You are on page 1of 18

O P T I M A L VALVE C O N T R O L IN W A T E R -

DISTRIBUTION NETWORKS

By Paul W. Jowitt 1 and Chengchao Xu 2


Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

ABSTRACT: The paper concerns the problem of minimization or leakage in water-


distributed networks. It has been reported that leakage from some networks may
account for a significant amount of the water put into supply. For some aging urban
networks, rates of up to 50% have been quoted, with average rates of 25% being
quite typical. These high rates of leakage represent a significant economic loss.
An algorithm for the determination of flow control valve settings to minimize leak-
age is presented. The nonlinear network equations describing nodal heads and pipe-
flows are augmented by terms that explicitly account for pressure'-dependent leak-
age and by terms that model the effect of valve actions. Successive linearization
of these equations using the linear-theory method allows a linear program that
minimizes leakage to be formulated and solved. The performance of the method
is demonstrated by application to an example network.

INTRODUCTION

The primary duty of a water-supply undertaking is the consistent provision


of water in sufficient quantity, at appropriate pressure, of acceptable quality,
as economically as possible. Whereas considerable effort has been invested
over the last decade in the development of optimization techniques to tackle
the economic pump-scheduling problem, the problem of operational-leakage
control has received relatively less attention. This situation is unfortunate,
since water leakage in a supply network may account for a large percentage
of the total water supplied and can consequently represent a significant eco-
nomic loss. Values as high as 50% have been reported in the case of some
aging and deteriorating urban distribution networks (NWC: "Leakage" 1980;
Twort et al. 1985). In the United Kingdom, the leakage levels vary from
one undertaking to another, depending on the system characteristics, with
average figures being on the order of 24% (Goodwin 1980; NWC: "Leak-
age" 1980).
The economic implications of leakage reduction are all too obvious. Water
leakage not only involves unnecessary expense in pumping and treatment
costs, but may also trigger premature investment to develop new sources or
to expand system capacity to keep pace with increasing demand. Reductions
in leakage will therefore result in lower annual operating costs and, fur-
thermore, deferment of demand-related capital expenditure. As an indication
of the potential benefits, it has been estimated that the total net savings avail-
able from the implementation of a proper leakage control strategy would be
on the order of $30,000,000/yr in Britain alone (Goodwin and McElroy
'Prof., Civ. Engrg. Systems Res. Centre, Dept. of Civ. Engrg., Heriot-Watt Univ.,
Riccarton, Edinburgh, EH14 4AS, U.K.
2
Res. Student, Civ. Engrg. Systems Res. Centre, Dept. of Civ. Engrg., Heriot-
Watt Univ., Riccarton, Edinburgh, EH14 4AS, U.K.
Note. Discussion open until December 1, 1990. To extend the closing date one
month, a written request must be filed with the ASCE Manager of Journals. The
manuscript for this paper was submitted for review and possible publication on March
23, 1989. This paper is part of the Journal of Water Resources Planning and Man-
agement, Vol. 116, No. 4 , July/August, 1990. ©ASCE, ISSN 0733-9496/90/0004-
0455/$ 1.00 + $.15 per page. Paper No. 24893.

455

J. Water Resour. Plann. Manage. 1990.116:455-472.


1983); even this estimate may be regarded as conservative. For these rea-
sons, together with the increasing contamination of existing supplies and
difficulty of obtaining new ones, more advanced control techniques for min-
imizing water leakage are needed.
It is well known that water leaking from a supply-and-distribution network
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

