer
program for the network analysis and optimization was written in Fortran and was
run on a Pentium MMX 330. Although the simplicity of the studied network, it
should be mentioned that for this simple network with eight pipes and a set of 14
commercial pipes, the total number of designs is 148 = 1.48 X 109. Therefore, it is
very difficult for any mathematical model to test all these possible combinations of
design and a very small percentage of combinations can be reached.
The optimal solution is obtained for the set of diameters of 18, 10, 16, 4, 16,
10, 10 and 1 inch for the links 1 to 8, respectively. Table 2 gives the diameters and
the optimum cost obtained by the current study in comparison with other authors.
It can be noticed that this optimal is identical to that obtained by Savic and Walters
(1997) and Cunha and Sousa (1999). In their studies and the present study,
identical pipes for each link are used. The first four researches used the split-pipe
design which considers that each pipe could be divided into two or more different
pipes, with different diameters. This kind of design is less realistic as noted by
Savic and Walters (1997).
The reliability of the network demands that each link of the network must
carry at least some minimum discharge, i.e. implying the minimum link flow
constraint. Suppressing this constraint, some previous studies (e.g. Bhave and
Sonak, 1992) showed that the global optimal solution can be obtained by canceling
68
some links from the network. Therefore, a looped network can be converted to a
branching configuration, i.e. a distribution tree (Alperovits and Shamir, 1977)
which its total number can be obtained by the graph theory. For the studied case
there are 15 trees and Bhave and Sonak (1992) showed that the network has 15
local optimum solutions. The solution corresponding to the tree without the links 4
and 8 ( Q4 = 0 and Q8 = 0 ) has the minimum cost of 400,129 units and therefore it
is the global optimum one.
Table 2. Optimal solutions for the case study
Pipe
1
2
3
4
5
6
7
8
Cost (units)
Alperovits and
Shamir (1977)
L (m)
256.00
744.00
996.38
3.62
1000.00
319.38
680.62
1000.00
D (in.)
20
18
8
6
18
8
6
16
784.94
215.06
1000.00
12
10
6
990.93
6
9.07
4
479,525
Goulter et al.
(1986)
L (m)
383.00
617.00
1000.00
D (in.)
20
18
10
1000.00
687.00
313.00
1000.00
16
6
4
16
98.00
12
902.00
10
492.00
10
508.00
8
20.00
2
980.00
1
435,015
Kessler and
Shamir (1989)
L (m)
1000.00
D (in.)
18
66.00
12
934.00
10
1000.00
16
713.00
3
287.00
2
836.00
16
164.00
14
109.00
12
891.00
10
819.00
10
181.00
8
920.00
3
80.00
2
417,500
Eiger et al.
(1994)
Savic and
Walters
(1997)
Cunha
and
Sousa
(1999)
Present
Study
L (m)
1000.00
D (in.)
18
D (in.)
18
D (in.)
18
D (in.)
18
238.02
761.98
1000.00
1000.00
12
10
16
1
10
10
10
16
4
16
4
16
4
628.86
371.14
989.05
10.95
921.86
78.14
1000.00
16
14
10
8
10
8
1
16
16
16
10
10
10
10
10
10
419,000
419,000
419,000
402,352
The output of various runs of the computer program is given in Table 3. This
is a part of the results that satisfies the constraints given in the problem. The
diameters are greater than 1 inch and the pressure heads at the nodes are not below
30 meters. The given results are for costs less than 600,000 units. The first result is
the optimal one and the other values of cost close to the optimal are presented. It
can be observed that many combinations of pipes set can give the same cost
because of the equality of pipes lengths.
Table 3. Diameters, discharges, node pressure heads and the costs

