1.
WATER SYSTEM OPTIMIZATION:
CONCEPTS AND METHODS
1.1 SYSTEMS DEFINITIONS
Engineering project design and optimization can be effectively approached using concepts of
systems analysis. A system can be thought of as a set of components or processes that transform
resource inputs into product (goods and services) outputs. The basic concept of a system is rep-
resented in Figure 1.1.
INPUT SYSTEM OUTPUT
(a)
I(x) 1 2
3
5 7
6 8 O(y)
(b)
1 2
x
3
y
4
quantity quantity
quality 7
quality
5
time char. 6 8 time char.
space char. space char.
others others
technological
economical
political
cultural
ecological
other
(c)
Figure 1.1: Representation of a System
In Figure 1.1b, the system is defined by a boundary which separates those components that are
an interrelated part of the system from those outside which are part of the "environment".
Determining the boundary depends on the physical system, the technological and spatial ele-
ments and the assumptions and the purposes for which the analysis is being conducted (see
Figure 1.1c). For example, in a water resources system, the analyst must decide which hydro-
logic basin and water sources, dams, reservoir, and conveyance systems, and service areas and
water uses to include in the “system”.
The inputs define the flow of resource into the system and the outputs and products from the
system. A system often has several subsystems. In the more detailed representation of Figure
1.2, the inputs include controllable or decision variables, which represent design choices that are
open to the engineer. Assigning values to controllable variables establishes an alternative.
The outputs describe the performance of the system or its consequences upon the environment.
They indicate the effects of applying design and planning decisions via the input variables and
are evaluated against system objectives and criteria in order to assess the worth of the respective
alternatives in terms of time, reliability, costs or other appropriate units.
2
Exogenous
Alternatives Variables E Rank
I1 Unc O1 v
ontr ENVIRONMENT Alternatives
I2 olle O2 a
A1 d A p1
... ... l
A2 u A p2
Policy Partially
... Variables
Ij SYSTEM Oi a ...
Controlled
... d ... t
rolle
Cont i
An A pn
In Om o
n
Inputs Outputs
feedback
Figure 1.2: Detailed Representation of a System
1.2 WATER RESOURCES SYSTEMS DESCRIPTIONS
Water resources systems modeling may be treated at various levels of specificity as illustrated by
Figure 1.3. If the design is concerned with local water supply planning, then the system bound-
ary would include the key elements shown by Problem 1 in Figure 1.3. If basin-wide multipur-
pose planning or operation is of concern, the system boundary must be expanded to include the
kinds of elements shown in Problem 2. The engineer might be interested in statewide allocation
of water among basins and uses as illustrated by Problem 3. As a further example, the specific
elements and interconnections of a multipurpose basin are further depicted in the system block
diagram of Figure 1.4. This type of diagram is useful in constructing the mathematical optimi-
zation or simulation models for the system. Table 1.1 summarizes many of the relevant input,
outputs, decision variables, and system constraints and components of water resources systems.
1.3 MATHEMATICAL MODELS OF SYSTEMS:
OVERVIEW AND CONCEPTS
Figure 1.5 is a representation of a “modeling space”, with each face of the cube representing an
important dimension of quantitative models. Depending on whether variable relationships are
probabilistic or deterministic, static or dynamic, and linear or nonlinear (as represented by the
faces of the cube) various analytical techniques (the corners) are required to handle them.