is directly related to the system service pressure. The relationship between


leakage and pressure shown in Fig. 1 is based on data from a set of field
experiments undertaken by the Water Research Centre (Goodwin 1980; NWC:
"Leakage" 1980). (It should be noted that leakage is not solely dependent
on the service pressure, and the vertical axis in the figure represents an index
of leakage rather than actual leakage volume.) It can be clearly seen from
the figure that the leakage ratio increases proportionately with the increase
in the average service pressure, with the curve steepening at higher pressure.
Thus, reductions in high service pressures will result in considerable reduc-
tions in leakage. From an economic viewpoint, it is therefore desirable to
regulate excessive system service pressures to acceptable levels; although
water companies are increasing their efforts in this regard (e.g., Wessex
Water, Thames Water), it is rare for the valve-control system to be integrated
into an operational-control and telemetry system. More usually, the valve is
of the constant-flow or constant-downstream-head type, actuated by local,
closed-loop feedback control [as described, for example, by Ratcliffe (1986)].
Pressure reduction can be achieved in a number of ways, from the re-
duction of pumping heads (for variable-speed pumps), establishing pressure
zones through the use of a variety of pressure-reducing valves, or by the
installation of break-pressure tanks. (The latter produce a discontinuity in
the hydraulic grade line by routing the flow via a float valve into a constant-
head tank.) With the general increase in the use of computerized control
systems and associated telemetry systems, the use of remotely controlled
flow-control valves may offer a cost-effective way of achieving pressure zoning
and reductions in system leakage. The operation and performance of such
valves are described by Ratcliffe (1986). Whilst valve action may be of the
plug or globe type, valve actuation is electronically controlled. It is this latter
feature that offers the opportunity for the form of integrated control described
in the present paper. Implementation of such pressure control can yield re-
ductions in leakage of up to 20-30% (Bargiela 1984; Bessey 1985a, b; Ger-
manopoulos and Jowitt 1989; Miyaoka and Funabashi 1984).
Previous research on the determination of valve settings to reduce network
pressures, and, inter alia, leakage, has formulated the problem as minimi-
zation of system overpressure subject to system operational constraints. Some
of these approaches (Bargiela 1984; Miyaoka and Funabashi 1984) have at-
tempted to resolve the problem without directly considering the leakage within
the network model. More recently, pressure-dependent leakage components
have been explicitly incorporated within the system model (Germanopoulos
and Jowitt 1989). This is believed to be an important step forward.
The present paper presents an algorithm that extends the work reported in
Germanopoulos and Jewitt (1989) by seeking to minimize the system leakage
directly, rather than merely minimizing the system overpressures. The po-
tential economic benefits from such a control scheme are evaluated by com-
parison of the leakage volumes resulting in the controlled and uncontrolled
cases. The usual nonlinear network equations, describing the nodal heads
and pipeflows, are augmented by terms that explicitly account for pressure-
456

J. Water Resour. Plann. Manage. 1990.116:455-472.


dependent leakage and by terms that model the effect of valve actions. Suc-
cessive linearization of these equations using the linear-theory method (Wood
and Charles 1972) allows a minimum-leakage linear program to be formu-
lated and solved. The linearization of the flow-control-valve terms used in
the present paper is different from that used by Germanopoulos and Jowitt
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

(1989) and leads to a more efficient solution to the prqblem.

100

90

80

70

60

"J 50
< 1
<
40

30

20

10

20 40 60 80 100
AVERAGE ZONE NIGHT PRESSURE (m)

FIG. 1. Relationship between Leakage and Pressure


457

J. Water Resour. Plann. Manage. 1990.116:455-472.


PROBLEM FORMULATION

A water-supply and distribution system consists of a collection of nodes


that are interconnected by various elements such as pipes, valves, pumps,
reservoirs. Each element in the network is characterized by a mathematical
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

function that describes the relationship between the element flow and the
head difference between the two ends of the element. The form of the re-
lationship depends on the physical characteristics of the element. The sys-
tem-governing equations can be formulated in accordance with the following
two familiar rules: (1) Nodal mass balance—the algebraic sum of all the
inflow and outflow at each node is equal to zero; and (2) Energy conser-
vation—the total sum of the all head losses around any loop in the network
is equal to zero.
There are several pipeflow formulas available that relate flow and head
loss under various flow regimes. The most typical are the Darcy-Weisbach-
Colebrook-White and Hazen-Williams equations (Twort et al. 1985). The
Darcy-Weisbash equation in conjunction with Colebrook-White formula is a
more accurate representation of wide-ranging flow conditions but requires
more computational effort; the Hazen-Williams equation is not really valid
for flows throughout the laminar, transition, and rough regimes, but it has
the advantage of computational simplicity and therefore receives wider ap-
plication. By using the Hazen-Williams equation, the flow-head-loss char-
acteristics of a pipe can be modeled as follows:

fiff (la)
~zr-sgn(ff,-ff,).|ff,-/ryr
Q9 = Rtj-SgniH, - Hj) • \H, - tf/'54 (lb)
In Eq. la and b, Qv represents the flow (1/s); Dtj = pipe diameter (m); Ltj
= length of the pipe (m); Ht - head at node i (m); CHWV = Hazen-Williams
coefficient for the pipe that connects nodes i and j ; a = a constant whose
value depends on the units used (for the units used herein, a has the value
278.534); Sgn(x) = sign of x. Some of the elements of Eq. la do not vary
with head or flow, and it is convenient to gather these together as a com-
posite term Rv, as shown in Eq. lb.
In the case of a flow-control valve located between nodes i and j , the
relationship between the flow and head loss can be expressed in a form
analogous to Eq. 1:
Q,j = V{k) -Ry Sgn(Ht - Hj) • \H, - Hj\0-54 (2)
where V(k) = a parameter that represents the setting of the Ath flow-control
valve. In this equation, the behavior of a fully closed valve is represented
by setting V(k) to zero. Alternatively, a value of V(k) of unity would rep-
resent a fully open valve, which produces no head loss beyond that in the
pipe element. In general, there may be physical limits to V(k) that restrict
its value to some range in the interval zero to unity. In the development to
follow, the question as to which type of valve is used is left open; the pur-
pose is simply to demonstrate the efficacy of operational valve control in
minimizing uncontrolled leakage. It should also be noted that some valves
may only permit flow in one direction. This aspect is not considered ex-

