1 MOD1 Optimization Techniques S5CSE Ktuguru
1 MOD1 Optimization Techniques S5CSE Ktuguru
COM Page 1
3. Pay-off: If it is possible to quantify each outcome, then for each course of action chosen
by the decision maker, a known pay-off will result with conditional on which SoN (or
event) has occurred. A tabular arrangement of these conditional outcome (pay-off) values
is known as pay-off matrix.
Courses of Action
Alternatives (CoA)
States of nature (SoN) S1 S2 · · · Sn
N1 p11 p12 · · · p1n
N2 p21 p22 · · · p2n
.. .. .. .. ..
. . . . .
Nm pm1 pm2 ··· pmn
Table 1: General form of Pay-off matrix
Decisions are made upon the information data available about the occurrence of events as well
as the decision situation (or environment).
1. Decision making under certainty: The decision maker knows with certainty the con-
sequences of every alternative or decision. In this model, assumed certainty means that
only one possible state of nature exist.
2. Decision making under uncertainty: Decision maker has no information at all about
various outcomes or states of nature, i.e., the decision-maker does not know the proba-
bilities of the various outcomes.
3. Decision making under risk1 : Decision maker has some knowledge regarding probabil-
ity of occurrence of each outcome or state of nature. Example, Russian Roulette, tossing
a coin etc.
WWW.KTUGURU.COM Page 2
4. Decision making under conflict1 : Decision maker neither knows completely about the
states of nature nor completely uncertain. E.g., Conflict involving two or more competi-
tors, marketing the same product.
The difference between making a decision under risk and under uncertainty is that in the case
of uncertainty, the probability distribution associated with the states of nature is either unknown
or cannot be determined. This lack of information has led to the development of the following
criteria for analyzing the decision problem:
1. Laplace (Equally likely decision): The Laplace criterion is based on the principle of in-
sufficient reason. Because the probability distributions are not known, there is no reason
to believe that the probabilities associated with the states of nature are different. The
working method is summarized as follows:
1
(a) Assign equal probability to each SoN i.e, P (Ni ) =
m
(b) Compute the average pay-off for each CoA.
(c) Select the best average (maximum for profit and minimum for loss)
(a) Locate the maximum (or minimum in case of profit) payoff in case of loss (or cost)
values corresponding to each CoA, then
(b) Select an CoA with the best anticipated pay-off value (maximum for profit and
minimum for loss).
3. Maximax or minimin (Criterion of optimism): In this criterion the decision maker se-
lects an alternative with largest (or lowest) possible pay-off value, hence it is known as
optimistic decision criterion. The working method is summarized as follows:
(a) Locate the maximum (or minimum) payoff values corresponding to each CoA, then
(b) Select an CoA with the best anticipated pay-off value (maximum for profit and
minimum for loss).
(a) From the given pay-off matrix, develop an opportunity-loss (or regret) matrix.
i. find the best pay-off corresponding to each state of nature, and
1
Not necessary for our course
WWW.KTUGURU.COM Page 3
ii. subtract all other entries (pay - off values) in that row from this value.
(b) For each course of action (strategy) identify the worst or maximum regret value.
(c) Select the course of action (alternative) with smallest anticipated opportunity-loss
value.
Tomorrow’s exam can be easy (s1 ), moderate (s2 ), or tough (s3 )’ depending on the professor’s
unpredictable mood. Hank anticipates the following scores:
s1 s2 s3
a1 85 60 40
a2 92 85 81
a3 100 88 82
1. Recommend a course of action for Hank. If the coefficient of optimism is 0.4 find the
alternatives which maximizes the score?
2. Suppose that Hank is more interested in the letter grade he will get. The dividing scores
for the passing letter grades A to D are 90, 80, 70, and 60, respectively. Would this
attitude toward grades call for a change in Hank’s course of action?
1. Laplace:
s1 s2 s3
a1 85 60 40 61.7
a2 92 85 81 86
a3 100 88 82 90←
2. Maximin
s1 s2 s3
a1 85 60 40 40
a2 92 85 81 81
a3 100 88 82 82 ←
3. Maximax
WWW.KTUGURU.COM Page 4
s1 s2 s3
a1 85 60 40 85
a2 92 85 81 92
a3 100 88 82 100 ←
4. Savage/regret
s1 s2 s3
a1 85 60 40 0 25 35 35
a2 92 85 81 0 7 11 11←
a3 100 88 82 0 12 18 18
s1 s2 s3
a1 85 60 40 85 · 0.4 + 40 · 0.6 = 47.2
a2 92 85 81 92 · 0.4 + 81 · 0.6 = 85.4
a3 100 88 82 100 · 0.4 + 82 · 0.6 = 89.2 ←
If Hank is interested in grades in letter then he can choose any alternatives from a2 and a3 .
