CIVL 2170: Infrastructure Systems

Engineering and Management

Lecture 2
Optimization Tools for Systems Development:
Introduction
Satish Ukkusuri
 Examples of optimization at various
phases of CE systems development
 Decisions include:
 What resources (manpower, money, materials, machinery use) are needed? what quantities?
when?
 What alternative should we adopt to enhance the system? Which year?

Terminal System Phase Needs Assessment,

Goals Identification, etc.
Selection of Optimal Termination Policy

System Planning
System Preservation
Selection of Optimal System
Selection of Optimal Plan, Location, or Policy
Preservation or Maintenance
Strategy for the System

System Operations
System Design
Selection of Optimal Operational
Procedure for the System Selection of Optimal
System Design or Material

System Construction
Selection of Optimal Construction Delivery Process 2
 Examples of optimization at various
phases of CE systems development
• Specific examples
– Which of these alternative highway routes should we select?
– What is the best combination of materials to use to line a sedimentation pond?
– What dimensions of structural elements provides the best support at least
cost?
• Common characteristics
– Goal is to maximize utility or minimize disutility
– Decisions are affected by constraints
• financial (budgets),
• physical (right-of-way limitations),
• institutional (legal and administrative requirements),
• political considerations.

– Decision variables: the factors that civil engineers manipulate in order to

maximize utility or minimize disutility
• Synonyms: design variables, input variables, control variables.

– The constraints are expressed in terms of the decision variables.

3
 How May we Categorize
Optimization Problems?
• Effort of the search
– Optimization means searching for values of control variable to
yield some optimum performance.
• Extensive search OR Short search

• Level of knowledge about the system outcome in response

to a certain stimulus.
– If outcome is known: optimization under certainty
– If outcome is unknown: optimization under uncertainty.
• Optimization with risk (where p of each possible outcome is known)
• Opt. with true uncertainty (where the p of each possible outcome is not
known).

• Attribute of the system under consideration

– Physical component or operational policy

4
 How May we Categorize
Optimization Problems?
• Number of criteria being considered
– The problem may be a single-criteria optimization problem, or a
multi-criteria optimization problem

• Discrete/continuous nature of the decision variable

– Discrete optimization vs. problem continuous optimization
• Discrete: count variable (eg, nr. of production units required to achieve
some objective)
• Binary variables (eg, whether to select a specific type of resource for a
given task)

• Network level or facility level

– The optimization or decision making problem may pertain to a
network of facilities or to a specific facility in the network.

5
 Some standard definitions
• Control variables / Decision variables (x1, x2, …., xn)
– A control variable is a term used to designate any parameter
that may vary in the design, planning, or management
process.

• Objective function
– f(x1, x2, …., xn) = f(x) is a single-valued function of the set of
decision variables (x)

• Constraint conditions
– These are the mathematical notation of the limitations places
upon the problem (e.g., financial, physical, institutional)

6
 Some standard definitions
• Feasible solution space
– Any combination of decision variables that satisfies the
set of constraint conditions is called feasible solution
space

• Optimum solution
– It is a feasible solution that satisfies the goal of the
objective function as well.
– The optimal solution is a statement of how resources or
inputs should be used to achieve the goal in the most
efficient and effective manner

7
 Setting up the optimization problem
• A city is growing in population. The public works department projects the
need for an increased water supply. There are three options:
– Source 1 Current supply, source
• Local stream of good quality
– Source 2 Aquifer and source
• Its hardness level is too high unless it is blended with a lower hardness source.
• The total pounds of hardness per million gallons is limited to 1,200 in the final blended
supply.
– Source 3 The distant stream
• Sufficient quality
• A pipeline would have to be built to convey it. The cost to pump water to the water
treatment plan is high.

Source 1 Source 2 Source 3

Cost 500 1,000 2,000
Supply Limit 25 120 100
Hardness 200 2,300 70

• A total of 150 additional million gallons per day is needed by the end of ten
years.
• What is the least cost strategy ensuring that water supply remains in
sufficient quality?
8
 Linear Programming with
Continuous Variables
• What is a feasible region? How to sketch one.