458

J. Water Resour. Plann. Manage. 1990.116:455-472.


plicitly in the development; it is anticipated that doing so would cause no
difficulties beyond those reported by others elsewhere regarding the con-
vergence of the numerical method used to solve the nonlinear equations de-
scribing the hydraulic behavior of the network.
The nonlinear relationship between the leakage and average service pres-
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

sure indicated in Fig. 1 can be approximated by the following function (Bar-


giela 1984; Germanopoulos 1988), and which derives from experimental data
(NWC: "Leakage" 1980):
QSy = CLyLy(Py)Lla (3a)
Lla
QSy = RSy(Py) ( (36)
In Eq. 3, QSy = water-leakage volume occurring in the pipe element of
length Ly spanning nodes / and j ; CLU = a coefficient that relates the leakage
per unit length of pipe to service pressure and depends on the system char-
acteristics (for example, the age and deterioration of the pipe and the soil
properties etc.); RSy = a composite parameter representing the product of
CLy and Lv; Py = average service pressure that can be approximated by the
average pressure relative to the ground level at two ends of the element:
Py = 0.5 • [(H, - G,) + (Hj - G,)] (4)
where G, = ground level at node i.
It might be noted at this point that the exponent in Eq. 3 is quite different
from the value of 0.5 which commonly characterizes the relationship be-
tween flow through an orifice and head difference. It is reiterated that Eq.
3 is based on field data, and it is speculated that the higher exponent of 1.18
incorporates any opening up of pipe joints/cracks caused by the difference
in internal and external pipe pressures.
The leakage QSy from element ij is apportioned equally to the nodes ;' and
j . The nodal mass balance thus takes the form:
2 Qu + 0.5 • 2 QSy + C, = 0; i=\,...,N (5)
j<ER, jeRi

where C, = demand consumption at node /; N = total number of nodes with


unknown heads; and i?, is the set of nodes connected to node i.
Substitution of Eq. 4 into 3, and then Eqs. 1-3 into Eq. 5, produces a
set ofN simultaneous equations in the N unknown heads. The heads, flows,
and leakages can be determined uniquely by solving these nonlinear equa-
tions; this is usually referred to as network static analysis. Several methods
have been developed for the solution of the resulting system of equations,
including the Hardy-Cross method (Cross 1936), Newton-Raphson (Martin
and Peters 1963; Shamir and Howard 1968), the linear-theory method (Wood
and Charles 1972; Isaacs and Mills 1980) and the gradient method (Todini
and Pilati 1987).
For the leakage-reduction-control scheme, the task of minimization of sys-
tem leakage can be regarded as determining the optimal flow-control-valve
settings V(k) under a system of network and operational constraints in order
to minimize the sum of the leakage terms QSy.
Mm^ QSy ' (6)

459

J. Water Resour. Plann. Manage. 1990.116:455-472.


where R = a set consisting of all pairs of connected nodes.
The minimization of Eq. 6 requires that the leakage terms QSy be func-
tionally related to the control variables V(k) and that the hydraulic and any
other constraints are satisfied. The effect of valve action is modelled in Eq.
2 and can be incorporated into the optimization problem by superimposing
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

the system-describing equation (Eq. 5) as a constraint. The control strategy


must not be at expense of failing to achieve continuity of the consumers'
supply; this can be ensured by requiring nodal heads to be maintained above
specified minimum values. Restrictions on some critical demand nodes such
as those with highest elevation, remoteness from the source, and maximum
load etc. will generally dominate the optimization problem. If the constraints
on these are satisfied, the head constraints elsewhere can be satisfied im-
plicitly. The minimum acceptable head profiles at these critical nodes can
be incorporated into the optimization problem as
Ht>Hf; i = 1, ..., NR (7)
where NR = number of the reference nodes selected; Hf = minimum pres-
sure requirement at selected reference node i.
These critical nodes can be chosen by close examination of the topography
and load pattern of the system. They can also be identified through inves-
tigating the operational records, examining which parts of the system fre-
quently experience low pressure. Alternatively, such nodes can be identified
through network simulation using maximum consumer demands. Constraints
on valve operational range can be expressed as
Vm[n(k) < V(k) < Vmax(k); k= 1, ..., K (8)
where K = number of flow-control valves in the network.
The minimization of the leakage is achieved through the solution of the
following mathematical program:
Min ^ QSy (total system leakage) (9a)
i/ER

subject to:

jeR,- J<ER,

i = 1, ..., N (nodal mass balance) (9b)


Ht>H*; i = 1, ..., NR (minimum head requirements) (9c)
min max
V (i) < V(i) < V (r); i=l,...,K (valve-control restrictions). .. (9d)
together with the associated equations (Eqs. 1-4):
Qy = Ry• Sgn(Ht - Hj) • \H, - # / 5 4
(flow/head loss relation in an element ij with no valve) (9e)
0 54
Qv = V(k)-R,j • Sgn(Ht - Hj) • \H, - Hj\ -
(flow/head loss in an element ij with a valve) (9/)
ln
QSy = Ry • (PtJ) (leakage/pressure relationship) (9g)