3
RIVER City
SPRING
IRRIGATED
AGRICULTURE
TREATMENT
PLANT
URBAN AREA
SERVICE AREA
WELL FIELD
INDUSTRY
PROBLEM 2: LOCAL WATER
SUPPLY SYSTEM CITY
BIRD REFUGE
Bear
River
Weber
River
Great Salt
Lake
Jordan Uintah
IRRIGATED
River AGRICULTURE
PROBLEM 2: Multipurpose System
Sevier River
West
West
Colorado
Cedar
Beaver
Lower
Southeast
Colorado
Colorado
PROBLEM 3: Statewide Water Resource
Allocation Plan
Figure 1.3: Levels of Specificity in Water Resources Systems Modeling
4
Output -- Economic-
Benefit B2 B1 B3 Political
Functions Interface
C3 C 4-7
City A B 4-7
Power
Plant Water Supply
Flow Augmentation
C8
C2 Watershed Flood Control
Waste Disposal
Res. Downstream
#1 Effects Q(t)
5
C1
Res. Physical-
Operating #2
Policies Irri- Economic
Cost -- gation Interface
(acres) B8
Input (tons/yr)
Functions
Objectives C10 C9
Efficiency;
Redistribution; B10
Environmental Quality; B9
etc.
Policy, Legislation, Decisions
Figure 1.4: Hierarchy of Systems and Systems Functions
Table 1.1: Elements of Water Resources Systems
Inputs to Water Resources Systems:
A. Water sources
1. Surface sources: for example, surface water flow, sedimentation, or salt load, precipitation
2. Underground sources
3. Imported sources: for example, desalting water, imported water
4. Reuse and recycling: for example, treated water from treatment plant, recycling water in irrigation
B. Other natural resources
1. Land
2. Minerals, etc.
C. Economic resources
Outputs of Water Resources Systems:
A. Water allocation to user sectors
1. Municipal
2. Agriculture
3. Industry
4. Hydroelectric power
5. Flood control
6. Navigation
7. Recreation
8. Fish and wildlife habitats
B. Quantity and quality of the water resource system
1. Flow of the stream
2. Quality of stream
System Decision Variables:
A. Management and planning
1. Operating strategies
2. Land use zoning
3. Regional coordination and allocation policy
4. Number and location of treatment plants
5. Sequence of treatments and treatment level achieved
B. Investment policy
1. Budget allocation to various subsystems
2. Timing of investment: for example, stages of development, interest rate
3. Taxing and subsidy strategies
Constraints on Systems Performance:
1. Economic constraints: for example, budget, B/C ratio
2. Political constraints: for example, tradeoff between regions
3. Law: for example, water rights
4. Physical and technology constraints: for example, probability of water availability
5. Standards: system output may have to meet certain standards: for example, effluent standards from wastewater treatment
plants
System Physical and Engineering Components:
A. Planning and management system components
1. Dam and control structures
2. Levees and other protecting structures
3. Distribution or collection systems comprised of (a) canals, (b) pipes, (c) pumping stations and other control struc-
tures
4. Treatment plants
B. Descriptive system components
1. Physical properties of stream: for example, roughness, slope
2. Biochemical properties of stream: for example, rate of aeration, rate of self-regeneration
3. Chemical properties of stream: for example, hardness, pH
6
Deterministic/Probabilistic
Probabilistic
Static
Dynamic Number
of
Variables
Deterministic
Static/Dynamic
Figure 1.5: Modeling Space (Cube)
Broadly speaking the purpose of modeling may be either predictive or prescriptive. Predictive
models of systems are constructed to clarify the internal structure of a system and predict its
behavior or response to input variables. On the other hand, prescriptive models strive not only to
reproduce the behavior of the system itself, but also to evaluate the consequences of design alter-
natives according to predetermined measures of performance.
Identifying the model structure for predictive or prescriptive models must be based either on
formal theory or some very strong plausibility arguments. Systems models cannot be devised by
simply using statistical manipulations of data and information to determine variable interactions.
Moreover, all the information relevant to the system may not be quantifiable as numerical data.
Hence, systems modeling techniques may be quantitative and nonquantitative or both.
Table 1.2 provides a general classification of modeling methods and techniques useful in systems
analysis. The entries in the Table are classified under the heading of predictive or prescriptive
models according to the theoretical basis for model construction. Table 1.3 relates the models to
the modeling cubic dimensions shown in Figure 1.5.
This overview of course cannot provide detailed descriptions of the various modeling
approaches. Whole textbooks are devoted to these subjects. Instead, this discussion simply tries
to provide a basic classification of the techniques in order to understand where optimization fit
among the various methods.
