DIMENSIONING OF BRANCHING IRRIGATION NETWORKS

BY MEANS OF LINEAR PROGRAMMING

by

1. hJAS

lnstitute of Water ~lanagement and Hydraulic Engineering, Technical "University, Budapest

(Received February 8, 1972)

Presented by Prof. 1. V. NAGY

In recent years, the method of linear programming has been used in

several countries for dimensioning branching pipe networks. Mathematical
models to this aim have been described by ZDRAZIL and SPITZ [13], LABYE
[11, 12], KAR:\IELI, GADlSH and MEYERS [7], KORSUN and SPITZ [9], KIKA-
CHEYSHVILI [8], CHUDZIK and OPALINSKI [1].
In Hungary linear programming has been used since 1966 for dimension-
ing branching networks. In the following, the mathematical model applied hy
the ahove authors and by the writer of this paper [2] with more or less changes
will be presented, referred to as the first model, fo~lowed by the description of
the ne'w model developed by the author, referred to as the second model [5],
permitting wide use of the method of linear programming.

1. Notations
lift of pumping station in the k-th service condition;
pumping level (on sucking side):
head loss of pumping station (from pump shaft to junction at main condnit in
the k-th service condition:
mark of critical point (critical hydrant):
elevation of critical point;
mark of service condition critical to point p:
service pressure above ground level to be maintained at the critical point;
head loss permissible up to critical point:
permissible minimum and maximum
velocity, respectively. in pipes:
number of pipe section considered;
by sections longitudinal elements of a pipe system are understood:
- exempt of branching:
- along which water discharge is constant in every service condition (no drawoffs o
branches are found within the length):
- wherein no hydrants to be checked for service pressure are inserted (e.g. from the
viewpoint of geodetic elevation).

Section symbols
length of section;
i-th pipe diameter in the section:
number of different pipe sizes in the section;
length of portion with the i-th diameter:

10*
350 I. IJJAS

= discharge in the i-th service condition;

specific head loss over the portion of i-th diameter in the kcth service condition;
head loss along the section in the k-th service condition;
nnmber of sections in the pipe network:
specific construction cost of pipe with the j-th diameter applied in the j-th section;
constrnction cost, understood as optimum cost of construction:
construction cost invoh-ing maximum pipe diameters.

2. The first mathematical model

The problem of dimensioning the pipe network of the irrigation system

is as follows:
Determine the optimum pipe sizes in a branching network (thus, in the
sections) if the head in the main pipe at the pumping station; the service
pressure to be maintained at the hydrants; the critical service conditions and
the corresponding overall discharge; as well as the critical points and their
elevations are known.

2.1. lVIatlzematical model in the case of a knolen pump lift

The problem is to select the pipe sizes so that the required discharge and
head is available even at the hydrants in the worst position, at a minimum
cost of construction for the pipe network.
The permissible head loss from the pumping station to the p-th critical
point is known:
(1)

The available pipe sizes (which may be used for the construction of the
pipe network) and the minimum and maximum velocity permissible in the
pipes are given.
Consider an arbitrary conduit section and assume it to be constructed
telescope-like of pipes of all the diameters available so that the velocity of
water cannot exceed the maximum permissible velocity V max even for the
maximum discharge in the section, but it is higher than the permissible
minimum velocity Vmin'
In the section of length lj' the portions built of different pipe sizes may
be of arbitrary length lij' provided two conditions are met:

(2)
and
r;
..:21ij = lj. (3)
=1
 BRAXCHI:YG IRRIGATIOX II'ETWORKS 351

Head loss in the conduit section may he written as follows:

(4)

where E; [Qkj] = specific head loss along the portion of i-th size in the k-th
service condition.
The head loss bet'ween the pumping station and the p-th critical point
(hydrant) must not surpass the value obtained from (1):