460

J. Water Resour. Plann. Manage. 1990.116:455-472.


Ptj, = 0.5 • [(Hi - G,) + (Hj - G,)] (pressure/head relationship) (9ft)
The foregoing program for optimal valve control is highly nonlinear, with
a nonlinear objective function and generally nonlinear constraints because of
the nonlinearity in Eqs. 1 , 2 , and 3. There are several iterative techniques
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

available for its solution, including the steepest-descent method, the con-
jugate-gradient method, and the sequential-minimization method (Lasdon
1970). However, the method employed herein is a combination of the linear-
programming and the linear-theory methods; the latter has been widely used
in the analysis of water networks (Wood and Charles 1972; Isaacs and Mills
1980).
i

LINEAR-THEORY METHOD

The linear-theory method was initially developed in a loop formulation to


determine the set of unknown flows (Wood and Charles 1972). More re-
cently, the method was developed to solve for the nodal heads (Isaacs and
Mills 1980). In both cases, the net inflows/outflows to the network were
assumed known, which is perhaps a reasonable assumption at the design
stage but is less tenable in the context of operational behavior. The extension
of the linear-theory method to accommodate pressure-dependent leakage has
been presented elsewhere by Germanopoulos and Jowitt (1989). The incor-
poration of pressure-related leakage and consumer demands within the New-
ton-Raphson solution scheme has also been developed by Germanopoulos
(1985). A summary of the linear-theory method for the problem under dis-
cussion here is the following:
The pipe flow/head difference relation in Eq. 1 can be rewritten:
\QlK • Qij = RlM -(Ht-Hj) (10)
For the pressure-dependent leakage terms, Eq. 3 after being combined with
Eq. 4 can be expressed in like fashion:
QSy015 • QS,j = 0.5 -RS^S -(Hi + Hj-Gt-Gj) (11)
(in which it can be assumed that the minimum-head constraints ensure that
Ht + Hj - G, - Gt> 0).
After n iterations, estimates of all nodal heads H" associated with the
estimates of pipe flow Q"j and pipe leakage QS"j are known. A linear ap-
proximation is imposed on this local condition. For a pipe element, the li-
nearization can be obtained from Eq. 10.

Qfx = A-JJ- (ff?+1" - Hfl) (12a)


where

*S = j ^ T i i - - - - ( 12fe )

Similarly, the leakage term can be linearized by


QSf = KSIj • (Hf1 + HJ+1 - d-Gj) (13)
where

461

J. Water Resour. Plann. Manage. 1990.116:455-472.


KSIj = 0.5-RSlj^-QSf5
The flow through the control valve can be approximated in a form similar
to Eq. 12 for a given fixed setting:
fif' = K'r(fl1+l - HJ+1) (14)
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

where

icsr
Substitution of Eqs. 12-14 into Eq. 5 results in a set of simultaneous
linear equations that can be solved readily. The process is repeated until
sufficient convergence is obtained. Numerical experiments show that con-
vergence can be accelerated and ensured if the estimates of flow and leakage
used to linearize Eqs. 12-14 take the following forms after the first iteration:
&j = 0.5 • (fig + Qf1) (15a)
QSl = 0.5 • (QS"9 + QSp) (15fe)
The computational flows Qy and leakage QSy satisfy the continuity rule since
Qh QSlj, Qf1, and QSnfl satisfy Eq. 5 after the first iteration. The iterative
procedure is terminated when the maximum flow discrepancy (Sg.j = Q"j —
Qlrl) in successive iterations is less than some specified tolerance e.
Max|82,y| < e (16)
ijER

Eqs. 12 and 13 can be used to linearize the pipe flow and leakage terms
within the program (Eq. 9). The only remaining nonlinearity is that involving
the unknown valve settings in Eq. 14. The original expression (Eq. 2) in-
troduced to model the effect of each valve can be rewritten as follows:
Qn = V(k)-qtj (17)
where qtj is termed the fictitious pipeflow.
By forming the total differential, the nominal change in the valve flow
can be expressed:
Afi,y = V{kf-Lqij + ,5 • AV(*) (18)
The valve flow at the current iteration can be considered
e r ' = Qlj + Afi„ (19)
Substituting Eqs. 17 and 18 into Eq. 19, and noting the identities AV(&) =
W + 1 • - V(k)n and bq0 = qf' - <fi, gives:
Qf = WW • <?r]] + [qlj • V(k)n+1] - [V(kf •«}] (20)
In a previous approach by Germanopoulos and Jowitt (1989), the valve
flow was given in the following manner
QT1 = «S+I - [i-o - v(k)n+l]-qi (2i)
Notice that the linearization formula (Eq. 20) is different from Eq. 21 in
that the valve setting V(k)n at the previous iteration is incorporated; the new
462

