Professional Documents
Culture Documents
C Springer 2006
RAUL P. LEJANO
Department of Planning, Policy, and Design, School of Social Ecology, SE-I, Room 218G,
University of California, Irvine, CA 92697-7075, USA.; E-mail: rplejano@yahoo.com,
lejano@uci.edu
Abstract. Branched distribution systems are found in a large number of situations, including
rural irrigation, reclaimed water distribution, and effluent disposal. Much research has been
developed around optimizing pipeline design assuming a predetermined geographical layout of
the distribution system. There has been less work done, however, on the problem of optimizing
the configuration of the network itself, which is a particularly crucial issue in lower-income
areas where the need for cost-effective irrigation systems is greatest. Generally, engineers
develop the basic layout through experience and sheer intuition. In this paper, a method is
developed for determining an optimal layout for a branched distribution system given only the
spatial distribution of potential customers and their respective demands. The technique utilizes
a mixed integer linear programming (MILP) algorithm to optimize an empirically derived
objective function. The method is tailored for practical application and allows the decision
maker to simultaneously solve for the optimal customer/user base and pipeline layout. The
method is illustrated with a simple example.
Key words: benefit maximization, branched pipeline systems, irrigation, optimization, water
distribution
Introduction
Note that task (1) is an added feature that generalizes the method, but that
the analyst also has the option of simply assuming that all the customer or
demand points need to be hooked up to the system – this is easily built into the
solution algorithm. The method should allow for solution of the more general
problem, however, which may entail solving an optimal set of demand points
to be served.
Method
a continuous route leading from the source to the ultimate delivery point.
The program switches on users, assigns flows to the appropriate links, and
optimizes the layout of the network so as to maximize net benefit.
Given the numerical efficiency of linear optimization routines, the objective
is to attempt to cast the problem using linear functional forms. The first task
is to formulate the objective function. The objective function expresses net
benefit as the total revenue from the water sold minus the total cost of the
system.
Benefit function
In the case of water reuse projects, water supply agencies will often express the
project’s benefit as the savings from reducing potable water demand (Lejano
et al., 1992). Each unit of water reclaimed is assumed to provide a saving
equal to the marginal cost of the next increment of potable water supply. The
result is a benefit function consisting of the product of the unit value of the
water, b, and the volume of water delivered, Qs .
B = bQ s (1)
where K and m are constants, D the characteristic diameter, and L the length
of the link. This expression is generally nonlinear.
In the present formulation, the model’s variables are the flows in each link,
qi j . It is then necessary to express pipeline costs as a function of flow. At the
same time, it is also useful to attempt to simulate the planning and design
process employed by an agency. In this light, two common heuristic design
rules are used:
Rule (1): Pipes are sized to maintain a maximum nominal velocity at their
respective design flows. In the example, we assume that the agency decides
on a design velocity of 1.8 m/s at peak flow. Note that it is common practice
for designers to use a target flow velocity to design pipe systems. This then
allows one to relate pipe diameter and flowrate directly.
Rule (2): To allow for future development and changes in the customer base,
agencies will commonly assume a minimum pipe diameter for installation.
In the example, we assume this to be 0.15 m.
130
The use of these two heuristic rules introduces some sub-optimality into
the solution. However, as will be discussed, the final layout will in most cases
not be sensitive to these assumptions.
The application of Rule (1) means that the pipeline network capital cost
function, C L , can be expressed as a function of flowrate:
CL = K L i j q m/2 for all adjacent i, j (3)
Note that we can even assume a fixed cost component, which is embodied in the
constant, K L 1 , the above expression (Equation (4)). However, this introduces
the constant K L 1 and, so, the the above expression is not, strictly speaking, a
linear but rather an affine function. How can this be cast in a form amenable
to a linear programming approach? The following expression provides one
solution.
CL = (K L1 Ii j + K L2 L i j qi j ) for all adjacent i, j (5)
Figure 2. Pipeline cost as a function of design flow (assumes design velocity of 1.8 m/s).