7
Table 1.2: An Overview of Approaches to Systems Modeling
Application or Use
Problem Type Predictive Prescriptive
Quantitative
Deterministic Systems Transformations Optimization Procedures
algebraic equations, Classical Optimization Theory;
differential equations, state differential calculus, lagrangians,
variable formulations, input- optimal control theory
output analysis
Networks Networks
graph theory CPM and PERT
Stochastic Stochastic Processes Decision Analysis
inventory theory, queuing Statistical (Bayesian) decision
theory, Markov processes theory, game theory
Statistical Models
regression analysis,
component and factor
analysis, stepwise multiple
regression, discriminant
analysis, econometric
analysis
Simulation Deterministic and stochastic Monte Carlo methods, search
model components and models techniques for dominant solutions
Non-quantitative Verbal Models Verbal Models
scenarios, survey research Delphi inquiries
People Models People Models
role playing operational gaming
8
Table 1.3: Types of Models in “Modeling Space”
State of
Time Random Domain Functional
Variance Variance Variance Variance
Single Variate
Deterministic
Probabilistic
Multivariate
Nonlinear
Dynamic
Linear
Static
Type of Model or Solution
Approach
Systems Transformation
Linear, nonlinear systems
1st order diff. equations X X X
state variables X X X
Input-output analysis X X X
Optimization
Classical Optimization
Differential Calculus X X X X
Lagrange Multipliers X X X X
Optimal Control Theory
Mathematical Programming
linear programming X X X X
nonlinear programming X X X X
integer programming X X X
dynamic programming X X X X
stochastic programming X X X X
genetic programming X X X X X
Stochastic
Inventory X
Queuing X X
Markov X X
Multivariate X X
Networks
Graph Theory
CPM and PERT
Simulation
9
1.4 A GENERAL MODEL OF SYSTEM OPTIMIZATION
1.4.1 System Design and Optimization
Engineering design problems can be mathematically described by three functions associated with
each of the design factors: the physical processes (design or production function), the resource
function's costs, and the output's or product's values (benefit functions) (see Figure 1.6). The
definition of each function is derived from different sources.
System Design Variables
(X1 , X 2 , ... , Xn )
Design
Physical (Production) Output: Goods
Resources Function and Services
(b1 , b 2 , ... , bm ) gm(x) <=> bm (z 1 , z 2, ... , z k )
g*(x) = Z
Resource Cost Evaluation Output Benefit
Functions Functions
C = c(x) V=B-C B = h(z) = h'(x)
Input Optimization Output
Cost Benefit
Utility Max V Utility
Figure 1.6: Model of Systems Design and Optimization
The design function, based on the physical nature of the system without regard to value,
describes the maximum product that can be obtained from the input of any given set of
resources. The resource cost function is usually defined by the economic market value of those
resources. The product valuation function may be determined either by a market or, in the case
of social benefits such as conservation--which often do not have a market--by a political process.
The first design step is to model the production process, the physical process for transforming
resources into products.
10
From a general mathematical description of the systems optimization problem can be stated as
follows:
The objective is to maximize the net value of output, or:
V= Benefit - Cost ...[1.1]
This is also expressed as Profit = Revenue - Cost, or
P=R-C ...[1.2]
The benefit (revenue) and cost functions are functions of a set of control design variables, x1, x2,
..., xn, and can be represented as nonlinear and linear mathematical functions. A general mathe-
matical description for maximizing net value is:
Objective function: Maximize Z = f(x) ...[1.3]
Subject to Design Function and Constraint set:
g(x) ≤,=,≥ b ...[1.4]
x ≥ 0 (nonnegativity condition) ...[1.5]
where: Z is a measure of effectiveness (Z = B - C) = f(x)
x is a vector of n control design variables, x1, x2, ..., xn
f(x) = f(x1, x2, ..., xn) is a function of control variables
Ï g1 (x) ¸
Ô g (x) Ô
g(x) = Ì 2 ˝ is a vector of m design and/or constraint equations of x
Ô ... Ô
Óg m (x)˛
Ï b1 ¸
Ôb Ô
b = Ì 2 ˝ is a resource constraint vector
Ô. . . Ô
Ób m ˛
1.4.2 The Design Function
The design and constraint functions represent the transformation of resource inputs to project
outputs or products. They are the core model, and represent the physical system design alterna-
tives.