(5)

where p indicates that along the critical line between the pumping station and
point p the head losses of all the sections are to be summed.
The sections of the pipe network should he built of pipe sizes such as to
minimize the construction cost, hence:

n Tj

B = .2S ~ bij '[ij = minimum. (6)

j=l ;=1

Accordingly, the mathematical model of the problem consists of the hydraulic

conditions type (5), geometric conditions type (3) and of the target fUllction type
(6). As many hydraulic conditions should be 'written down as there are critical
points. (Exceptionally, several hydraulic condition equations may be related
to the same critical point.) Tht' geometric conditions equal in number the
sections. To meet conditions type (2) is self-evident.
If needed, also upper limiting hydraulic conditions of head may be in-
corporated in the model.
Solution of the model starts from a secondary target function, the first
step being the search of a possihle solution, that is, determination of the
lengths of portions lij meeting geometric condition equation type (3) and cor-
responding to conditions type (4-). Hereafter, the computation follows the
original function according to the so-called modified simplex method.

2.2. iVIathematical model for the determination of the optimum lift of the pumping
station

In the previous mathematical model the lengths tu made with pipes i

are unknown. No'w, a new unknown value is emerging, i.e., the-lift H of the
pumping station. The purpose of the analysis is to minimize the water lifting
3 J_
~')
I. IJJAS

costs during T years. Thus, the target function (6) tak('s the form:

n T;
K = B.J...TU ~ ...:;.
= ...,;;; "';;'bZ+THC.
lJ IJ. , (7)
j=l i=l

where B = cost of construction of pipe network;

tJ = annual cost of pumping:
T = local refund standard limit (per year):
C = yearly cost of lifting the total amount of water to unit height.
Also the hydraulic conditions are transformed:

s;
H- .::E ;2 Ei[Qlij] Iij H,,+Hp--Hl +Hb , (8)
p i=l

The geometric condition equations are unchanged.

3. The second mathematical model

Development of the model set out from that in every section of the
network the largest pipe size complying with the limit velocity Vmin has been
applied. In this case, the construction cost of the pipe network is

n
Bma" = ;; blj Ij (9)
j=l

Head loss to critical point p:

h~ = ;; E)[QkJ Ij (10)
p

where El [Q"j] specific head loss along the j-th conduit section for the
largest pipe size (meaning of p is here the same as in (5)).
If the head loss concomitant to the largest pipe size IS higher than the
permissible one, i.e.
(11)

then the problem cannot be soh'ed bv th(' utilization of the gIven series of
pipes.
Application of the largest possible diameters in the pipe network would
be uneconomic. Therefore, as far as possible, in certain sections smaller dia-
meters should be applied. But neither the total length of the new, narrower
 BR.LVCHLYG IRRIGATIO,V :VET WORKS 353

portions can be longer than the entire length of the section. Thus, the folIo'wing
condition is to be met:

(12)

wherein lij = length to change over to the i-th diameter (the largest pipe size
i = 1).
Changing over from the largest to a smaller diameter results in an in-
creased head loss. The amount of this increase along the length to the critical
points cannot exceed the difference between the head available and the head
loss due to the application of the largest pipe size.

hp :E El [Q/ij] ljo (13)

p

Besides, pipe sizes should be changed in a way to achieve a total of

construction cost saving; for each section as big as possible, in order to provide
the smallest first cost possible, as expressed by the target function:

11 rj

B' = ~ ~ (b1j-b i ) lij = maximum. (14)

j=l i=2

Solving the mathematical model consisting of geometric conditions type (12),

hydraulic conditions type (13) and target function type (14), the optimum
cost of construction of the pipe network is:

B = Bmax B:nax, (15)

where Bmax = maximum cost of construction if the largest diameters are

applied (9); and B:nax = maximum of the target function (14).
The primal variables of the simplex table consisting of the matrix con-
taining the coefficients of conditions (12) and (13), the right-hand sides of the
conditions and the target function are those portion lengths of the sections
lvhere the change-over from the largest to minor diameters is realized.
The variables corresponding to the condition rows are, from the view-
point of the simplex algorithm, dual variables, those corresponding to the
equality rows are, however, primals with respect to the outcome. Among them those
kept in the optimum program mean the lengths where the largest diameters possible
in the section result in the optimum cost of construction.
If it is wanted to solve the problem for several pump lifts, then the
right-hand side of the initial simplex table, corresponding to all pump lifts, is
to be produced. 0nly right-hand sides of the hydraulic conditions differ, that
 W
CJI
,j.>.