Exercise 1. For the upcoming planting season, Farmer McCoy can plant corn (a1 ), plant wheat
(a2 ), plant soybeans (a3 ) or use the land for grazing (a4 ). The payoffs associated with the
different actions are influenced by the amount of rain: heavy rainfall (s1 ), moderate rainfall
(s2 ), light rainfall (s3 ) or drought season (s4 ). The payoff matrix (in thousands of dollars) is
estimated as
s1 s2 s3 s4
a1 -20 60 30 -5
a2 40 50 35 0
a3 -50 100 45 -10
a4 12 15 15 10
Develop a course of action for Farmer McCoy. If the coefficient of optimism is 0.65 find the
alternatives which maximizes the profit?
Under conditions of risk, the payoffs associated with each decision alternative are described by
probability distributions. For this reason, decision making under risk can be based on the ex-
pected value criterion, in which decision alternatives are compared based on the maximization
of expected profit or the minimization of expected cost.
WWW.KTUGURU.COM Page 5
2.1 Expected Value Criterion
The expected value criterion seeks the maximization of expected (average) profit or the mini-
mization of expected cost. The data of the problem assumes that the payoff (or cost) associated
with each decision alternative is probabilistic.
Example 2. Suppose that you want to invest $10,000 in the stock market by buying shares
in one of two companies: A and B. Shares in Company A, though risky, could yield a 50%
return on investment during the next year. If the stock market conditions are not favorable (i.e.,
“bear” market), the stock may lose 20% of its value. Company B provides safe investments
with 15% return in a ”bull” market and only 5% in a “bear” market. All the publications you
have consulted (and there is always a flood of them at the end of the year!) are predicting a
60% chance for a “bull” market and 40% for a “bear” market. Where should you invest your
money?
Figure 2: Decision-tree representation of the stock market problem. A square () represents a
decision point and a circle (
) represents a chance event.
In general, a decision problem may include n states of nature and m alternatives. If Pj (> 0)
is the probability of occurrence for state of nature j and pij is the payoff of alternative i, given
WWW.KTUGURU.COM Page 6
state of nature j (i = 1, 2, ..., m; j = 1, 2, ..., n.), then the expected payoff for alternative i is
computed as
EVi = pi1 P1 + pi2 P2 + ... + pin Pn , i = 1, 2, ..., m
The best alternative is max EVi .
i
Exercise 2. Farmer McCoy can plant either corn or soybeans. The probabilities that the next
harvest prices of these commodities will go up, stay the same, or go down are .25, .30, and
.45, respectively. If the prices go up, the corn crop will net $30,000 and the soybeans will net
$10,000. If the prices remain unchanged, McCoy will (barely) break even. But if the prices go
down, the corn and soybeans crops will sustain losses of $35,000 and $5000, respectively.
Exercise 3. You have the chance to invest in three mutual funds: utility, aggressive growth, and
global. The value of your investment will change depending on the market conditions. There is
a 10% chance the market will go down, 50% chance it will remain moderate, and 40% chance it
will perform well. The following table provides the percentage change in the investment value
under the three conditions:
Exercise 4. 1. Discuss the difference between decision making under certainty, uncertainty
and risk. Explain using examples.
2. Define expected monetary value (EMV), expected opportunity loss (EOL) and incremen-
tal (marginal) analysis. Explain each one with suitable examples.
3. What do you mean by decision tree analysis? Explain with a suitable example.
Queuing theory deals with problems which involve queuing (or waiting). Typical examples
might be:
WWW.KTUGURU.COM Page 7
1. banks/supermarkets - waiting for service
2. computers - waiting for a response
3. failure situations - waiting for a failure to occur e.g. in a piece of machinery
4. public transport - waiting for a train or a bus
Queuing theory is not an optimization technique, rather, it determines the measures of perfor-
mance of waiting lines, such as the average waiting time in queue and the productivity of the
service facility, which can then be used to design the service installation.