J. Water Resour. Plann. Manage. 1990.116:455-472.


approximation leads to more rapid convergence. The fictitious pipe flow can
be approximated in a similar way to Eq. 12:
qfn = Klj • (H?+1 - HJ+1) (22a)
where
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

Kj {22b)
' ~wir--
SOLUTION VIA LINEAR PROGRAMMING

Substitution of Eqs. 12, 13, 20, and 22 into the program (Eq. 9) leads,
after n iterations, to the following linear program for the valve-control prob-
lem:
Min 2 KSIj'iHr1 + HJ+1) - ^ KS"U (G, + G}) (23a)
ijER IjeR

subject to
0.5 • 2) KSnv• [H?+l + HJ+1 - (G, + Gj)] + C, + ^ Kny- (H?+1 - H]+i)
j^Rt jeRi
+ ^ iV(k)n • Klj • (Hfl - H'J+l) + q\ • V{k)"+l - ql • V(k)"] = 0;
MRl

i = 1, ..., N (23b)
+i
H? >Hf; i = 1, ..., NR (23c)
min n+l raa
V(k) < V(k) < V(fc) *; k = 1, ..., K (23d)
and
H, > 0; i=l,...,N (23e)
where the objective function term X KS"j (G, + G,) = a constant during the
optimization and can be stripped out; R] and Rj = subsets of Rt that represent
the sets of connections to node i through a pipe and a flow-control valve,
respectively. The quantities in the program (Eq. 23) with a superscript n are
regarded as constants; those with a superscript of n + 1 represent the de-
cision variables.
After solution of the program (Eq. 23), the new values for the heads and
valve settings are used to relinearize the hydraulic constraints for the next
iteration of the linear program. The linear-theory method is thus implicit
within the linear program. Iterations continue until convergence is obtained.
The iterative procedure of the proposed approach can be summarized as fol-
lows:

1. Set iteration number n = 0. Guess initial valve settings V(k). (Typically,


the valves are initially set fully open.)
2. Initialize the element-flow and leakage terms and carry out a static analysis
using the linear-theory method to produce a set of hydraulically consistent nodal
heads H°, element flows gy, and leakage values QS°,. [Isaacs and Mills (1980),
suggest that initialization of the element flows at values equivalent to flow ve-
463

J. Water Resour. Plann. Manage. 1990.116:455-472.


locities of about 1 m/s is generally satisfactory.]
3. Check the resultant nodal heads Hi to ensure that they satisfy the mini-
mum-head constraints (Eq. 7). If not, reset the valve settings and return to step
step 2.
4. Using the estimates of leakage terms QS"Jt linearize the objective function
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

according to Eq. 13.


5. Linearize the system hydraulic constraints, Eqs. 12, 13, 20, and 22, using
the computational flows Q"j and leakage QS"j.
6. Obtain the new nodal heads H"+1 and valve settings V(k)n+1 by solving the
resulting linear program (Eq. 23).
7. Update the element flows and leakage terms using Eqs. 12, 13, 20, and
22 based on the new heads and valve settings obtained at step 6.
8. Calculate the discrepancies in the element flows and leakage terms between
the last two successive iterations. If all the differences are within the specified
tolerance, the program terminates. Otherwise update the iteration counter n = n
+ 1, obtain the computational flows and leakage terms from Eq. 15 and go to
step 4.

NUMERICAL EXAMPLES

The foregoing procedure has been used to test a number of networks of

Legend
_r~|_ Fixed head node
0 Junction node
_ c < _ Flow control valve

FIG, 2. Example Network with 25 Nodes

464

J. Water Resour. Plann. Manage. 1990.116:455-472.


TABLE 1. Node Information of Example Network
Node number Ground level (m) Demand (L/s)
(1) (2) (3)
1 18.0 5.0
2 18.0 10.0
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

3 14.0 f 0.0
4 12.0 5.0
5 14.0 30.0
6 15.0 10
7 14.5 0.0
8 14.0 20.0
9 14.0 < 0.0
10 15.0 5.0
11 12.0 10
12 15.0 0.0
13 23.0 0.0
14 20.0 5.0
15 8.0 20.0
16 10.0 0.0
17 7.0 0.0
18 8.0 5.0
19 10.0 5.0
20 7.0 0.0
21 10.0 0.0
22 15.0 20.0