~-;;ru;:ther right - hand sizes of the first simplex table

Read and store data: roughness value, vmoxo Vmin, d, b
_______ f
i-Optimize 0,@ by modified simplex method
W Read data of first alternat-iV
I
~ linear programming; transform always each right-hand side
I

Select d for each section; compute head losses Compute optimum cost for first pump lift from @Ond@a,
B=Bma,-B'max
pu~~--;;r;--J

J
Start computing first simplex table for first
Print results for the first pump lift
.
rJ
Compute hydraulic conditions 0
r.r (E i [QkJ] -E 1 [Qk)] )'I ii ~ hp - H1 [Qkj] 'Ij ,...
p~ 'p No
Are there still pump lifts? I:::
~
Print results r;;
'Compute geometric
rJ
conditi~~s CB I L______

I I ~I Solve dual problem for right- hand side

i"2 'J J
of subsequent pump liff
(starting from last simplex toble)

Establish target function @

n rj No --- Are
B'=1: 1: (br -b .. ) Ii=moximum ------ there still
j"l i.2 J IJ U . variants 1
~------ .--------------~

Compute maximum investment cost (0 ~ Do ~2

of pipe network (with greatest d values) input dolo
Bmax = ~ b1j '[J'
clionge .
)"1
 355

of the geometric conditions is invariable. From the right-hand side belonging

to the first pump head these pertaining to the subsequent ones may be cal-
culated by adding the consecutive pump head differences.
The mathematical model will be solved by means of the modified simplex
method of linear programming, nevertheless in certain details of the method
the peculiarities of the mathematical model of pipe network design, the
special structure of the hydraulic and geometric conditions permitted the
application of unusual steps of algorithm, at a substantial reduction of the
solution time, hence, of the computer time.
The optimi::ation begins according to the right-hand side corresponding
to the lift of the first pump. After achieving the optimum program and printing
the outcome, the right-hand sides corresponding to the subsequent pump lifts
are handled as solution of the dual problem. Thereby the computer time
required for the second and further heads could be reduced to a fraction of
that needed for the first head.
The operations appear from the flow chart on p. 354.

4. The second model and pecnliarities of the method applied

The second model is more eon-venient than the first one hy its fewer
iteration8, the 8maller extent of the problem and the shorter computer timc.
For this problem, se-veral programs ha-ve heen developed in ALGOL 60
language whieh are applied by the Water Management Organization and
Computing Bureau and hy the Institute for \\1 ater Management and Hydraulic
Engineering of the Teehnical "University, Budapest.
The model is computer constructed climinating the cumbersome data
setup.
The program permits the determination of optimum diameter com-
binations corresponding to lifts of seyeral pumps, in-voh-ing several right-hand
sides. The optimum related to the first lift having been determined, com-
putation affects the second right-hand side, i.e., the dual problem. The algo-
rithm quickly approximates the optimum for the second and further pump
lifts. The number of iterations and therehy, the computer time needed for the
second and further lifts make up a minor fraction of those required for the first
lift.
This program helped already to design sprinkler irrigation systems for an
area over 150000 ha in Hungary. Economy in first costs amounted to 2500
f orints per ha.
356 I IJJAS

5. Comparison of the linear programming and other methods for optimizing

hranching irrigation networks

For the optimization of branching networks, programs operating accord-

ing to different methods have been established (LA-BYE'S discontinuous method,
linear programming, dynamic programming) [2 to 6]. In the case of identical
input data and conditions, the programs operating according to different
methods gave identical results. Each method has its field of application.
The benefit of the dynamic programming [6] is to deliver at the same
time the optimum trace and cost of the conduit.
The advantage of the discontinuous method and dynamic programming
is a reduced computer time.
The linear programming is the most general and most versatile method.
Cases where e.g. some conduit sections have different water discharges under
different service conditions, can only be approximated either by dynamic
programming or by the discontinuous method. This is a prohlem for linear
programming.
Dynamic programming and the program using the discontinuous method
help to find the optimum cost of construction for all heads possible at the
pumping station (minimum cost polygon of the system) while the program
based on linear programming offers the optimum cost of construction cor-
responding to given lifts.