The principal actors in a queuing situation are the customer and the server. Customers are gen-
erated from a source. On arrival at a service facility, they can start service immediately or wait
in a queue if the facility is busy. When a facility completes a service, it automatically “pulls”
a waiting customer, if any, from the queue. If the queue is empty, the facility becomes idle
until a new customer arrives. From the standpoint of analyzing queues, the arrival of customers
is represented by the inter-arrival time between successive customers, and the service is de-
scribed by the service time per customer. Generally, the inter arrival and service times can be
WWW.KTUGURU.COM Page 8
Figure 4: The basic queuing process
Queue size plays a role in the analysis of queues, and it may have a finite size, as in the buffer
area between two successive machines, or it may be infinite, as in mail order facilities. When a
new customer’s arrival into the service system is restricted by the number of customers already
in the system (customer’s in the queue and those in service), the input source is considered
as finite (e.g., machines requesting the service of a repair person). Alternatively, if a new
costumer’s arrival is independent of the number of customers in the system, then input source
is considered to be infinite (e.g., calls arriving at a telephone exchange).
Queue discipline, which represents the order in which customers are selected from a queue, is
an important factor in the analysis of queuing models. The most common discipline is
Figure 5: An elementary queueing system (each customer is indicated by a C and each server
by an S).
The queuing behavior of customers plays a role in waiting-line analysis. “Human” customers
can be patient or impatient customers.
WWW.KTUGURU.COM Page 9
1. Balking: customers deciding not to join the queue if it is too long.
2. Jockeying: customers switch between queues if they think they will get served faster by
doing so.
3. Reneging: customers leave the queue if they have waited too long for service.
4. Collusion: Several customers may cooperate and only one of them may stand in the
queue.
The design of the service facility may include parallel servers (e.g., post office or bank opera-
tion). The servers may also be arranged in series (e.g., jobs processed on successive machines),
or they may be networked (e.g., routers in a computer network).
A notion for summarizing the characteristics of the queue is given by the following format:
(a/b/c) : (d/e/f )
where,
a = Arrival distribution
b = Departure (service time) distribution
c = Number of parallel servers (1, 2, ..., ∞)
d = Queue discipline
e = Maximum number (finite or infinite) allowed in the system
f = Size of the calling source (finite or infinite)
The standard notation for representing the arrivals and departures distributions(a and b) is
M = Markovian (or Poisson) arrivals or departure distribution (or equivalently exponential in-
terarrival or service time distribution)
D = Constant (deterministic) time
Ek = Erlang or gamma distribution of time (or, equivalently, the sum of independent exponen-
tial distributions)
GI = General (generics) distribution of interarrival time
G = General (generic) distribution of service time.
For example, (M/D/10) : (GD/20/∞) used Poisson arrivals, constant service time, and 10
parallel servers. The queue discipline is GD, and there is a limit of 20 customers on the entire
system. The size of the source from which customers arrive is infinite.
Exercise 5. 1. In each of the following situations, identify the customer and the server:
10
WWW.KTUGURU.COM Page 10
(a) Planes arriving at an airport.
(b) Taxi stand serving waiting passengers.
(c) Tools checked out from a crib in a machining shop.
(d) Letters processed in a post office.
(e) Registration for classes in a university.
(f) Legal court cases.
(g) Check-out operation in a supermarket.
(h) Parking lot operation.
For each of the above situations identify the following: (a) nature of the calling source
(finite or infinite), (b) nature of arriving customers (individually or in bulk), (c) type of
the interarrival time (probabilistic or deterministic), (d) type of service time, (f) queue
capacity (finite or infinite), and (g) queue discipline.
2. Explain the basic queuing process. What are the important random variates in queuing
system to be investigated?
3. What do you understand by (i) queue discipline, (ii) arrival process (iii) service process?
4. What is queuing theory? In what situations can queuing theory be applied successfully?
Give appropriate examples.
Simulation deals with a computerized imitation of the random behavior of a system for the
purpose of estimating its measures of performance.
Within the overall task of simulation, there are three primary sub-fields: model design, model
execution and model analysis
1. the model is very complex with many variables and interacting components;
11
WWW.KTUGURU.COM Page 11
Figure 6: Steps of Simulation Process.
Monte Carlo simulation is a computerized mathematical technique that allows people to ac-
count for risk in quantitative analysis and decision making. The technique is used by profes-
sionals in such widely disparate fields as finance, project management, energy, manufacturing,
engineering, research and development, insurance, oil & gas, transportation, and the environ-
ment.
