Beruflich Dokumente
Kultur Dokumente
(Received 28 February 1992; final version received 10 August 1992; accepted 10 November 1992)
ABSTRACT
KEY WORDS
SOFTWARE AVAILABILITY
The software has been developed in FORTRAN language and is available for use on IBM
PC-XT/AT compatible personal computers under MS-DOS as well as on HP-9050-AM computer
system under HP-UX operating system (Hewlett-Packard, USA).
The software is available from the authors at the National Environmental Engineering
Research Institute, Nehru Marg, Nagpur - 440 020, India.
[2] developed user friendly software based on the trial estimation of frictional bead-loss in pipes. It has been
and error approach to identify the op6mal combination shown by Jain et at. [9] that the use of this formula
of pipe sizes. results in errors up to =1=.30% and :f.55% in the estimation
of velocity and frictional bead-loss respectively for
Quindry et al. [3] developed an algorithm based extreme conditions of velocities and pipe roughness
on the gradient search technique incorporating a linear encountered in practice. The researchers also presented
programming approach. Swamee et al. [4] developed the Modified Hazen-Williams formula which obviates
an algorithm to reorganize a distribution system the inherent limitations of the Hazen-Williams
where the network is synthesized via linear formula and has been employed in this research.
programming to yield the new pipe diameters and the Thus, frictional bead-loss in a pipe link can be
input point pumping beads. estimated as,
L qt.So99
Watanatada [5] used Davidon-Fletcber-Powell
method to solve an unconstrained version of a 54 f ~ ........................... ...(la)
node network and reported on the unsuitability of the 994.62 Ca I " ~ D4 " ~
approach for large size networks. C,enedese et al. [6]
formulated a multiobjective optimization problem -- Z D--A'aog9 ...(lb)
considering construction cost (capital cost) and
management cost (operational cost) as the two in which, f is frictional bead-loss in pipe in m; L
objectives and obtained a Pareto optimal set (i.e. a is length of pipe in m; q is flow in pipe in
series of compromise solutions which take into m3/s; C r is coefficient of roughness, D is internal
account the existing relation between different points diameter of pipe in m; and Z is resistance
of view) for a direct pumping water distribution coefficient term for the pipe given by (Lq t's°99) /
system. Chiphinkar et al. [7] developed an algorithm (994.62 CRl"s°°9).
using interior penalty function with Davidon-Fletcber-
Powell method and demonstrated the efficacy of the
Cost Model
approach through a case study for a large sire system.
Lartsey and Mays 181 developed a methodology
The total cost of a water distribution system is the
based on reduced gradient method using an augmented sum of individual cost components in the system, i.e.
Lagrangian approach to incorporate pressure head
pipe network, pumping systems, power, and
bounds in the objective function. distribution reservoirs.
This paper addresses itself to the development Capital cost of pipe, including the cost of laying
of software for optimal design of a new, or optimal and jointing, can be expressed as,
augmentation of an existing, combined pumping and
gravity system with multiple reservoirs. For a new
Cpip~ = K I L D" ...(2)
design problem, given a network configuration, the
program computes the pipe diameters and reservoir
heights which minimize the total cost of the system in which, C p,pe
. is the cost of pipe in Indian Rupees
while satisfying the quantity and head requirements. (1 US $ = 25.95 Indian Rupees); and K I and a am
For reorganization, since the height of the existing coefficient and exponent of the regression equation and
reservoir and the existing pipeline must remain fixed, are functions of pipe material.
functional design is achieved through additional
pipes provided in parallel with existing ones Cast iron (CO, asbestos cement (AC) and galvanized
wherever necessary and providing new reservoir(s) or iron (GI) pipes have been considered for regression
enhancing the pumping capacity. The efficacy of analysis. Cost data have been taken from the Schedule
the algorithm is demonstrated through an example of rates (1990-91) provided by the Maharashtra Water
considering reorganization of a typical water Supply and Sewerage Board, India. The values of
distribution system. coefficients for different pipe materials obtained by
regression analysis (Eq.2) are presented in Table 1.
contains all the nodes but no loop. The nootree edges k = (2p-n+2) ..... (2p+l); j = l,...,n
constitute the elements of the ~ which is the
complement of spanning tree. gk(X) = Zfji + H I - H., = 0 ...(15)
Using graph theoretic formulation (Boulos and k = (2p+2) ..... (2p+s); j = 2,...,s
Airman [11]) for analysis of looped networks, Eqs.