131
There are also site-specific costs associated with hooking up each customer
or, in the case of water reuse projects, retrofitting users for a dual supply. This
is included in the objective function as the following expression (where C j
refers to the on-site cost for a user at node j).
Cj = Kj Sj for all nodes j (6)
where S j is a switch for each potential customer. These expressions are only
needed for each node j where a potential customer is found. S j is also used as
a binary switch – when a customer is present at node j, then S j = 1, and we
obtain a cost of K j (specific to the kind of user found at node j).
This formulation assumes, a priori, the existence of a pump station. The capital
cost of a pump station is often depicted with a function like the following
(Sanks, 1989):
C p = P K 1 P n = K 1 P n+1 (7)
where P is the total horsepower of the pumping plant, and P is linear with total
flowrate, Qs , and total dynamic head, TDH, for which the plant is designed,
and n is a constant usually less than one. That is, P = K 2 Q s (TDH), and
Equation (7) can be expressed as (8):
K 1 [K 2 Q s (TDH)1+n ]
Cp = (8)
η
the corresponding cost per horsepower from the ideal cost curve. If the as-
sumed unit cost approximates the actual, then the solution is kept. If not, then
it is repeated with a new unit cost. This procedure is recommended by the
relatively flat slope of the unit cost curve (as a function of flow). Equation (8)
then simplifies to
C p = K p Qs (9)
since, for a given TDH and efficiency η, the expression K 1 [K 2 Q s (TDH)n+1 ]/η
in Equation (8) can be expressed as the constant, K p .
In checking for how reasonable the assumed pumping plant TDH is,
we back-calculated the required TDH in the pump station given the sys-
tem layout and pipe sizes returned by the optimization routine. We simply
assume a delivery pressure at each user delivery point and back-calculate
the required head at the pump station, which is equal to the largest head
loss calculated through every route from the users back to the pump station.
That is, the required TDH of the pumping plant is given by the expression
below:
where TDHreq = the required total head, calculated for the pump station;
k = 1, 2, . . . for all points of delivery, k; hk = the total head loss along the
route from k back to the pump station (in m); Pk = the total delivery pressure
requirement at point k (expressed as required pressure head, in m).
This procedure is illustrated in the example below. Note that, for flat terrain,
Pk is simply the minimum service pressure required at each point of delivery.
However, we can vary Pk to account for changes in elevation. For example,
if delivery point 2 is elevated 10 m above the pump station, then P2 is simply
the sum of the difference in elevation and the required delivery pressure (the
latter expressed as required head). If we assume the required delivery pressure
to be 240 kPa (or 24 m, expressed as head), then the total head requirement is
given by P2 = 24 + 10 = 34 m Differences in required pressures at delivery
can also be accommodated this way. For example, if the user at point 2 does
not require the minimum delivery pressure, then a lower required head can be
assumed for this user.
Note that the annual operation and maintenance (O&M) cost can usually be
reasonably expressed as a linear function of horsepower. Expressed in present
value, the O&M component can also be included in the constant, K p .
133
Putting all the various elements together, the problem is expressed as:
maximize bQ s − (K L1 Ii j + K L2 L i j qi j ) − K p Q s
− Kj (S j ) for all i, j (11)
such that qi j = S j Q j for each j and all adjacent i; qsi = Q s for all i
adjacent to s (pump station location); 0 ≤ S j ≤ 1 for each j; S j an integer
variable; 0 ≤ Ii j − qi j /Nlarge < 1 for all adjacent i, j; Ii j an integer variable
all other variables ≥0.
The solution returns the optimal suite of customers to be served, the
pipeline links to be built, and the resulting flows, qi j , in each link. Note
that the first constraint allows more than one link to supply a node (e.g., when
qi j > 0 and q jk < 0 for i = k).