Monte Carlo simulation furnishes the decision-maker with a range of possible outcomes and the
probabilities they will occur for any choice of action. It shows the extreme possibilities—the
outcomes of going for broke and for the most conservative decision—along with all possible
consequences for middle-of-the-road decisions.
The technique was first used by scientists working on the atom bomb; it was named for Monte
Carlo, the Monaco resort town renowned for its casinos. Since its introduction in World War II,
Monte Carlo simulation has been used to model a variety of physical and conceptual systems.
1. Continuous models deal with systems whose behavior changes continuously with time.
These models usually use difference - differential equations to describe the interactions
among the different elements of the system. A typical example deals with the study of
world population dynamics.
2. Discrete models deal primarily with the study of waiting lines, with the objective of
determining such measures as the average waiting time and the length of the queue.
These measures change only when a customer enters or leaves the system. The instants at
which changes take place occur at specific discrete points in time (arrivals and departure
events), giving rise to the name discrete event simulation.
12
WWW.KTUGURU.COM Page 12
1. Monte Carlo simulation performs risk analysis by building models of possible results by
substituting a range of values—a probability distribution—for any factor that has inherent
uncertainty.
2. Then, it calculates results over and over, each time using a different set of random values
from the probability functions.
3. Depending upon the number of uncertainties and the ranges specified for them, a Monte
Carlo simulation could involve thousands or tens of thousands of recalculations before it
is complete.
1. Normal - Or “bell curve.” The user simply defines the mean or expected value and a
standard deviation to describe the variation about the mean. Values in the middle near
the mean are most likely to occur. It is symmetric and describes many natural phenomena
such as people’s heights. Examples of variables described by normal distributions include
inflation rates and energy prices.
2. Lognormal - Values are positively skewed, not symmetric like a normal distribution. It
is used to represent values that don’t go below zero but have unlimited positive potential.
Examples of variables described by lognormal distributions include real estate property
values, stock prices, and oil reserves.
3. Uniform - All values have an equal chance of occurring, and the user simply defines
the minimum and maximum. Examples of variables that could be uniformly distributed
include manufacturing costs or future sales revenues for a new product.
4. Triangular - The user defines the minimum, most likely, and maximum values. Values
around the most likely are more likely to occur. Variables that could be described by a
triangular distribution include past sales history per unit of time and inventory levels.
5. Program (or Project) evaluation and review technique or PERT - The user defines the
minimum, most likely, and maximum values, just like the triangular distribution. Values
around the most likely are more likely to occur. However values between the most likely
and extremes are more likely to occur than the triangular; that is, the extremes are not as
emphasized. An example of the use of a PERT distribution is to describe the duration of
a task in a project management model.
6. Discrete - The user defines specific values that may occur and the likelihood of each. An
example might be the results of a lawsuit: 20% chance of positive verdict, 30% change
of negative verdict, 40% chance of settlement, and 10% chance of mistrial.
13
WWW.KTUGURU.COM Page 13
During a Monte Carlo simulation, values are sampled at random from the input probability
distributions. Each set of samples is called an iteration, and the resulting outcome from that
sample is recorded. Monte Carlo simulation does this hundreds or thousands of times, and the
result is a probability distribution of possible outcomes. In this way, Monte Carlo simulation
provides a much more comprehensive view of what may happen. It tells you not only what
could happen, but how likely it is to happen.
1. Probabilistic Results. Results show not only what could happen, but how likely each
outcome is.
2. Graphical Results. Because of the data a Monte Carlo simulation generates, it’s easy to
create graphs of different outcomes and their chances of occurrence. This is important
for communicating findings to other stakeholders.
3. Sensitivity Analysis. With just a few cases, deterministic analysis makes it difficult to
see which variables impact the outcome the most. In Monte Carlo simulation, it’s easy
to see which inputs had the biggest effect on bottom-line results.
4. Scenario Analysis: In deterministic models, it’s very difficult to model different combi-
nations of values for different inputs to see the effects of truly different scenarios. Using
Monte Carlo simulation, analysts can see exactly which inputs had which values together
when certain outcomes occurred. This is invaluable for pursuing further analysis.
Exercise 6. 1. Distinguish between solution derived from simulation model and solution
derived from analytical models.