TABLE 2. Reservoir Nodes


Maximum level Normal level Minimum level
(m above-ordnance (m above-ordnance (m above-ordnance
Node datum) datum) datum)
(1) (2) (3) (4)
23 56.0 54.66 54.5
24 55.5 54.6 54.5
25 55.5 54.5 54.0

different sizes. The performance of the proposed method is illustrated herein


using a network used by others elsewhere (Bargiela 1984; Germanopoulos
and Jowitt 1989), and shown schematically in Fig. 2; the network has 25
nodes, 37 elements, and 3 flow-control valves. The system configuration
and parameters are given in Tables 1, 2 and 3. The coefficients CLy relating
the leakage to service pressure are chosen as 10~5, which leads to 16% leak-
age under uncontrolled conditions. In practice, this value can be estimated
through measurement of night-line flows. The convergence criterion is pre-
specified as 0.05 L/s. The reference nodes are selected as nodes 19, 22, 21
and 13. The minimum acceptable pressure requirements at the selected ref-
erence nodes are defined as 30 m above ground level.
To simulate the dynamic character of a real network, a control scheme is
developed over a 24 hr period. A typical diurnally varying daily demand
465

J. Water Resour. Plann. Manage. 1990.116:455-472.


TABLE 3. Linkage Information of Example Network
Pipe Length (m) Diameter (m) HW coefficient Comment
0) (2) (3) (4) (5)
23-1 606.0 0.457 110.0 .—
23-24 454.0 0.457 110.0 —
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

24-14 2,782.0 0.229 105.0 —


25-14 304.0 0.381 135.0 —
10-24 3,382.0 0.305 100.0 —
13-24 1,767.0 0.475 110.0 —
14-13 1,014.0 0.381 135.0 —
16-25 1,097.0 0.381 6.0 —
2-1 1,930.0 0.457 110.0 —
3-2 5,150.0 0.305 10.0 —
12-13 762.0 0.457 110.0 VI
15-16 914.0 0.229 125.0 —
17-16 822.0 0.305 140.0 . —
18-17 411.0 0.152 100.0 •—
20-18 701.0 0.229 110.0 —
19-17 1,072.0 0.229 135.0 •—
20-19 864.0 0.152 90.0 —
21-20 711.0 0.152 90.0 —
21-15 832.0 0.152 90.0 —
22-15 2,334.0 0.229 100.0 —
12-15 1,996.0 0.229 95.0 V2
11-12 777.0 0.229 90.0 —
10-11 542.0 0.229 90.0 —
8-12 1,600.0 0.457 110.0 —
8-10 249.0 0.305 105.0 —
9-8 443.0 0.229 90.0 —
6-8 743.0 0.381 110.0 —
22-8 931.0 0.229 125 —
22-21 2,689.0 0.152 100.0 V3
4-3 326.0 0.152 100.0 —
5-4 844.0 0.229 110.0 —
6-3 1,274.0 0.152 100.0 —
5-6 1,115.0 0.229 90.0 —
7-6 615.0 0.381 110.0 —
5-22 1,408.0 0.152 100.0 —
5-7 500.0 0.381 110.0 —
6-9 300.0 0.229 90.0 —

variation is prescribed as shown in Fig. 3. The prescribed reservoir deple-


tions are indicated through Figs. 4-6, representing the usual practice of re-
filling reservoirs during periods of cheaper nighttime pumping. The optimal
valve operations determined by the optimization procedure presented earlier
are illustrated in Figs. 7-9. The reduction in leakage is clearly shown in
Fig. 10. It is worthwhile noting that leakage reduction during the peak de-
mand period is marginal; leakage reduction is maximized during the night
when consumer demands are lower and system pressures tend to be higher.
The overall reduction in leakage is about 20%. In practice, of course, the
actual cost savings that could be achieved would depend on a number of
466

J. Water Resour. Plann. Manage. 1990.116:455-472.


1 .60
1 .40 •

1 .20
1 .00
0.80
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

O.SO f
0.40
0.20
0.00
0 2 4 6 8 10 12 11 IG 18 20 22 21

Time '

FIG. 3. Daily Demand Variation

(U
>
(D
Hi
l-l

U
0)
<n
0)
PC

FIG. 4. Reservoir (Node 23) Depletion

>

U
•H
0
>
l-l
0)
CO
0)
«

FIG. 5. Reservoir (Node 24) Depletion

467

J. Water Resour. Plann. Manage. 1990.116:455-472.


57,

56L
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

<u
> 55L
(U

u
•H
o 54
>
u
0)
to
<D
Pi c^^| , , , , , , , , , , , I
0 2 4 6 8 10 12 H IS 18 20 22 21
Time

FIG. 6. Reservoir (Node 25) Depletion

factors, including the magnitude and mechanism of the antecedent leakage,


the location of control valves, network topology/topography and the actual
costs of pumping and water treatment. It might be noted, however, that in
many cases the primary cost is that associated with distribution (i.e., pump-
ing), and that the justification of pump-scheduling optimization is often based
on anticipated cost savings of 10-15%. It is clear that for systems with high
leakage rates, the potential savings from leakage control are of a comparable
order.