11
Specifically, the system design (production) function is the mathematical description of the out-
put that can be obtained from any given set of resources. A general design function may be
expressed as:
gi(x1, x2, ..., xn) ≤,=,≥ bi ...[1.6]
Money and value are not part of the expression; gi(x) is in units of production output, and the xj
represent physical rather than monetary resources. For example, g(x) could be the maximum
product for given quantities of land (x1) and water (x2). Other examples of design functions that
can be represented as an algebraic equation are the power output from a hydro-generating station
as a function of rainfall intensity and duration over a drainage basin, or the amount of BOD
removal in a wastewater treatment system as a function of influent BOD concentration and
detention time in treatment. The design (production) function is a relationship between physical
quantities alone.
Thus g(x) may be a series of physical relationships which are based on theoretical knowledge
(hydraulics, mass balance, etc.) or may be statistically based on probability distributions, regres-
sion function, etc. These are no different than the functions engineers have always used in tradi-
tional problem solving. There is a difference in how they are related to the formal way in which
the equations or inequalities are written so that a solution algorithm (usually a separate piece of
computer software) can be used to generate all possible outputs of interest including the "best" or
optimal solution. This contrasts with traditional approaches where usually only one or a few
solutions are produced. A great advantage of modeling is the ease with which "what if" question
can be asked to explore alternate assumptions on resource limits or production efficiency with no
further mental effort once the system is described mathematically.
Each point on the design function represents the maximum output that can be obtained for any
given set of resources. The function, therefore, dominates any lesser amount of product that
would be obtained from a wasteful or technologically poor use of these resources. The produc-
tion function is thus, by definition, the locus of all technically efficient combinations of
resources.
The significance of the design function can be simply illustrated in Figure 1.7. The amount of a
crop that might be produced on a given parcel of land, x, is g(x) tons. This point on the design
production function dominates other feasible outputs, such as A and B, that might be achieved
with the same land, x, if, for example, the land were not all cultivated. Conversely, it is infeasi-
ble to produce any more than g(x) tons with a quantity of land equal to x. The point (x*,g*(x)) is
at the edge of feasible and infeasible amounts of production for x. The production function can
thus be conveniently visualized as the boundary between the feasible and infeasible regions in
the input-output space.
A design function can relate any number of resource inputs to one or several product outputs.
Graphically, it is difficult to visualize whenever there are more than two inputs--since our usual
perception is limited to three dimensions--but its functional significance remains the same.
12
g(x) Design Function
[X*,g*(x)]
g*(x)
Region of
Output
Feasible
A Combinations
Resource X
Figure 1.7: Example of Design Function
The concept of the design function is general: it does not necessarily have any particular form
and cannot always be written as an algebraic expression or system of equations. Sometimes the
maximum product for any set of resources may be simply tabulated. Whether the design func-
tion is determined by formula, by detailed design, or by complex simulation methods, its mean-
ing is the same. It represents the limit on what can be achieved with available technology and a
given set of resources.
1.4.3 Evaluation Models and Design Optimization
In a design analysis, evaluation of alternatives may have to be conducted at several different
levels of decision. This is depicted in the generalized model of design evaluation in Figure 8.
The lowest level of evaluation is in terms of system performance.
Performance or effectiveness analysis simply looks at the capability of the physical system to
meet the specified needs or requirements; for example, the ability of a structure to carry the
design loads, or a water system to maintain a certain flow and pressure, or a production process
to turn out a given quantity and quality of project.