Application
the situation at hand and agency practice. A curve relating actual pipeline cost
to design flowrate can be developed, such as illustrated in Figure 2. As shown
by the solid curve in Figure 2, the second rule makes the curve flatten out as
flow approaches zero. A regression line is drawn through the data points and
is seen to approximate the ideal curve, with the deviation being less than 7%
all throughout the curve. The resulting cost equation for the pipeline capital
cost in dollars is given below.
where Qave is average flowrate through reach i, in m3 /s; Li j the length of reach
connecting nodes i and j, in m.
At this point, we need to approximate the total dynamic head (TDH) at the
pump station. Assuming a TDH of about 61 m, we obtain the following cost
functions (in present values over a 20 year time horizon) of the pump station
O&M and capital costs in dollars, respectively.
users 5, 7, and 9, with a total flowrate of 0.0255 m3 /s. Recall that the objective
is not to minimize cost, but to maximize net benefit. The least-cost user base is,
of course, serving none of the users, but then benefits would also be zero. Note
that customers 3, 6, and 8 are not served since they have very low demand,
i.e., the cost of serving them exceeds the benefit of doing so.
Given the link flowrates, one can then solve for the ideal pipe diameters
using a standard pipeflow equation (assuming a target design velocity) and
obtain a second-best solution by utilizing the nearest commercially available
pipe sizes. One can then solve for the actual head losses and back-calculate
the required TDH at the pump station and compare this to the assumed value.
If the calculated and assumed values are too divergent, then a new TDH can
be assumed and the process repeated.
Assuming a minimum delivery pressure at each point of use of 414 kPa
and using a minimum pipe diameter of 0.15 m, one calculates a total head
requirement at the pump of 54 m, which is close to the assumed value of
61 m. At any rate, additional iterations can be performed if a closer match
is desired. Note that this is a second-best solution – it involves no trade-off
between the costs of pumping and piping.
It should also be noted that, given the finite choice of commercially avail-
able pipe sizes, the resulting solution should be robust. That is, the final
configuration will not change. This procedure is illustrated in the example
below, much, even if we were to vary some assumptions (e.g., the design
velocity, the unit benefit, etc.). For example, lowering the unit benefit by as
much as 20% one still obtains the same solution. The MILP method provides
a systematic method for finding an optimal layout, in contrast to the common
practice of choosing from a small number of alternative layouts developed
“by feel.”
We should point out that the use of heuristics in the methodology results
in a type of “second-best” optimization. For example, assuming a design ve-
locity precludes more optimal pipeline designs that might result if we were
to consider a different design velocity or varying velocities. However, there
are no existing procedures that allow one to ideally optimize both pipe diam-
eters and geographic layout. Moreover, since such a heuristic is commonly
used in the actual design of pipelines, by using this simplifying assumption,
we are able to create an optimization algorithm that is readily applicable and
an improvement over designing these systems “by hand” which, given the
staggering number of potential layouts that even a simple example can yield,
will either be insurmountable task or far short of the optimal yeild. The pro-
cedure provided herein is a reasonably easy procedure that can actually tell
the designer, in a short period of time, which layout to use.
As seen in the example, given the static nature of the problem formulation,
the solution will generally be a branched configuration. One can, of course,
136
set constraints such that every link carries a pipe and redundancy introduced,
but at the price of sub-optimality.
Conclusions
(1) The method seeks out the optimal geographical layout of a water distri-
bution system and optimal user base.
(2) It can be applied to cases where not all potential customers need be served
(such as in the case of water reuse systems). In this case, the model selects
the users which are most efficiently served by maximizing a net benefit
function.
(3) Several heuristic rules normally used for pipeline design are employed in
the model – indeed, the close linkage to real-world design processes is a
strength of this methodology. These rules entail setting a minimum pipe
diameter and limiting velocity.
(4) The objective function can be formulated to reflect site-specific costs,
such as connection or storage costs.
(5) The use of integer constraints and the large number of constraints that even
a small problem can generate adds considerably to the solution time.