for the case study
Node
Cost
419000
Pipe
D (in.)
1
18
2
10
3
16
4
4
5
16
6
10
7
10
8
1
1120
336.87
53.24
12
683.13
30.46
16
32.57
43.45
2
530.56
33.8
16
-200.56
30.44
10
236.87
30.55
10
-0.56
434000
Q (m3/h)
H (m)
D (in.)
364.07
53.24
14
655.93
37.17
16
4.23
43.79
3
531.7
39.03
16
-201.7
30.78
10
264.07
30.83
10
-1.7
363.9
53.24
14
656.09
40.38
16
9.9
43.79
3
526.2
42.25
16
-196.2
30.83
10
263.9
31.14
12
3.8
373.17
53.24
16
646.83
40.24
16
5.53
43.91
2
521.29
46.68
16
-191.3
31.00
10
273.17
31.52
10
8.7
368.78
53.24
16
651.22
41.71
16
3.12
43.85
2
528.1
43.3
16
-198.1
30.87
10
268.78
31.09
10
1.89
371.75
53.24
14
648.25
41.69
16
3.19
43.89
3
525.06
43.1
16
-195.06
30.94
12
271.75
31.3
12
4.94
370.79
53.24
14
649.21
40.28
16
5.37
43.88
3
523.84
46.77
16
-193.84
30.94
12
270.79
34.05
12
6.15
376.18
53.24
10
643.82
40.2
20
5.74
43.95
12
518.08
46.56
12
-188.08
31.07
10
276.18
34.28
8
11.92
179.77
55.96
18
840.23
41.96
18
300.25
48.58
14
419.97
49.33
14
-89.97
30.67
8
79.77
34.56
8
110.03
213.11
53.24
14
450.29
53.24
10
237.24
55.96
16
467.38
53.24
16
806.88
42.93
14
569.71
39
16
782.76
39.28
16
552.62
40.87
16
270.54
44.56
14
116.75
41.68
16
364.53
44.78
2
1.82
45
16
416.34
47.91
16
332.96
46.33
12
298.23
48.28
16
430.79
47.95
16
-86.34
30.89
14
-2.96
30.41
12
31.77
30.58
14
-100.8
32.96
10
113.11
32.84
14
350.29
35.41
12
137.24
35.65
14
367.38
37.7
10
113.66
297.9
53.24
16
722.09
42.21
16
198.62
42.93
10
403.47
47.44
16
-73.47
31.12
10
197.9
35.36
10
126.52
332.41
53.24
14
687.59
41.98
14
149.38
43.39
14
418.21
45.55
14
-88.22
31.45
14
232.41
35.38
14
111.78
454.54
55.96
565.46
41.63
152.77
44.48
292.69
48.9
37.3
32.56
354.53
37.61
237.3
450000
Q (m /h)
H (m)
D (in.)
3
468000
Q (m /h)
H (m)
D (in.)
3
474000
Q (m /h)
H (m)
D (in.)
3
477000
Q (m /h)
H (m)
D (in.)
3
486000
Q (m /h)
H (m)
D (in.)
3
489000
Q (m /h)
H (m)
D (in.)
3
550000
Q (m /h)
H (m)
D (in.)
3
579000
Q (m /h)
H (m)
D (in.)
3
580000
582000
585000
586000
Q (m /h)
H (m)
D (in.)
Q (m3/h)
H (m)
D (in.)
Q (m3/h)
H (m)
D (in.)
Q (m3/h)
H (m)
D (in.)
3
18
1120
18
1120
18
1120
18
1120
18
1120
18
1120
18
1120
20
1120
18
1120
18
1120
20
1120
18
1120
18
1120
586000
Q (m /h)
H (m)
D (in.)
1120
590000
Q (m3/h)
H (m)
D (in.)
Q (m3/h)
H (m)
1120
18
20
14
197.03
12
231.77
14
99.2
10
16
14
69
70
Note: The initial directions for the flow in the pipes are proposed as follows: from
nodes 1 to 2, 2 to 3, 3 to 5, 5 to 7, 2 to 4, 4 to 6, 4 to 5 and 7 to 6. The negative sign
in the results means the contrary direction.
CONCLUSIONS
The determination of the optimal design for water distribution networks is
computationally complex. In this paper a nonlinear programming is applied for
pipe network optimization. The Newton-Raphson method for the hydraulic
analysis of the network and the sequential unconstrained minimization technique
(SUMT) of Fiacco and McCormick (1964) are used to solve the optimal design of
network. The application of the method on a simple fed-gravity network shows the
capability of the approach to solve such optimization problems. The first
evaluation of the method is satisfactory, although of the shortcoming of the
method, as in many other methods, is that the inability of finding the global
optimum.
In future studies, the optimization of networks with pumps and reservoirs
will be considered as these are the actual real-world problems.
NOMENCLATURE
C
c( Di , Li )
c j ( x)
Hazen-Williams coefficient
cost of the pipe i
constraints
Di
Dmin
Hj
diameter of pipe i, (m)

minimum diameter, (m)
head at node j, (m)
H j , min
hf
L
Li
M
m
N
Qe
Qi
minimum required head at the node j, (m)

head loss, (m)
objective function
length of pipe i, (m)
total number of nodes in the network
number of constraints
total number of pipes in the network
demand at the junction node, (m3/s)
flow in pipe i, (m3/s)
Qin
Qout
r
71
flow into of the junction node, (m3/s)