Significance analysis relates the quantitative measure of an output to its qualitative value. To use
the economist’s language, it is the process of describing the utility function for the particular
outputs. These processes describes the degree to which individuals or society as a whole places
positive value or negative value on, or are indifferent to, project outputs.
13
A Generalized Model of Decision Evaluation
DESIGN
(methods and
materials)
RESOURCES ALTERNATIVES RESULTS
COSTS INPUTS SYSTEM OUTPUTS BENEFITS
C1 I1 S1 O1 B1
C2 I2 S2 O2 B2
... ... ... ... ...
market $ capital ... products market $
extra-market $ material Sn performance non-market $
intangibles labor ... intangibles
... ... On ...
Cn In Bn
(effectiveness) (significance)
14
Minimize variable fixed
Cost
Maximize
fixed variable
Benefits
Maximize
cost variable variable benefits
Profit
(ECONOMIC EFFICIENCY $ MODELS)
(SOCIAL EFFICIENCY-MULTICRITERION MODELS)
Figure 1.8: System Performance and Optimization
Cost Effectiveness: In cases where project costs are monetary but benefits are measured in some
other unit, then cost-effectiveness analysis can be used for single criterion evaluation. For
example, assume that the objective is to reduce sedimentation of a reservoir resulting from
present watershed development. A systems model is constructed and three possible alternatives
are tested to determine the effectiveness in reduction of sediment level as a function of the
designs and associated costs. Then either a level of effectiveness must be specified and then the
cost minimized for that level, or the limit on cost specified and the effectiveness maximized. For
example in Figure 1.9 if the cost cannot exceed C2 and reduction of sediment to level E1 is all
that is required for the reservoir uses then Alternative 2 operated at cost C1 is the best choice.
On the other hand, if the goal were to reach some minimum level of effectiveness E2 regardless
of the cost then Alternative 3 at cost C2 would be the choice. However, some flexibility should
be allowed in the analysis for if C3 is a reasonable cost to pay, then by only a slight increase in
costs to C3 large gains in effectiveness can be achieved with Alternative 1. In fact, the approach
of setting costs at the place where slope of the cost-effectiveness curve flattens is a judicious one
since little is gained by further expenditures past that point.
E3 1
Effectiveness, E, in Reduction
2
E2
3
of Pollution Level
E1
C1 C2 C3 Cost, C
Figure 1.9: Cost-Effectiveness Curves of Reservoir
Sediment Control Measures
Economic Benefit Analysis: If consequences of alternatives can be valued on an economic
(monetary) scale, then the preferred alternative is the one which produces the largest net
economic benefit (revenues minus costs). With both benefits and costs measured in monetary
units, evaluations can be performed using the tools of engineering economic analysis (Grant et
al., 1987). Since investment in project facilities and the returns from project operation occur
over long periods of time, to correctly evaluate both present and future benefits and costs they
must be compared at the same point in time. One method to accomplish this is by "discounting"
15
the costs and benefits using an appropriate interest rate applied over the useful project life to
obtain the net present value of project outputs. This general procedure stated in equation form is
as follows:
If the question involves the present value of a single future quantity (C), the equation is:
C
PV = ...[1.7]
(1 + r)t
where r is the interest rate per time period, and t is the number of time periods.
If the present value of a uniform series of future amounts is needed, the equation is:
(1 + r)t - 1
PV = C ...[1.8]
r (1 + r)t
Also, if rather than present value, one is doing the reverse--that is, a capital recovery analysis, the
factor needed is the reciprocal of the factor in the second equation. Therefore, the capital recov-
ery factor is:
r (1 + r)t
RC = ...[1.9]
(1 + r)t - 1
These relationships are all that are needed for the vast majority of the engineering economics
analyses that one encounters, including those in this course.
