Professional Documents
Culture Documents
ABSTRACT: A new approach to the least cost design of water distribution net-
works has been developed. The method is based on traditional techniques of
pipe network analysis by computer; initially assumed pipe diameters are suc-
cessively adjusted using an optimization technique which is based on a cost
function incorporating the various capital and operating costs. Final design total
costs for a given network have been found to be approximately the same re-
gardless of the initially assumed values of pipe diameter although the final ar-
rangements of pipe diameter are not necessarily identical. Since the method
considers all physical components of the network, both from the point of view
of the flow and pressure distribution and the costs, the design represents the
global optimum as opposed to a typical element optimum which has been con-
sidered by other workers. The method contains no constraints such as assumed
fixed pressure heads or pipe flows and may be applied to either closed loop
or open loop networks of any size. Provided a computer program for analyzing
pipe networks is available, the optimization technique developed represents a
simple addition to the basic program.
INTRODUCTION
221
sum of the equivalent lengths in each and every loop equals zero. How-
ever, Raman and Raman (6) have shown mathematically that the alge-
braic sum of Le/Q must be zero instead of 2Le. However, the require-
ment of known pressure surface in this method leads to some limitation.
Jacoby (4) in 1968 proposed a nonlinear programming technique of
gradient variables. The final design is obtained by selecting the size of
commercial pipe closest to the theoretical diameter. The method is very
complex and was only applied to a simple two-loop system. Deb and
Sarkar (3) developed the equivalent pipe diameter method for network
optimization using an assured pressure surface profile capital cost func-
tions for pipes, pumps, and reservoirs. This method, as in the case of
the equivalent length method, has two major drawbacks: firstly it lacks
mathematical justification for cost equivalent pipes; and second, a hy-
draulic pressure surface over the network must be artificially created
(Swamme and Khanna (7)).
While others have investigated the application of nonlinear and dy-
namic programming techniques there does not appear to be a method
applicable to large closed loop networks, which at the same time incor-
porates all the important factors affecting the total cost of the mains,
pump capacity, friction losses, energy, and operating costs. Any method
which achieves these objectives should be, at the same time, amenable
to automatic computation and simple to use by the water engineers. The
method of optimization described in this paper adopts the cost functions
of the most important elements of the water supply scheme and is based
on a concept that there is a unique hypothetical hydraulic gradient
which results in an optimal design.
The method overcomes certain limitations of previous methods in that
the initial flow distribution does not remain constant until the end of
the solution but is correctly balanced using the Hardy Cross or nodal
method. It is not necessary to assume a pressure surface profile since
this is computed from the network analysis. As opposed to concepts of
equivalent diameter or length, actual lengths are used in the analysis
and assumed diameters are finally transformed into actual commercial
sizes using an explicit function. The mathematical concept of optimi-
zation is based on the whole system thus resulting in global minimum
as opposed to other methods in which the optimization is based on a
typical elemental loop. All cost components including pipelines, pumps,
tanks, as well as operating costs, are incorporated in the total cost
function.
FIG. 1.—Rotation of Linear and Nonlinear Energy Lines About Network Inlet
line, the diameters of the individual lengths between the submains cor-
responding with the known discharges and pressure gradients being
obtained from the Hazen Williams formula. However the cost difference
between the results of using the straight energy and the optimal energy
gradient lines was found to be of the order of only 2%.
Cenedese and Mele (1) proposed a method which starts by assuming
that the most economical distribution system is always an open network.
However, Lischer (5) commented that the method was applicable only
to a system with a constant input head without pumping and that the
cost function included only capital cost parameters. In the present work
therefore, the basic concept is that a hypothetical linear gradient for a
particular balanced network exists by which the initial design can be
corrected to produce an optimal relation between the scheme component.
The equation for total annual cost of the scheme is obtained by sum-
ming the functions for fixed and variable costs. The annual costs can be
determined as the product of the initial investment and the capital re-
covery factor which is in turn a function of the rate of interest and the
expected life of the installations. Operational costs are generally asso-
ciated with pumping and are functions of power consumption and as-
sociated labor.