flow out of the junction node, (m3/s)
scale factor, Eq. (10)
REFERENCES
1. Alperovits, E., and Shamir, U., "Design of Optimal Water Distribution
Systems," Water Resources Research, Vol. 13, No. 6, 1977, pp. 885-900.
2. Bhave, P.R., and Sonak, V.V., "A Critical Study of the Linear Programming
Gradient Method for Optimal Design of Water Supply Networks," Water
Resources Research, Vol. 28, No. 6, 1992, pp. 1577-1584.
3. Carroll, C.W., "The Created Response Surface Technique for Optimizing
Nonlinear Restrained Systems," Operations Research, Vol. 9, 1961, pp. 169184.
4. Cenedese, A. and Mele, P., "Optimal Design of Distribution Networks,"
Journal of the Hydraulics Division, ASCE, Vol. 104, No. HY2, 1978, pp. 237247.
5. Chiplunkar, A.V., and Khanna, P., "Optimal Design of Branched Water
Supply Networks," Journal of the Environmental Engineering, ASCE, Vol.
109, No. 3, 1983, pp. 604-618.
6. Cunha, M.D.C., and Sousa, J., "Water Distribution Network Design
Optimization: Simulated Annealing Approach," Journal of Water Resources
Planning and Management, ASCE, Vol. 125, No. 4, 1999, pp. 215-221.
7. Deb, A.K., and. Sarkar, A.K., "Optimization in Design of Hydraulic
Networks," Journal of the Sanitary Engineering Division, ASCE, Vol. 97, No.
SA2, 1971, pp. 141-159.
8. Eiger, G., Shamir, U., and Ben-Tal, A., "Optimal Design of Water
Distribution Networks," Water Resources Research, Vol. 30, No. 9, 1994, pp.
2637-2646.
9. Featherstone, R.E., and El-Jumaily, K.K., "Optimal Diameter Selection for
Pipe Networks," Journal of Hydraulic Engineering, ASCE, Vol. 109, No. 2,
1983, pp. 221-234.
10. Fiacco, A.V., and McCormick, G.P., "Computational Algorithm for the
Sequential Unconstrained Minimization Technique for Nonlinear
Programming," Management Science, Vol. 10, 1964, pp. 601-617.
72
11. Fujiwara, O., and Dey, D., "Method for Optimal Design of Branched
Networks on Flat Terrain," Journal of the Environmental Engineering, ASCE,
Vol. 114, No. 6, 1988, pp. 1465-1475.
12. Fujiwara, O., Jenchaimahakoon, B., and Edirisinghe, N.C.P., "A Modified
Linear Programming Gradient Method for Optimal Design of Looped Water
Distribution Networks," Water Resources Research, Vol. 23, No. 6, 1987, pp.
977-982.
13. Goulter, I.C., Lussier, B.M., and Morgan, D.R., "Implications of Head Loss
Path Choice in the Optimization of Water Distribution Networks," Water
Resources Research, Vol. 22, No. 5, 1986, pp. 819-822.
14. Hathoot, H.M., "Minimum-Cost Design of Pipelines," Journal of
Transportation Engineering, ASCE, Vol. 112, No. 5, 1986, pp. 465-480.
15. Jacoby, S.L.S., "Design of Optimal Hydraulic Networks,", Journal of the
Hydraulics Division, ASCE, Vol. 94, No. HY3, 1968, pp. 641-661.
16. Karmeli, D., Gadish, Y., and Meyers, S., "Design of Optimal Water
Distribution Networks," Journal of the Pipeline Division, ASCE, Vol. 94, No.
PL1, 1968, pp. 1-10.
17. Kessler, A., and Shamir, U., "Analysis of the Linear Programming Gradient
Method for Optimal Design of Water Supply Networks," Water Resources
Research, Vol. 25, No. 7, 1989, pp. 1469-1480.
18. Quindry, G.E., Brill, E.D. and Liebman, J.C., "Optimization of Looped
Water Distribution Systems," Journal of the Environmental Engineering
Division, ASCE, Vol. 107, No. EE4, 1981, pp. 665-679.
19. Samani, H.M.V., and Naeeni, S.T., "Optimization of Water Distribution
Networks," Journal of Hydraulic Research, Vol. 34, No. 5, 1996, pp. 523-632.
20. Srbu, I., and Borza, I., "Optimal Design of Water Distribution Networks,"
Journal of Hydraulic Research, Vol. 35, No. 1, 1997, pp. 63-79.
21. Savic, D.A., and Walters, G.A., "Genetic Algorithms for Least-Cost Design of
Water Distribution Networks," Journal of Water Resources Planning and
Management, ASCE, Vol. 123, No. 2, 1997, pp. 67-77.
22. Sonak, V.V., and Bhave, P.R., "Global Optimum Tree Solution for SingleSource Looped Water Distribution Networks Subjected to a Single Loading
Pattern," Water Resources Research, Vol. 29, No. 7, 1993, pp. 2437-2443.
23. Swamee, P.K., and Khanna, P., "Equivalent Pipe Methods for Optimization
Water Networks-Facts and Fallacies," Journal of the Environmental
Engineering, ASCE, Vol. 100, No. EE1, 1974, pp. 93-99.

An Investigation of The Optimization of Potable Water Network-Libre

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

An Investigation of The Optimization of Potable Water Network-Libre

Enviado por

Direitos autorais:

Formatos disponíveis

Fifth International Water Technology Conference IWTC 2000, Alexandria, Egypt