• How to find corners (vertices) of a feasible region

• Objective functions and Constraints

• Graphical solution of LP optimization problems

• Some standard terminology

9
 Linear Programming with
Continuous Variables
Example 1: Sketch the following region:
y -2>0
• Solution
– First, make y the subject
– Write the equation of the critical boundary
– Sketch the critical boundary
– Indicate the region of interest

y
y>2
5 Equivalent to
4 y -2> 0
3
Critical Boundary y=2
2
1
x
-1

-2 10
 Linear Programming with
Continuous Variables
Example 2: Sketch the following region:
x -5<0

x
1 2 3 4 5

x <5 x=5
11
(Critical Boundary)
 Linear Programming with
Continuous Variables
• How to Sketch a Region whose Critical Boundary is a bi-variate
(x,y or x1,x2) Function
– First, make y the subject of the inequality
– Write the equation of the critical boundary
– Sketch the critical boundary (often a sloping line)
– Indicate the region of interest
• Note that …
– y < f(x) means the region below the sloping line (e.g. y < 9x+5)
– y > f(x) means the region above the sloping line (e.g. y > 6x-1)
• Examples:
Region: y - x ≤ 0 Region: y > 2x + 1
y y
y=x y = 2x+1
(Critical Boundary) (Critical Boundary)

1
x
x
-1/2

12
 Linear Programming with
Continuous Variables
• How to sketch a region bounded by two or
more critical boundaries
– First make y the subject of each inequality
– Write the equation of the critical boundary
– Sketch the critical boundaries for each inequality
– Indicate the overlapping region of interest
• See next example…

13
 Linear Programming with
Continuous Variables
– Example y
Sketch the region bounded (or constrained)
by the following functions
y>0
x>0 (0, 5)
y < -3x + 5
1. y > 0
Thus, the critical boundary is:
y=0 (5/3, 0)
y=0
(0, 0) (Critical Boundary)
2. x > 0
Thus, the critical boundary is:
x=0

3. y < -3x + 5
Thus, the critical boundary is: x=0 y= -3x + 5
y = -3x + 5 (Critical Boundary) (Critical Boundary)
When x = 0, y = 5 → CB passes thru (0, 5)
When y = 0, x = 5/3 → CB passes thru (5/3, 0)

4. The resulting region where all points satisfy all the three constraining
functions is the FEASIBLE region

14
 Linear Programming with
Continuous Variables
• Other examples
Region Region
y>0 y>3
y < - 0.2x + 5 y < -0.2x + 6
y > -0.5x + 5 y<x+1
y
y = -0.2x + 6 y= x + 1
(Critical Boundary) (Critical Boundary)
Feasible Region 6

y=3
(Critical Boundary)
y=0 -1 3
(Critical Boundary) x

Feasible Region
y= -0.5x + 5 y = -0.2x + 5
(Critical Boundary) (Critical Boundary)
15
 Linear Programming with
Continuous Variables
• What are the vertices of a feasible region?
– Simply refers to the corner points
• How do we determine the vertices of a feasible region?
– Plot the boundary conditions carefully on a graph sheet and read off the values at the
corners, OR
– Solve the equations simultaneously
– Example
y>0
x>0
y < 0.33x + 1
y > 2x - 5 y (Critical Boundary)
y= 0.33x + 1 Vertex X Y
C (3.6, 2.2) (Critical Boundary)
A 0.0 0.0
B (0, 1) Feasible
Region B 0.0 1.0
C 3.6 2.2
x
A (0, 0) (Critical Boundary) D 2.5 0.0
D (2.5, 0)

y = 2x - 5
(Critical Boundary)
16
 Linear Programming with
Continuous Variables
• The Objective Function
– For example:
• W=x+y
• W = 2x + 3y
• W = x2 + y
• W = x0.5 + 3y2
• W = (x + y)2
• …
– So we typically seek to optimize (maximize or minimize) the value of W.
– In other words, W is the objective function.