Besides comparison of present worth of benefits minus present worth of costs, engineering eco-
nomic analysis may also be formulated as a benefit cost ratio, equivalent uniform annual costs
(benefits), and rate of return including incremental rates of return. These methods are presented
and discussed by various writers (Grant et al., 1987; Winfrey, 1969; DeGarmo et al., 1984;
Howes, 1971; and others). All of the methods when correctly applied will give equivalent
answers. The principal difficulties in benefit cost studies are the selection of an appropriate time
period and discount rate, since the results of the analysis are often sensitive to these factors.
Social Efficiency (Multiple Criteria Analysis). Project alternatives that have several noncom-
mensurate outputs involving both market and non-market values require multiple criteria com-
parisons in evaluation. Insight into this very important problem of trade-offs between the pluses
and minuses of noncommensurate project consequences is provided the theory of welfare eco-
nomics. To illustrate the problem, consider the joint optimization of two objectives correspond-
ing to the outputs, O1 = f(x) and O2 = g(x), where x is a set of input levels associated with a
range of alternatives. O1 and O2 are plotted in Figure 1.10 as a function of input levels for the
alternatives. These constitute a pair of objective functions to be “maximized” simultaneously:
16
È O1 = f1(x) ˘
Maximize Í ˙ ...[1.10]
ÍÎO = f (x)˙˚
2 2
subject to any applicable design functions and constraints:
gi(x) ≤ 0 ...[1.11]
g(x)
Output Magnitudes, O1 , O2
f(x)
a b X
Input Levels, X, of Alternatives
Figure 1.10: Joint Optimization of Multiple Outputs
Examination of Figure 1.10 indicates that some alternatives (input levels) can be immediately
eliminated from further consideration because they are dominated by better combinations. This
includes the area to the left of “a” and to the right of “b”, since in these regions the functions O1
and O2 are both decreasing. This reduces the range of alternatives to between “a” and “b”, called
the “efficient region”. However, to select the joint optimum point within the efficient region
depends on the tradeoffs or relative weights for the outputs O1 and O2. The tradeoffs or weight-
ings of the objectives cannot be deduced analytically, but are value judgments that must be
supplied by the decision maker.
17
1.4.4 Summary
An engineering systems analysis is the basis for choosing among alternative project designs.
The analysis may be performed at one or several levels depending on the nature of the project
and the decision criteria. For most all engineered projects, a criterion that must be satisfied is
that of maximizing the economic or monetary return. For this reason, optimization tools and
methods usually attempt to maximize the monetary benefits of projects. In many instances,
analysis of projects must be broadened to non-monetary environmental and social consequences
to fully evaluate the worth or projects.
1.5 CLASSICAL OPTIMIZATION--A REVIEW
Before proceeding to modern operations research methods of optimization, it is useful to briefly
review the concept of optimization (finding maximums or minimums) of functions from the clas-
sical approach of calculus. Engineers are usually familiar with calculus approaches to finding
best facility capacities by taking a first derivative, setting it equal to zero and solving for station-
ary points. For example, consider the problem of a fixed flowrate to be pumped through a pipe.
The pipe cost increases with diameter while the energy cost decreases. The optimal diameter is
where df/dx = 0.