In the present work the cost functions as derived by the UK Water
Research Centre (WRC) (2) have been adopted but any other empirical
functions relevant to local conditions can be used. C is the cost which
224
Pumping Installations
Installation cost
C = hQ^tf3 (3)
3
in which Q = discharge, in m /s; and H = total head, including static
lift (w).
For UK conditions the WRC obtained:
C = 0.0229 Q081H0-43 (4)
In the case where, in the first instance, the head, H, is not known by
the designer the following cost function was obtained, from typical char-
acteristic pump (HvQ) relationships:
0 = ^0* (5)
and for UK conditions:
C = 0.160 Q 077 (6)
Operating Costs:
pgQHTF
C=— (7)
lOOOri ^ '
in which T = number of hours of operation per annum; F = energy cost
per kWh; and T| = overall pump efficiency.
If the capital recovery cost is based on a life of Y years then the op-
erating cost of the pumps is given by:
pgQHTFY
C=— (8)
v
lOOOii '
Balancing Tanks
C = ^V* (9)
3
in which V = volume of tank, in m . The values of dx and d2 depend on
the type of construction material and the country in which the tank is
built.
225
FIG. 2.—Shortest Actual Flowpath between Inlet and Point of Minimum Pressure
Head Elevation
Gradient, So
FIG. 3.—Variation of Global Cost Function, CT, with Hypothetical Optimum Hy-
draulic Gradient S„
227
x
X
© ©
1000
Li x © JU6O°© y
7 | 6000 l/mln
i i
© ©
Legend:
Pipe n u m b e r - * » M J
length , m. - ^ gg
node number
outflow,
ow, Il / m l n . « ^ g ^ ,
elevation ,
1. Identify the pipe layout and input the pipe parameters, pumps,
and tanks, define the minimum permissible pressure head elevation,
and specify the cost functions.
2. Adopting equal diameters for the network elements, carry out a
hydraulic analysis using the Hardy Cross or nodal method as appro-
priate, using an arbitrary inlet pressure head elevation.
3. From the computed pressure head elevations at the nodes identify
the shortest actual flow line leading from the inlet to the point of min-
imum pressure head elevation. Adjust the latter to be equal to the min-
imum permissible head elevation and thus determine the actual inlet
pressure head elevation and the pumping head.
4. Calculate the cost of the initial design.
5. Calculate the optimal 'dummy' hydraulic gradient of the initial
design.
6. Determine the diameters of the network elements using Eq. 17 and
round off to nearest commercial size.
7. Carry out a hydraulic analysis using the revised diameters and re-
peat step 3.
8. Calculate the cost of the new design and compare with the previous
cost.
9. Repeat steps 5 to 8 as necessary until the least cost design is
obtained.
229
230
2 25 20
3 25 7.5
4 25 15
5 25 15
6 25 15
7 25 15 304,406.1 299,892.2
8 25 15
9 25 15
10 25 7.5
11 25 35
12 25 30
13 25 25
at the same elevation (but this is not essential since the hydraulic anal-
ysis does not depend on pipe elevations and the level of the minimum
permissible pressure head elevation is specified); (2) the minimum re-
quired residual pressure head is 10.0 m above the pipe axes; (3) the
maximum water demand rate is 10,000 L/min with a daily average of
3,000 L/min; (4) the life of the scheme is 15 yr; (5) the supply is pumped
from a local ground water source the the level of which, during pump-
ing, is taken to be 15.0 m below the level of the pipes (The head at the
pump is therefore H = Hf + 15 + 10 + h^ in which Hf = maximum head
loss between the inlet and the point of minimum pressure head eleva-
tion across the network; and hfp = head loss in the pumping main); (6)
231
1 25 25
2 25 25
3 25 7.5
4 25 20
5 25 20
6 25 20
7 25 15 412,560.0 408,899.0
8 25 15
9 25 20
10 25 7.5
11 25 35
12 25 35
13 25 25
the cost functions are based on 1977 figures; and (7) pump operating
cost is $0.05 per kWh. Using the program described, the design diam-
eters for the minimum total cost are shown in Tables 2, 3, 4, and 5 for
different assumed initial diameters.