2. What is Monte Carlo simulation? Explain the steps of Monte carlo simulation.
3. Define simulation.
4. State major reason for using simulation. Briefly describe the application in any 3 fields.
5. Discuss the Monte Carlo method of solving a problem. Describe the kind of problem for
which Monte Carlo will be an appropriate method of solution.
5 Introduction to Optimization
Optimization - (noun) - The act of obtaining the best result under given circumstances.
Operation research is a part of mathematics concerned with the application of scientific meth-
ods and techniques to decision making problems and with establishing the best or optimal
solutions. There is no single method available for solving all optimization problems efficiently.
14
WWW.KTUGURU.COM Page 14
Hence a number of optimization methods have been developed for solving different types of
optimization problems.
1. Data (Costs or demands, fixed operation conditions of a reactor, Capacities and so on.)
4. Objective function
maximize objective
by choosing variables
subject to constraints
When building mathematical models for optimization will usually lead to structured problems
such as: Linear Programming (LP), Mixed Integer Linear Programming (MILP), Non Linear
Programming (NLP), Mixed Integer Non Linear Programming (MINLP). Besides Building a
model and classifying the problem a solver is needed, i.e. a piece of software which has a
set of algorithms capable of solving the problems. The terms mixed integer optimization, and
mixed integer programming and discrete optimization are synonyms. Discrete Optimization
involves variables restricted to integer values, for example counts (no. of reactors), columns,
tanks, ships) and decisions (Yes or No).
There are also heuristic 2 methods, which can find feasible points of optimization problems.
However, optimality of these points can only be proved when used in combination with exact
mathematical optimization methods. That is why these methods are not optimization methods
in the strict sense. Heuristic methods include simulated annealing, tabu search, evolution strat-
egy, constraint programming, ant colony optimization, neural networks and genetic algorithms.
The hybrid approaches combine elements form mathematical optimization and heuristic meth-
ods. The hybrid methods may have great impact on supply chain and scheduling problems in
the future.
Genetic algorithms can be regarded as an optimization method that uses evolutionary tech-
niques, in which the fittest solutions survive. In genetic programming, it is different programs
that compete against each other. Chemistry and Chemical Technology have been among the
most important application areas for genetic algorithms.
15
WWW.KTUGURU.COM Page 15
• Mathematical programming methods. These are useful in finding the minimum of a func-
tion of several variables under a prescribed set of constraints. Eg., Calculus of variation,
Linear programming, Dynamic programming, Game theory, Genetic algorithm, Neural
Networks, etc
• Stochastic process techniques. These are used to analyze problems which are described
by a set of random variables of known distribution. Eg., Statistical decision theory, Qeu-
ing theory, Simulation methods etc
• Statistical methods. These are used in the analysis of experimental data and in the con-
struction of empirical models. Eg., Pattern recognition, design of experiments, etc.
Some typical application from different engineering disciplines are given below:
3. Design of civil engineering structures such as frames, foundations, bridges, towers, etc.
4. Minimum weight design of structures for earthquake, wind, and other other random load-
ing.
7. Design of pumps, turbines, and heat transfer equipment for maximum efficiency, pipeline
networks.
10. Analysis of statistical data and building empirical models from experimental results to
obtain the most accurate representation of the physical phenomenon.
11. Allocation of resource and services among several activities to maximize the benefit.
12. Controlling the waiting and idle times and queuing in production lines to reduce the costs.
13. Planning the best strategy to obtain macimum profit in the presence of a competitor.
16
WWW.KTUGURU.COM Page 16
Figure 7: Classifying Optimization Problem
• Problem to find the values to parameters that make some prescribed function of
these parameters minimum subject to certain constraints.
• Problem to find the parameters which are all continuous function of some other
parameter, that minimizes an objective function subject to a set of constraints.
5. Permissible values of the design variables - Depending on the values permitted for the
design variables, it can be classified as integer and real valued programming problems
17
WWW.KTUGURU.COM Page 17
6. Deterministic Nature of the variables - classified into deterministic and stochastic pro-
gramming problems.
7. Separability of the functions - separable and non separable programming problems bases
on the separability of the objective and constrain function.
References
[1] Hamdy A Taha. Operations Research An Introduction. Pearson Prentice Hall, 2010.
[2] J. K. Sharma. Operations Research: Problems And Solutions. Macmillan India Limited,
2003.
[3] Singiresu S. Rao. Engineering and Optimization: Theory and Practice. New Age Interna-
tional Publishers, 2005.
18
WWW.KTUGURU.COM Page 18