COMPUTATIONAL DETAILS
The overall procedure just outlined has been coded in Fortran 77 and im-
plemented on the VAX/VMS-8700 mainframe computer at Heriot-Watt Uni-
versity, Edinburgh, United Kingdom. The linear-theory component for the
static analysis used NAG library subroutines F01BRF, F01BSF, and F04AXF
for solving the resulting simultaneous linear equations, which are capable of
exploiting the structure of sparse matrices. The linear-programming problem
is solved by the revised simplex method using NAG library subroutine H01ADF
in which degeneracy is resolved by using the perturbed-problem technique.
The execution time [central processing unit (CPU)] for the example shown
is about 0.4 sec for each iteration. Convergence is achieved in 8-15 itera-
tions.
It has been indicated earlier that linearization of the valve action embodied
in Eq. 20 takes a different form to that used in previous research (Geman-
opoulos and Jowitt 1989). The relative performance of the two methods can
be seen by comparing the maximum discrepancy in the flow and leakage at
consecutive iterations, as shown in Fig. 11. It can be seen that Eq. 20 gives
improved convergence.

CONCLUSION AND ANALYSIS

Water-supply networks are planned and designed to cope with current and
future peak demands; it is inevitable that excessive pressures will exist and
468

J. Water Resour. Plann. Manage. 1990.116:455-472.


J. Water Resour. Plann. Manage. 1990.116:455-472.
69fr
Valve Setting Valve Setting
o o o o o o o o o o — -
O CO 00
O o 8
oO o o
p p
CO »4 J'
T
-t»

f S en .L~~1
N 00
»
Q. 1
_
o
N)
_
I
—.
[ -*
I I CO
—.
00
_.

O

p Leakage Volume (1/s) Valve Setting
o — -
o N) CO oo o
O O o o
oo o
p
i
CO
00 •
Q. f o
! M
i t= a
o _ •
•o 3 -b
0 3
3 CQ •

ont
CO 0)
N'
M
0 00 CO
H
H fl>
0)
< M N)
R
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

o < o
<D
< rs) •
IO
o
o
&ro
p
r
3 s
32

o
a
m
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

a.
<D
U

Iteration

FIG. 11. Convergence Characteristics of Alternative Forms of Valve Lineariza-


tion

will frequently occur, typically at the diurnal scale. Whilst adequate pressure
is essential to maintain a satisfactory service, these excess pressures are un-
desirable because they can lead to significant levels of leakage. To increase
economic efficiency, leakage should be reduced to appropriate levels. The
work described in this paper outlines a method of minimizing leakage through
optimal valve control. Reducing leakage leads to financial saving in annual
operational costs and by the deferment of capital investment to increase net-
work capacity. An additional benefit arising from pressure regulation is the
reduction in the frequency and consequent cost of repair of pressure-induced
mains bursts.
The control scheme for minimizing leakage can be implemented alone or
j in conjunction with other leakage-control and detection methods. The latter
include regular sounding (using a simple sounding rod or a "leak noise cor-
j|! relator"), district metering, waste metering, and combined district and waste
il'll metering (NWC: "Leakage" 1980; Twort et al. 1985). Waste metering in-
ijj volves valving off the supply to a group of 1,000 or so consumers, and
jjj.'l installing a waste meter to record night-line flows. The technique is both
u, , time-consuming and expensive, and in any event serves mainly to locate
! jj1 ! districts with high losses. It should be noted at this point that the current
i!' mathematical representation of pressure-dependent leakage term used in this
paper derives from data collected from a specified set of field experiments,
1
MI i and may not be representative of other circumstances [for example, for data
I j! from some networks in Japan (Miyaoka and Funabashi 1984) water leakage
i !!;,! is found to be a function of service pressure to the power 1.15 instead of
h!' ! 1.18]. For many networks, a calibrated pressure-leakage relationship may
I !''!! not be available, but it is hoped that the technique described in the present
([ ], paper will stimulate further interest, both analytical and experimental, in the
estimation of leakage in distribution networks.
The method developed in this paper can also be used to identify the most
appropriate location of flow-control valves at the planning stage. The prob-
lem of optimal valve location is being investigated and results will be pre-
sented on another occasion. In addition, many actual networks contain pumps
and the effect of valve operation on these would need to be taken into ac-
470

J. Water Resour. Plann. Manage. 1990.116:455-472.


count. Specifically, measures intended to reduce costs attributable to leakage
should not compromise pump operation (either in terms of hydraulic feasi-
bility or cost). It should be noted, conversely, that pump-schedule optimi-
zation (which generally results in increased nighttime pumping) often gives
rise to higher system heads, particularly at times of low demand, and will
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

tend to increase leakage costs. The leakage model usejl in this paper can be
used to quantify any such effects explicitly.

ACKNOWLEDGMENT

The financial support provided in the form of a postgraduate studentship


by the Education Commission of the People's Republic of China and the
British Council is acknowledged.

