Modeling & Simulation

Download as pdf or txt
Download as pdf or txt
You are on page 1of 51

1

UNIT 1
INTRODUCTION AND OVERVIEW









2

1.1 Concept of System


Simulation is the representation of a real life system by another system, which
depicts the important characteristics of the real system and allows experimentation on it.
In other words, simulation is an imitation of the reality.
A system is defined as an aggregation or assemblage of objects joined in some
regular interaction or independence.
As an example of a conceptually simple system, consider an aircraft flying under
the control of an autopilot. A gyroscope in the autopilot detects the difference between
the actual heading and the desired heading. It sends the signal to move the control
surfaces. In response to the control surface movement the airframe steers toward the
desired heading.

Desired Heading

ActualHeading

As a second example, consider a factory that makes and assembles parts into a
product. Two major components of the system are the fabrication department making the
part and the assembly department producing the products. A purchasing department
maintains a supply of raw materials and a shipping department dispatches the finished
products. A production control department receives orders and assigns work to the other
departments.
Customer Orders



Raw Finished
Materials Products
Gyroscope ControlSurfaces Airframe
Production
controlDept.
Purchasing
Dept
Fabrication
Dept
Assembly
Dept
Shipping
Dept
3

One more example of system can be taken of an engine governor, which is


required to keep the speed constant within limits, at varying loads of the engine. As the
load is changed, the speed changes, governor balls lift or lower the sleeve, which controls
the fuel supply and in turn speed. It is an automatic control system.

Load



In looking at these systems we see that there are certain distinct objects, each of
which possesses properties of interest. There are also certain interactions occurring in the
system that cause changes in the system.
In case of governor system, the components of the system, the engine, governor,
fuel pump, etc. are all physical. Their relationships are based on well-known physical
laws. Such systems are called physical systems.
The production system, on the other hand comprises of a large number of
departments, with manmade inter-relationships which cannot be represented by physical
objects. Such systems are called non-physical systems. Management systems, social
system, political system, education system etc. are all non-physical systems.
1.2 System Environment
The behavior of one component of a system affects the performance of other
components due to interconnections and hence, when a system is identified it has to be
described and studied as a whole. This methodology of tackling the system is called systems
approach. We may also call it scientific approach. The term system engineering has been
used for engineering employing systems approach.
A system is often affected by changes occurring outside the system. Some system
activities may also produce changes that do not react on the system. Such changes occurring
Engine
(Speed)
Governor(Lift
ofsleeve)
FuelPump(Controlof
FuelSupply)
4

outside the system are said to occur in system environment. An important step in modeling
systems is to decide upon the boundary between the system and its environment. The
decision may depend upon the purpose of the study.
In the case of the factory system, for example, the factors controlling the arrival
of orders may be considered to be outside the influence of the factory and therefore part of
the environment. However, if the effect of supply on demand is to be considered, there will
be a relationship between factory output and arrival of orders, and this relationship must be
considered an activity of the system.
Similarly, in the case of a bank system, there may be a limit on the maximum
interest rate that can be paid. For the study of a single bank, this would be regarded as a
constraint imposed by the environment. In a study of the effects of monetary laws on the
banking industry, however, the setting of the limit would be an activity of system.
Another example would be if in a manufacturing system the effect of a particular
work place lay out is to be studied then, activities like inspection, purchasing, packaging, in
process inventory level etc., are all external to the system. On the other hand if the idle time
of bank teller is to be studied, the arrival of customers is an activity internal to the system.
The term endogenous is used to describe activities occurring within the system
and the term exogenous is used to describe activities in the environment that affect the
system. A system for which there is no exogenous activity is said to be a closed system in
contrast to an open system which does have exogenous activities.
1.3 Elements of system
The term entity is used to denote a distinct object of interest in a particular
investigation in system.
The term attribute denotes a property of interest of an entity. There can be many
attributes to a given entity
The term activity is defined as any process that causes changes in the system.
The term event is defined as the different process or operations that take place in
the system.
5

The terms state or state of the system or state variables is used to mean a
description of all the entities, attributes and activities as they exist at one point in time.
The table 1.1 below lists the examples of entities, attributes, activities, events and
state variables for a few systems. This table does not show a complete list of all entities,
activities, events or states of the system, as the complete list can be made only when the
system is properly defined.
System Entities Attributes Activities Events State Variables
Banking Customers Balance,
Credit, status
Making
deposits,
withdrawals
Arrivals,
departures
Number of
customers
waiting, number
of busy tellers
Traffic control
lights
Vehicles Distance,
speed, type of
vehicles
Driving Stopping,
Starting
Status of traffic
signal, number
waiting, time to
green
Production Machines,
Work Pieces
Processing
rates,
breakdown
times
Machining,
welding,
sampling,
moving of
work pieces
Work arrives
at machine,
processing
starts, ends
Machine busy,
work piece
waiting, machine
down
Super Market Customers,
trolleys,
baskets
Shopping list Collecting
items, checking
out
Arrival in
store, collect
basket, end
shopping
Availability of
stock, variety,
number of
shoppers waiting
for check out
Communication Messages Length,
priority,
destination
Transmitting Sending time,
arrival at
destination
Messages waiting
to be transmitted