Summary
In Hungary, linear programming has been applied since 1966 for dimensioning branching
irrigation pipe networks. The presented mathematical model (first model) has been used in
several countries with more or less modifications, yet, another mathematical model (second
model) has been developed at the Institute of Water Management and Hydraulic Engineering of
the Technical University. Budapest. For the optimization of the branching pipe networks,
programs using several different methods (Labye's discontinuous method, linear programming)
have been established. In the case of identical input data and conditions, the programs operat-
ing according to different methods gave identical results. The application fields of the methods
are different. In many cases, the choice of the convenient method depends on the nature of
the problem.

References
1. CHliDZIK, B.-OPALIl'\SKI. Cz.: Techniczno-ekonomiczna optimalizacja projektu sieci
rurociag6w deszczownianych. Zeszyty Xaukowe Wyzszej Szkoly Rolniczej we \\'ro-
clawiu. Wroclaw. 1970. Nr. 90.
2. IJJA5. I.:' Investigation of underground pipe network of sprinkler irrigation systems.
(In Hungarian). (Dr. Techn. Thesis 1966).
3. IJJA5, I.: Determination of the optimum lift of pumping stations by means of dynamic
programming. Periodica Polytechnica. Civ. Eng. 14 (1970) 243.
4. IJJA5, I.: Okreslenie srednic podziemnej sieci rurociag6w deszczowni. Zeszyty :\'aukowe
.. \\iyzszej Szkoly Rolniczej we Wroclawiu, Wroclaw, 1970. :\'r. 90.
,).IJJA5, I.: Application of computers in agricultnral water resources management. (In
Hungarian). Lecture notes for special engineering students. Tankiinyvkiad6 1971
6. IJJAS, I.: Determination of the optimal route of the main conduit of branching irrigation
pipe networks by means of dynamic programming. ICID YIlI. European Congress, 1971.
 BRASCHISG IRRIGATIOS SEnmRKS 337

7. KARlIELL D.-GADISH, Y,-MEYERS, S.: Design of optimal water distribution networks.

Proe. ASCE 1968. PL 1.
8. KIKACHEYSHVILI, G. E.: Tehniko-ekonomichesky raschet razvetvlennikh vodoprovodnikh
setey metodom lineynogo programmirovaniya. Vodosnabzhenie i Sanitarnaya Teknika,
1969/6.
9. KORSUK, S.-SPITZ, P.: Pouziti linearniko programovani pri dimenzovani zavlahovych
trubnich Riti. Vodni hospodarstvi-A, 2 (1969).
10. l\10LK"\.R, L.: Some experiences on the sprinkler irrigation system in Hajduszovat. (In
Hungarian). Viztigyi Kozlemcnyek, 3 (1970).
11. LABYE, Y.: Etude des procedes de calcul ayant pour but la distribution d'cau. La Houille-
Blanche, 5 (1966).
12. LABYE, Y.: L'utilisation en France des calculatrices eIectroniques pour les projets d'irri-
gation et d'assainissement. Essai de synthese, ICID Symposium, Mexico City, 1969.
13. ZDRAZIL, K.-SPITZ, P.: Ekonomicke dimcnzovani trubni site metodou linearniko progra-
movani. Vodni Hospodarstvi 9 (1965).

Senior _~ss. DJ' Ish-an lJJAS, 1111 Budapest, lVIiiegyetem rkp. 3, Hungary