9 and l0 could be combined in a single matrix equation.
gk(X) = Rmin R.J -< 0 ...(16)
The number of simultaneous hydraulic nonlinear
equations and the number of variables to be solved
k = (2p+s+l) ..... (2p+s+d); j = 1..... d
in this method is equal to the number of couee pipes
in the network (p - n + 1), in which p and n are
gk(X) = H.j Hmax _< 0 ...(17)
the number of pipes and nodes respectively. An energy
equation is added to the full equation set for each
node where energy grade or pre~ure is specified except k = (2p+s+d+l) ..... (2p+s+d+e); j = 1..... e
tbr the reference node. Modified Newton's method
(Dennis and Schnabel [121) has been used in this gk(X) = H.J < 0 ...(18)
research to solve the resulting system of nonlinear
equations. k = (2p+s+d+e+l) ..... (2p+s+d+2e); j = 1 ..... e
1 1
SET STEP LF.NGTII o f t I: r r E R t l
SET DIRECTION MATRIX J AS
1
SOLVE SYSTEM OF NONLINFAR
1 COMPUTE V ~
.
AT X
EQUATIONS BY MODIFIED
NEW'TOWS METIIOD. COMPUTE DIRE(TTION VECTOR .q
FIND FLOW IN ALL PIPES AND NORMAI.ISE S
S = - J 9@: S, t S/4(y...~;~)
I ,,E^,, , I 1
CAI.I. ONEIIIM . COMPUTE TIlE
()PTIMAL ,RTE.P I.EN(TI'I! ~" AND
NEW DF.~I(|N VECTOR X m
C()MI'U'rE GRAI)IENT V ~ . AT X.
?
(lm,,~) tO achieve convergence given by,
O-F
I = 6 + INT [. . . . . . . . . * 5] ...(20)
F
YES
. Form the pseudo-circuit which is a chain of of the upper triangular matrix are stored resulting in
connected tree branches between the reference node substantial reduction in memory storage.
and a node with specified grade
The DFPM subroutine based on the interior
4. Solve the resulting simultaneous equations using penalty function method with recourse to Davidon-
modified Newton's method to obtain the flow Fletcher-Powell method comprises the following steps:
in all the pipes
1. Initialize direction matrix to identity matrix (J)
5. Compute the residual heads at all the nodes
2. F'md the search direction (S)
Newton's method requires the inversion of the
Jacobian which is symmetric and hence only the elements 3. Find the optimal step length (ct*)
Optimization of water distribution system 107
: s~w,,.1,..I , o / ~
i^
EVALUATE 0 , AT I
CALL ANAL TO OBTAIN XF~LOWS I A-B'; F, - F,,{
AND R. EVALUATE (~, AT X, [ FA, - S~VO, ]
CONV - I (R*V~N) I (1~11 lV~.I) I[
NO
IS~ x - x. I
4. Compute the new solution (X N) the value of optimum step length c~', subject to a
maximum of three iterations. Convergence criterion for
5. Test the new solution for optimality and if cubic interpolation is given by,
optimal, return
(ST V @)
6. Update the direction matrix and go to step 2
_< 0.001 ...(21)
IISll IIV~ll
The convergence criterion for unconstrained
minimization is taken as II VO N II <: 0.001 subject
to a maximum number of iterations equal to the number Three situations can be identified for design/
of design variables, where., II VO N U = "~,Y__,(VONIz). reorganization of a water distribution system, viz. new
Cubic interpolation technique is employed for obtaining design, augmented design and selective augmentation.
108 L Gupta, J. K. Bassin, A. Gupta, P. Khanna
@
L can be split in two sections of lengths L u and
(L - L ) in the next upper and lower commercial ®). ' '
sizes (D u and Dr) such that their combined hydraulic SO 15
characteristics are the same as fl~at of the non- t4 (~)
@ ,
commercial diameter pipe. Noting that the head-loss ! t
t |
in the non-commercial diameter pipe is equal to the |
@
sum of head-losses in two sections, i.e. f = f+f~, in
t !
which f is the head-loss in the upper diameter section ! i
t !
and f~ is the head-loss in the lower diameter section,
the following expression is obtained for computing the 1, @ ,' 18
The computer code has been written in --- NEW PIPE I..INK
FORTRAN-77 for use on an IBM-PC/AT-386 @ 33
MHz computer, with or without math coprocessor Fig. 5. Typical water distribution system.
support, 640K bytes memory, 1.2M bytes high capacity
floppy disk drive and a hard disk drive, under MS- The network, depicted in Fig. 5, comprising 30
DOS operating system. The programs have been pipes (24 existing and 6 new) and 18 nodes including
compiled using Microsoft FORTRAN Optimizing 16 demand nodes (all nodes except nodes 7 and 13),
Compiler Version 4.1 and linked using Microsoft is supplied through a single reservoir of 20 m height
Segmented-Executable Linker Version 5.01.20. The at node 7 with ground level of 1(11.3 m. The ground
maximum size of the distribution system that can he level at different demand nodes varies between 100.5
designed using the software is limited to 150 pipes, m and 102 m. The network is similar to the one
105 nodes, 100 demand nodes and 5 reservoirs. considered in Swamee et at. [41. Constraints on
Incorporating these limitations the size of the executable minimum nodal pressure and pipe diameter are taken
program is 440K bytes. The programs have also been as 12 m and 50 mm respectively. Cast iron (CI) pipes
compiled on HP-9050-AM computer system under I-IP- have been employed in the design for diameters of
UX operating system. 80 mm and above whereas galvanized iron (GI) pipes
have been used for diameters less than 80mm. The
coefficient of roughness for old and new pipes are
EXAMPLE taken as 0.7 and 0.9 respectively. The demand peak
factor (ratio of peak and average daily demands) of
To illustrate the usefulness and applicability of the 2.5 has been considered for design.