To show the sensitivity of the design to pumping cost the pumping
cost per kWh was doubled and the new optimal design obtained for the
set of initial diameters given in Table 4. The results are summarized in
Table 6.
A method for the optimal least cost design of new water distribution
pipe networks has been developed. The design procedure is based on
either the head balance or quantity balance methods of analysis of flows
and pressure distributions in the network of initially assumed pipe sizes.
These initial diameters, individually, are successively and systematically
adjusted until the global cost of the network is at a minimum. After each
adjustment the network is reanalyzed. The new diameters are obtained
explicitly using an expression which is a function of the pipe resistance
equation in which the hydraulic gradient is a hypothetical or dummy
optimum. This latter gradient is based on the concept that the minimum
cost of a pipeline consisting of elements of different diameter in series
is obtained when the diameters are such that the hydraulic gradient over
the whole length forms a single straight line. The equivalent optimal
dummy gradient for the network is calculated from an objective function
which is a combination of the first derivative of the global cost function
of the network and the pipe-resistant function. The cost function can
include the capital and operating costs of the elements of the network
including pipes, pumps, and storage tanks.
An example of the application of the method to the design of a water
232
sizes if desired or, if certain pipes are given a large initial diameter in
relation to the rest they will tend to remain larger in the final design.
Provided that a computer program for pipe network analysis is avail-
able the optimization procedure presented in this paper represents a
straightforward additional subroutine to the basic program. The m e t h o d
contains no prescribed constraints such as fixed pressure h e a d s or pipe
flows and considers the computed flows a n d h e a d in the whole net-
work—thus producing a global optimum design.
APPENDIX I.—-REFERENCES
1. Cenedese, A., and Mele, P., "Optimal Design of Water Distribution Net-
works," Journal of the Hydraulics Division, ASCE, Vol. 104, No. HY2, Proc.
Paper 13568, Feb., 1978, pp. 237-247.
2. "Cost Information for Water Supply and Sewerage Disposal," Technical Report
TR61, Water Research Centre, Medmenham, U.K., Nov., 1977.
3. Deb, A. K., and Sarkar, A. K., "Optimization in Design of Hydraulic Net-
work," Journal of the Sanitary Engineering Division, ASCE, Vol. 97, No. SA2,
Proc. Paper 8032, Apr., 1971, pp. 141-159.
4. Jacoby, L. S. S., "Design of Optimal Networks," Journal of the Hydraulics Di-
vision, ASCE, Vol. 94, No. HY3, Proc. Paper 5930, May, 1968, pp. 641-661.
5. Lischer, V. C , "Optimal Design of Water Distribution Networks," Journal of
the Hydraulics Division, ASCE, Vol. 105, No. HY1, Jan., 1979, pp. 113-114.
6. Raman, V., and Raman, S., "New Method of Solving Distribution System
Networks Based on Equivalent Pipe Lengths," Journal of the American Water
Works Association, Vol. 58, No. 5, May, 1966, pp. 615-626.
7. Swamme, P. K., and Khanna, P., "Equivalent Pipe Methods for Optimizing
Water Networks, Facts and Fallacies," Journal of the Environmental Engineering
Division, ASCE, Vol. 100, No. EE1, Proc. Paper 10375, Feb., 1974, pp. 93-99.
8. Tong, A. L., et al., "Analysis of Distribution Networks by Balancing Equiv-
alent Pipe Lengths," Journal of the American Water Works Association, Vol. 53,
No. 2, Feb., 1961, pp. 192-210.
9. Wu, I. P., "Design of Drip Irrigation Main Lines," Journal of the Irrigation and
Drainage Division, ASCE, Vol. 101, No. IR4, Proc. Paper 11803, Dec, 1975,
pp. 265-278.
APPENDIX II.—NOTATION
s = hydraulic gradient;
= computed hydraulic gradient along shortest flow
path between inlet and point of lowest pressure
head elevation;
S0 = hypothetical optimal hydraulic gradient;
T = number of hours of pump operation per year;
V = volume of tank;
Y = life of scheme in years;
T) = pump efficiency; and
p = density of liquid.
Subscripts
D = daily value (e.g., of discharge, Q);
i = pipe identifier (integer); and
T = total (e.g., of cost, C).
234