This is however an over simplified problem because usually several constraints exist (standard
pipe diameters and pump sizes) and flow may very seasonally with demands so total cost is a
function of several flow rates which cannot be calculated until diameter is known. We, there-
fore, need approaches to handling many constraints simultaneously. If the constraints are
equalities, we can include them in the objective function using the Lagrangian concept. Consider
the constrained problem:
Minimize cost = f(x) ...[1.12]
subject to g(x) = b ...[1.13]
Define the Lagrangian function as:
L = f(x) - l [g(x) - b] ...[1.14]
where l is an artificial variable to be discussed later. Since we now have only a single function
(L) we can take its derivative and set each partial derivative equal to zero as before for identify-
ing the optimal value (the minimum possible value of f(x), which does not violate the constraint)
as follows:
∂L
=0 ...[1.15]
∂x
∂L
=0 ...[1.16]
∂l
18
A simultaneous solution of the above equations yields x = X*, where f(X*) will be either a
minimum or maximum of f(x). If there are more than one x variable and/or multiple constraints,
we have:
Max or Min f = f(x1, x2, ..., xn) ...[1.17]
subject to: gi(x1, x2, ..., xn) = bi (i = 1, 2, ..., m) ...[1.18]
For example, consider:
Maximize f = 2 x1 + x1 x2 + 5 x2 ...[1.19]
subject to the design constraint:
3 x1 + x2 = 10 ...[1.20]
The Lagrangian for this problem can be formed as:
L = 2 x1 + x1 x2 + 5 x2 - l (3 x1 + x2 - 10) ...[1.21]
Note, in forming the Lagrangian, all we have done is subtract a penalty from f(x) equal to l for
each unit by which g(x) ≠ b. Therefore, l can be defined as the change in f(x) for each unit of
change in the right hand side of g(x). This is what economists call an imputed value or shadow
price at the optimal x:
∂L
= 0 = 2 + x2 - 3 l ...[1.22]
∂x1
∂L
= 0 = x1 + 5 - l ...[1.23]
∂x 2
∂L
= 0 = 3 x1 + x2 - 10 (the original constraint) ...[1.24]
∂l
A simultaneous solution of this system of equations is:
x1 = -0.5
x2 = 11.5
l = 4.5
19
f(x1,x2) = 50.75
Note that we could easily solve this problem because we had three linear equations, but even
three nonlinear equations could have been very difficult to solve and most real-world water
problems have many (perhaps hundreds) of variables. We clearly need a procedure for handling
much larger problems. Also, the constraint was an equality, but most water problem constraints
are inequalities (upper or lower bounds on resources or capacities on pipes, etc.). The above
approach can be generalized to handle inequalities by using Kuhn-Tucker conditions (which are
discussed in more detail in Section 11), but the solution then becomes even more difficult. The
classical calculus approach to general optimization problems can be summarized as:
Problem Solution Technique
1. Max or Min f(x) Set:
df
=0
dx i
for each xi, and solve the resulting system
of equations.
2. Max or Min f(x) Form the Lagrangian:
s.t.: g(x) = b (equality constraints) L = f(x) - Â [li (gi(x) - bi)]
Set the partial derivatives of L with respect
to each xi and lj equal to zero and solve the
resulting simultaneous system.
3. Max or Min f(x) Form the Lagrangian:
s.t.: g(x) ≤,=,≥ b (inequalities) L = f(x) - Â [li (gi(x) - bi)]
and employ the Kuhn-Tucker conditions,
including:
L’(x,l) ≥ 0
li Si = 0
where Si are slack or surplus variables.
In general we are not assured of a global optimum because of the nature of f(x), as illustrated in
Figure 1.11.
20
Z = f(x)
Global
max
Local
max
Valley
Ridge
Inflection
Local point
min
Global
min
Figure 1.11: Possible Behaviors of a Nonlinear Function
1.6 MODEL DEVELOPMENT APPROACH
The proper way to approach development of the mathematical model of any system is:
1. Identify the essential parameters that must be modeled in order to capture quantitatively
the decision variables of interest. For example, a thermal electric generating plant is a
tremendously complex system with several sub-systems. If, however, our objectives is
to model only cooling water requirements, these can be calculated by simple functions
relating water supply to: capacity in KW, type of cooling tower, salinity at which efflu-
ent is removed.
2. Describe the system as accurately as possible as a series of equations and/or inequalities
which capture the interaction of these essential parameters without regard to type of
solution techniques which is to be used. If at this stage one sets out to build a linear pro-
gramming (LP) model, then the solution tool is distorting (increasing errors in) the
model. The solution tool rather than the real system is dictating the model structure.
One should define the system as well as possible in mathematical terms using symbols
for all parameters, which may vary, rather than assuming any constant values.