References
Alperovits, E. & Shamir, U. 1977. Design of optimal water distribution systems. Water
Resources Research 13(6): 885–900.
Bhave, P.R. & Lam, C.F. 1983. Optimal layout for branching distribution networks. Journal of
Transport Engineering ASCE 109(4): 534–545.
Cornish, G. 1998. Pressurised irrigation technologies for smallhoders in developing countries –
a review. Irrigation and Drainage Systems 12(3): 185–201.
Deb, A.K. 1973. Least cost design of water main systems in series. Journal of the Environmental
Engineering Division ASCE 99(3): 405–409.
Deb, A.K. 1974. Least cost design of branched pipe network system. Journal of Environmental
Engineering Division ASCE 100(4): 821–835.
137
Eiger, G., Shamir, U. & Ben-Tal, A. 1994. Optimal design of water distribution networks. Water
Resources Research 30(9): 2637–2646.
Eusuff, M.M. & Lansey, K.E. 2003. Optimization of water distribution network design using
the shuffled frog leaping algorithm. Journal of Water Resources Planning and Management
ASCE 129(3): 210–225.
Fujiwara, O. & Dey, D. 1988. Method for optimal design of branched networks on flat terrain.
Journal of Environmental Engineering ASCE 114(6): 1464–1475.
Gupta, I. 1969. Linear programming analysis of a water supply system. Transactions of Amer-
ican Institute of Industrial Engineering 1(1): 56–61.
Hamberg, D. 1974. Optimal location of pumping stations in a branching network. M.Sc. thesis,
Israel Institute of Technology, Technion, Haifa.
Karmeli, D., Gadish, Y. & Meyers, S. 1968. Design of optimal water distribution networks.
Journal of Pipeline Division ASCE 94(PL1): 1–10.
Land, A.H. & Doig, A.G. 1960. An automatic method for solving discrete programming prob-
lems. Econometrica 28: 497–520.
Lejano, R., Grant, F., Richardson, T., Smith, B. & Farhang, F. 1992. Assessing the benefits of
water reuse. Water Environment & Technology 8(4): 44–50.
Murphy, L.J., Simpson, A.R. & Dandy, G.C. 1993. Pipe network optimization using an im-
proved genetic algorithm. Research Rep. No. R109, Department of Civil and Environmental
Engineering, University of Adelaide.
Ormsby, L.E. & Wood, D.J. 1986. Hydraulic design algorithms for pipe networks. Journal of
Hydraulic Engineering ASCE 112(12): 1195–1207.
Pereira, L., Calejo, M., Lamaddalena, N., Douieb, A. & Bounoua, R. 2003. Design and perfor-
mance analysis of lower pressure irrigation systems. Irrigation and Drainage Systems 17:
305–324.
Quindry, G.E., Brill, E.D. & Liebman, J.C. 1981. Optimization of looped water distribution
systems. Journal of Environmental Engineering ASCE 107(4): 665–679.
Sanks, R.L. 1989. Pumping Station Design, Butterworth-Heinemann, Boston.
Savic, D.A. & Walters, G.A. 1977. Genetic algorithms for least-cost design of water distribution
networks. Journal of Water Resources Planning and Management ASCE 123(2): 67–77.
Sherali, H.D. & Smith, E.P. 1997. A global optimization approach to a water distribution
network design problem. Journal of Global Optimization 11: 107–132.
Sherali, H.D., Subramanian, S. & Loganathan, G.V. 2001. Effective relaxations and partition-
ing schemes for solving water distribution network design problems to global optimality.
Journal of Global Optimization 19: 1–26.
Tanyimboh, T.T. & Sheahan, C. 2002. A maximum entropy based approach to the layout
optimization of water distribution systems. Civil Engineering and Environmental Systems
19(3): 223–253.
Walski, T.M. 2001. Editorial: The wrong paradigm – why water distribution optimization
doesn’t work. Journal of Water Resources Planning and Management ASCE 127(4):
203–205.