algorithm and software, it was used to design a typical
water distribution system having existing and proposed Two types of studies have been carried out using
pipe links including expansion into new areas. this network to meet an increase in demand.
Optimi=ation of water distribution system 109
TABLE 2a
O P T I M A L DESIGN W I T H S P U T C O M M E R C I A L D I A M E T E R PIPES
TABLE 3a
diameters ate the decision variables. In most of the 4. Swamee. PK. and Sharma" A.K. Reorganization of Water
Distribution System. Journal of Environmental Engineering.
water supply systems serving areas with high population
ASCE 1990, 116(3), 588-600.
density the cost of reservoirs, pumps and power are
dominant. Accordingly, the objective function includes 5. Watamat~a. T. l..~a,.~tCost D~ign of Water Distribution System.
cost of pumps, power and reservoirs along with the Journal of the Hydraulics Division. ASCE 1973. 99(I-IY9),
1497-1513.
cost of pipes. The software developed runs on IBM
PC/AT-386 in a reasonable time (approximately 1 hou0 6. Cenede~, A., Gallerano, F., and Misiti, A. Multiobjecdve
for a medium size network. Analysis in Optimal Solution of Hydraulic Networks, Jonrnal
of Hydraulic Engineering. ASCE 1987, 113 (9), 1133-1143.
The example study presented in the paper 7. Chiplunkar. A.V., Mehndiratta" S.L, and Khanna" P.. Looped
demonstrates the efficacy of the optimization algorithm. Water Dixtribution System Optimization for Single Loading.
Various options of augmentation/expansion of an Journal of Environmental Engineering, ASCE 1986, 112(2)
existing system can be considered, in a reasonable 264-279.
time using this software on an inexpensive computer 8. Lan.sey, K.E. and Mays. L.W~ Optimization Model for Water
like an IBM-PC/AT, for appraisal of different solutions, Distributi,tm Sy~em Design, 1ournal of ltydraulic Engineering.
such as, whether a new reservoir is required or ASCE 1989, 115(10), 1401-1418.
strengthening the existing network with parallel pipe
9. Jain. A.K., Moran, D.M., and Khanna, P. Modified i[a~en-
lines is adequate for meeting the increased demand. Williams Formula, Journal of the Environmental Engineering
Division. ASCE 1978. 104(EEl). 137-146.
In summary, use of the algorithm presented in
I0. Schedule of Rates, Maharaahtra Water Supply and Sewerage
this r e . a r c h would provide a useful tool t0 the
Board. Office of tim Superintendent "Engineer, Environmental
practicing engineers and decision makers in less Engineering Works Division, Nagpur. 1990-91.
developed countries in arriving at least cost solutions
to water distribution system design problems. II. Bouh~. P. and Airman. T. A Graph-theoretic Approach to
Explicit Nonlinear Pipe Nctwork Optimization, Applied
Mathematical Modelling. 1991, 15. September, 459-466.
I:eatl~tstone. R.E. i~d F,l-lumaily, K.K. Opllmal Diameter 13. Rat, S.S. Optimization theory and applications. 1987, Wiley
Selection fin" Pipe Networks. Journal of the Ilydraulics Division, F..a.~tem Lid. Bombay.
1983, ASCI~ IO9(|1Y2), 221-234.
14. llimmelblau. D.M. Applied Nonlinear Programming. 1972.
2. Walski. T.M.. C,ex~ler. J. and Sj~trom, J.W. Selecling Optimal McGraw-llill Book Company, New York.
Pipe Sir,,t for Water Distribution Syxtems. Journal of American
Water Works Association, 1988, 80(2), 35 - 39. 15. Manga.,~arian, O.L., Nonlinear Programming, 1969, McGraw
llill, New York.
3. Ouindry, G.E., Brill E.D. and Liebman, J.C. Optimization
of looped Water Distribution Sy~tem.~. Journal of the 16. Gupta, [. and Vartak, M.N., Kuhn-Tucker and Fritz-John type
Environmental Engineering Division ASCE 198 I. 107(EF.,4). sufficicnt optimality conditions foe generalized .~mi-locally
665-679. ctmvex programs, Opsearch. 1989, 26(I), 11-27.