• Variables Notation
– x and y are referred to as Control variables or Decision variables
• In some books, (x1, x2) is used instead of (x,y)
• (x1, x2, x3) is used instead of (x, y, z)
• (x1, x2, x3 , x4) is used instead of (x, y, z, v)
• etc. x2
x2

x1 x1 x3
17
 Linear Programming with
Continuous Variables
• An optimization problem with n decision variables  n-dimensional

x2
x2

x1
Cannot be visualized
x1 x3
2-dimensional 3-dimensional n-dimensional

2 Decision Variables 3 Decision Variables n Decision Variables

Intersecting lines yield Intersecting planes yield Intersecting objects yield
vertices (problem solutions) vertices (problem solutions) vertices (problem solutions)

18
 Linear Programming with
Continuous Variables
Example of 2-dimensional problem
• Given that Z = 8x + 5y
• Find the maximum value of Z subject to the following:
y>0
x>0
y < 0.33x + 1
y < 2x - 5
Vertex X Y Max Z = 8x + 5y
Vertex X Y Max Z = 8x + 5y
y A 0.0 0.0 = 8(0) + 5(0) = 0
(Critical Boundary) A 0.0 0.0 = 8(0) + 5(0) = 0
y= 0.33x + 1 B 0.0 1.0 = 8(0) + 5(1) = 5
B 0.0 1.0 = 8(0) + 5(1) = 5
C (3.6, 2.2) (Critical Boundary)
C 3.6 2.2 = 8(3.6) + 5(2.2)
B (0, 1) Feasible C 3.6 2.2 = 8(2.5) + 5(0) = 20
= 39.8
Region
D 2.5 0.0 = 8(2.5) + 5(0) = 20

x
A (0, 0) (Critical Boundary)
Therefore, the maximum value of the
D (2.5, 0) function is Z = 39.8
y = 2x - 5
This happens at the point
(Critical Boundary) (x, y) = (3.6, 2.2)
19
 Linear Programming with
Continuous Variables
• Common Methods for Solving Linear Programming Problems
– Method 1 Graphical Methods y

• e.g. Previous example

x

– Method 2 Simultaneous Equations

• Employ the technique of substitution or elimination to solve the constraints
simultaneously
• Then plug in the vertex values (i.e., value of the decision variables) into the
objective function
• Determine which vertex (set of decision variables) yields the optimal value of the
objective function.
• Method is easy when there are few decision variables and even fewer constraints.

– Method 3 Using Linear Algebra (Matrices)

• In this method, the vertices of the feasible region are determined as follows:
– Set up the objective function and constraints as a set of linear algebra equations
– Develop the corresponding matrices.
– Solve the set of linear equations using vector algebra. This yields the optimal value of the
objective function.
– Simplex method can be employed to help in rapid solution of the matrix
– Method is easy when there are few decision variables and even fewer constraints.
20
 Linear Programming with
Continuous Variables
• Method 4 Using MS Solver
– Steps
1. Construct a Constraints matrix
2. Put in initial (or seed ) values for your control
(or decision) variables
3. Call up Solver to determine which values of
the control variables give the optimum value
of Z and what that optimum is.

21
 Applications of LP optimization in
Civil Engineering
• Civil engineers constantly engaged in allocating resources
in the most effective manner, within given constraints
– Resources:
• Manpower
• Money
• Materials
• Machinery
• etc.
– Constraints
• Financial
• Physical
• Institutional
• Political
• etc

22
 Applications of LP optimization in
Civil Engineering
• Example of Resources
– How many items of type X should be used?

– How much money to invest in various aspects of that venture?

– How wide should the new highway be?

– What maximum slope can we lay a pipeline for a certain project?

– What is the minimum section of a channel to hold a certain amount

of flow?

– Minimum area of steel needed to satisfy bending requirements?