APPENDIX I. REFERENCES

Bargiela, A. (1984). "On-line monitoring of water distribution systems," thesis pre-


sented to the University of Durham, at Durham, United Kingdom, in partial ful-
fillment of the requirements for the degree of Doctor of Philosophy.
Bessey, S. G. (1985a). "Some developments in pressure reduction." J. Inst, of Water
Engineers, 39(6), 501-505.
Bessey, S. G. (1985b). "Progress in pressure control." Aqua, Vol. 6, 325-330.
Cross, H. (1936). "Analysis of flow in networks of conduits or conductors." Bulletin
No. 286, Univ. of Illinois Engineering, Experimental Station, Urbana, 111.
Germanopoulos, G. (1985). "A technical note on the inclusion of pressure dependent
demand and leakage terms in water supply network models." Civ. Engrg. Systems,
2(3), 171-179.
Germanopoulos, G. (1988). "Modelling and operational control of water supply net-
works," thesis presented to the University of London, at London, United King-
dom, in partial fulfillment of the requirements for the degree of Doctor of Phi-
losophy.
Germanopoulos, G., and Jowitt, P. W. (1989). "Leakage reduction by excessive
pressure minimization in a water supply network." Proc, Institution Civ. Engrs.,
Part 2, 87 (June), 195-214.
Goodwin, S. J. (1980). "The result of the experimental programme on leakage and
leakage control." Technical Report TR 154, Water Res. Centre, Swindon, United
Kingdom.
Goodwin, S. J., and McElroy, S. (1983). "Reducing the water budget loss." World
Water, 7(11), 31-33.
Isaacs, L. T., and Mills, K. G. (1980). "Linear theory method for pipe network
analysis." / . Hydr. Div., ASCE, 106(7), 1191-1201.
Lasdon, L. S. (1970). "Optimization theory for large systems." The Macmillan Com-
pany, New York, N.Y.
"Leakage Control Policy and Practice." (1980). Standing Technical Committee Re-
ports, Number 26, Dept. of the Envir., National Water Council, Britain.
Martin, D. W., and Peters, G. (1963). "The application of Newton's method to net-
work analysis by digital computer." J. Inst, of Water Engineers, 17(2), 115-129.
Miyaoka, S., and Funabashi, M. (1984). "Optimal control of water distribution sys-
tem by network flow theory." IEE Trans. Automatic Control, AC-29(4), 303-311.
Ratcliffe, B. (1986). "The performance and selection of pressure reducing valves,
part 1." WRC Tech. Report 238, Water Res. Centre, Swindon, United Kingdom.
Shamir, U., and Howard, C. D. D. (1968). "Water" distribution analysis." J. Hydr.
Div., ASCE, 94(1), 219-234.
Todini, E., and Pilati, S. (1987). "A gradient algorithm for the analysis of pipe
networks." Proc, Int. Conf. of Computer Applications for Water Supply and Dis-
tribution, Leicester, United Kingdom.

471

J. Water Resour. Plann. Manage. 1990.116:455-472.


Twort, A. C , Law, F. M., and Crowley, F. W. (1985). "Water supply." 3rd Ed.,
Edward Arnold, London, United Kingdom.
Wood, D. J., and Charles, C. O. A. (1972). "Hydraulic network analysis using linear
theory." /. Hydr. Div., ASCE, 98(7), 1157-1170.
Downloaded from ascelibrary.org by ARIZONA,UNIVERSITY OF on 01/05/13. Copyright ASCE. For personal use only; all rights reserved.

APPENDIX II. NOTATION

The following symbols are used in this paper:

Ci = consumer demand at node i;


CHWjj = Hazen-Williams coefficient for pipe linking nodes i and j ;
CLy = constant in pressure-leakage relationship for pipe that connects
nodes i and j ;
— diameter for pipe that connects nodes i and j ;
G, = ground level at node i;
Hi = head at node i;
Hf = minimum head restricted at node i;
K = number of flow-control valves used in network;
= constant in the iterative linearization of head-flow; relationship
for pipe that links nodes i and j ;
= length of pipe that connects nodes / and j ;
U = number of pipe-joining junction nodes;
N = number of selected reference nodes;
NR = average service pressure in pipe that connects nodes i and j ;
P<j = pipeflow between nodes i and j ;
QSy = leakage occurring in pipe that connects nodes ;' andj;
= fictitious pipe flow between nodes i and j corresponding to
valve flow with valve being fully open;
R = set representing all pairs of connected nodes in the network;
Rt = set containing all nodes connected to node i;
Rj = subset of Rt containing all nodes connected to node i through
a nonvalved pipe element;
R* = a subset of R,, containing all nodes connected to node i through
a valved pipe element;
Ru = resistance parameter for pipe that connects nodes ;' and j ;
RSV = composite parameter for leakage in pipe element ij;
Sgn(X) = sign of X;
V(k) = control parameter for Mi flow-control valve in network;
a = dimensionally dependent constant;
A = nominal change;
8G = discrepancy in successive estimates of some value Q; and
= tolerance.

| l! : Superscripts
| ],[;;! Min = minimum;
Max = maximum; and
n = nth iteration.

Subscripts
i,j = node numbering.

472

J. Water Resour. Plann. Manage. 1990.116:455-472.

You might also like