3. Now consider what solution approach may be best. The objective may be to find the
“best” (the optimal) solution or more commonly it may be to analyze a range of “good”
21
solutions which are within the domain of the resources available. In either case a simu-
lation or an optimization model may be appropriate. The essential differences is that an
optimization model has one or more formal objective functions and all of the model’s
constraints must be solved simultaneously (unless dynamic programming or some or
decomposition technique is used to solve a series of smaller systems of equation
sequentially). All parameters over which some control is possible are considered to be
variables and we seek the solution that maximizes or minimizes our objective.
4. Simulation models in contrast usually consist of a series of equations that are solved
one at a time until all potential variables are fixed. This much simpler solution is
obtained by fixing enough variables that equations can be solved, thereby giving a par-
ticular (but not optimal) solution. One can then change the fixed parameters, calculate a
new solution and interactively develop the surface of the solution space (an objective
function is not formally included in the model).
Simulation models are more flexible than optimization (due for example, to the capability of
using “if-then-else” statements and “do” loops) and can produce more detailed solutions. How-
ever, for problems with many variables, manually varying multidimensional parameters may be
very slow, very costly in terms of computations, and will never assure that one has really found
the region of the solution space that is optimal. It is very common to use optimization models
for preliminary design (identifying the domains of good solutions) and simulation models for
more detailed design and operation analysis.
Non-linearities cause no difficulty in simulation models but may be very troublesome in most
optimization models. If, however, an optimization model is considered appropriate and the sys-
tem is non-liner there are several techniques available to either use piecewise linear segments of
non-linear functions or to do truly non-linear optimization. The most common water resource
planning models are those with linear constraints and non-linear objectives (due for example to
economies of scale in cost functions). Another very common water resource model characteris-
tic is discrete step functions due to standard pipe or treatment component sizes. Here we can use
0 or 1 variables for non-build or build decisions. This requires integer programming.
22
1.8 PROBLEMS
1. The storage volume for a proposed reservoir is given by the equation:
S = 10 A h
where A is the area of the reservoir site in acres, and h is the height of the dam in feet.
The cost of the land is $1000 per acre, and the cost of the dam is $5000 per foot of height.
The total budget for the project is $200,000.
a. Determine the optimal size for the reservoir area and dam height within the given
project budget.
b. If the project budget were increased by $1000, what corresponding increase in storage
could be obtained?
2. Flood control for a region can be provided by some combination of reservoir storage and
levees. It has been determined that:
a. Cost of storage is inversely proportional to the flow released (or passed) from the res-
ervoir.
b. Cost of levees is directly proportional to the flow released.
Your objective as project designer is to find the minimum cost project.
a. Formulate equations and find an analytical solution.
b. Show the problem graphically and illustrate the solution.
3. A water company wants to pump water from its well field to the city, which is three miles
away. The pipe in place will cost $10D per foot, where D is the diameter of the pipe in
feet. The annual pumping cost is $150h per year, where h is the head loss in the pipe. The
head loss is given by:
0.02 L V 2
h=
D 2g
where L is the pipe length in feet and V is the average flow velocity in feet/second. The
pipe will last 15 years and the interest rate is 10 percent. The company wants to pump
12,000 gpm. What size pipe should they use to minimize cost?
23
4. A sedimentation tank is circular in plan with vertical sides above the ground and a conical
hopper below the ground (refer to the figure, below). The slope of the conical part is 3
vertical to 4 horizontal. Determine the proportions to hold a volume of 4070 cubic meters
for a minimum cost (which, in this case, will be a tank with a minimum total area of bottom
and sides).
H
G.L.
3 D
5. Consider a trapezoidal canal whose cross-sectional area, A, is to be 100 ft2 (see the figure,
below). It is desired to maximize discharge for the given cross-sectional area. The mean
flow velocity increases with hydraulic radius, A/P, where P is the wetted perimeter. The
discharge is a maximum when the mean velocity is maximum. Therefore, the design opti-
mization reduces to minimizing the wetted perimeter, P.
For the trapezoid,
P = b + 2d csc j
and
A = db + d2 cot j
Find the canal design parameters d, b, and j.
Canal of trapezoidal section.
A d
f f
24