– What sample size to use for data collection in order to minimize cost
but to maximize confidence about our hypothesis conclusion?, etc.
etc.

23
 Applications of LP optimization in
Civil Engineering
• Example of constraints
– Financial
• Oh man, our budget this year is only $1,000,000!

– Physical
• Material strength does not exceed 5N/mm2
• Maximum area of beam should be 1.5m2

– Institutional:
• Do not build the interstate just nearby Jefferson Elementary
School!
• In the airport area of this town, do not put up buildings higher
than 10 floors
• Cement-water ratio for this structure should exceed 0.5
• At any signalized intersection in Busytown, the green time on
the primary road should be at least twice that for the secondary
road.

24
 Applications of LP optimization in
Civil Engineering
• Note

– In civil engineering, the variables we may change in order

to obtain a certain objective are also referred to as
• Design variables,
• Decision variables, or
• Control variables

– Constraints are expressed in terms of the decision

variables.

– Remember that the number of decision variables dictates

the number of dimensions of the optimization problem.

25
 Applications of LP optimization in
Civil Engineering
• Typical Objectives (Goals):
– General
• Maximize profit
• Maximize Social Welfare
• Minimize Environmental Damage (air pollution,
noise, groundwater pollution, etc.)
• Maximize Equity
• Maximize Economic Efficiency (B/C ratio, NPV, etc.)
• Minimize Maintenance and Operational Costs
• Maximize Safety, Mobility, Aesthetic Appeal
• Minimize all Costs incurred over Cash Flow Period

26
 Applications of LP optimization in
Civil Engineering
• Typical Objectives (Goals):
– Specific
• Minimize deflections in a structural member
• Maximize flow in an open or closed channel
• Minimize consolidation settlement
• Minimize possible degree of error associated with a
measurement
• Maximize surface slope to enhance drainage
• Maximize load carrying capacity while minimizing
dead load

27
 Applications of LP optimization in
Civil Engineering
• General steps in a typical problem
Verbal statement of an optimization problem

Formulation

Mathematical statement of the optimization problem

Solving the problem

Optimal values for all the variables

Convert the optimal

variable values to a verbal
answer

Verbal answer 28
 Categories of Solutions to LP
Optimization Problems

LP

Feasible Infeasible
No solution exists which
satisfy all constraints

Bounded Unbounded
Finite optimum Infinite optimum
objective value objective value

29
 Some more interesting stuff about
LP Optimization Problems
• A constraint is said to be redundant if dropping the
constraint does not change the feasible region. In other
words, it does not form a unique boundary of the feasible
solution space

• A binding constraint is a constraint that forms the optimal

corner point of the feasible solution space

• Sometimes, the decision variables are all integers, then the

problem becomes an integer programming problem

• Sometimes, we cannot have exact solutions to a

programming problem, then we use techniques that are
collectively called heuristic programming.
30
 Some more interesting stuff about
LP Optimization Problems

Mathematical Programming

Linear Non-linear
Programming Programming

Linear
Programming with Discrete Linear Mixed
Continuous Programming Programming
Variables

Binary Linear
Programming

31
 Some more interesting stuff about
LP Optimization Problems
• Integer Programming
– When the decision variables are positive integer numbers

• Binary Programming
– When the decision variables take on only values of either 1 or 0 (examples,
“yes” or “no”, married” or “not married”).
– Also called Zero-one Programming.

• Mixed Programming
– When some decision variables are continuous while others are discrete.

– Non-linear Programming
– When the constraints are non-linear. The solution approach to non-linear
optimization problems is different from that of LP problems.
– For example, the optimal solution is not necessarily a vertex of the feasible
region.

32
 Some more interesting stuff about
LP Optimization Problems
• Some Applications of Linear Programming
– Network Analysis
– Transportation
– Hydraulics
– Environmental
– Construction
• Resource Allocation

• Note
– Linear Programming
• Is only one of many tools that can be used for optimization.
– Other tools include
• Calculus
• Trial and Error
• Simulation (typically with aid of computer)

33