Some other terms which are frequently encountered while dealing with a system,
are given below.
Queue: It is a situation, where entities wait for something to happen. It may be
physical queue of people, or of objects, or a list of tasks to be performed.
6

Creating: Creating is causing an arrival of a new entity to the system at some


point in time.
Scheduling: Scheduling is the process of assigning events to the existing entities,
like the service begin and service and times.
Random Variables: It is a variable with uncertain magnitude, i.e., whose
quantitative value changes at random. The inter arrival times of vehicles arriving at a
petrol pump, or the life of electric bulbs are random variables.
Random Variate: It is a variable generated by using random numbers along with
its probability density function.
Distribution: It is a mathematical law which governs and defines probabilistic
features of the random variables. It is also called probability density function.
Let us take the example of a petrol pump, where vehicles arrive at random for
getting petrol or diesel. In this case,
Entities - vehicles which will comprise of various types of four, three and two
wheelers.
Events arrival of vehicles, service beginning, service ending, departure of
vehicles
States or state variables the number of vehicles in the system at a particular
time, number of vehicles waiting, number of workers busy, etc.
Attributes Type of vehicle i.e. four wheeler, two wheeler, petrol or diesel, size
of fuel tank, filing rate of machine, etc.
Queue Vehicles waiting in front of the pump.
Random Variables Inter arrival times of vehicles, service times and billing
times, etc.
7

Distribution The distribution may be one of the many statistical probability


density functions. It is generally assumed to be exponential in case of queuing systems.
1.4 System Modeling
A model is defined as the body of information about a system gathered for the
purpose of studying the system. In the case of a physical model, the information in
embodied in the properties of the model, in contrast to the symbolic representation in a
mathematical model. A model is not only as substitute for a system, it is also a
simplification of the system.
The task of deriving a model of a system may be divided broadly into two
subtasks:
1. Establishing the model structure
2. Supplying the data
Establishing the structure determines the system boundary and identifies the
entities, attributes and activities of the system. The data provide the values the attributes
can have and define the relationships involved in the activities. Assumptions about the
system direct the gathering of data, and analysis of the data confirms or refutes the
assumptions. Quite often, the data gathered will disclose an unsuspected relationship
that changes the model structure.
1.5 Types of Models
The models used for the analysis of various types of systems have been classified
in a number of ways, as shown in fig
8


The first major classification of models is into Physical models and Mathematical
models.
Physical Models: In physical models, physical objects are substituted for real things.
Physical models are based on some analogy between the model and simulated system. These
models may comprise of only physical objects without any dynamic interaction between
them, or physical objects where there is dynamic interaction between the components or the
attributes of the system. Representation of mechanical systems by electrical systems, of
hydraulic system by electrical systems, and vice-versa are examples of physical models.
Here, the mechanical attributes like pressure, spedd and load etc. are represented by
properties like voltage, current and resistance etc. It can be said that in physical models, a
physical representation is prepared for studying the system.
Mathematical Models: The mathematical models comprise of symbolic notations and
mathematical equations to represent the system. The system attributes are represented by the
variables like the controlled and dependent variables, while the activities or the interaction
between the variables are represented by mathematical functions. The mathematical models
9

can be static as well as dynamic and further analytical, numerical or simulation models,
depending upon the techniques employed to find its solution.
Static Models: A static model represents a system, which does not change with
time or represents a system at a particular point in time. Static models describe a system
mathematically, in terms of equations, where the potential effect of each alternative is
ascertained by a single computation of the equation. The variables used in the computations
are averages. The performance of the system is determined by summing the individual
effects.
Since the static models ignore the time dependent variables, these cannot be used
to determine the influence of changes which occur due to the variations in the parameters of
the system. The static models, do not take into consideration the synergy of the components
of the system, where the action of separate elements can have a different effect on the
modeled system than the sum of their individual effect could indicate.
Dynamic Models: The dynamic simulation models represent systems as they
change over time. Dynamic modeling is a software representation of the dynamic or time-
based behavior of a system. While a static model is a mathematical model and involves a
single set of computations of the equations, the dynamic model involves an interactive
procedure of solution. Dynamic model constantly re-computes its equations as the variation
in parameters occur with time.
Dynamic models can predict the performance of the system for a number of
alternative courses of action and can take into consideration the variance or randomness.
Most of the activities occurring in nature are random, their occurrence cannot be controlled,
but dynamic models help to predict the outcome of their behavior.
Analytical and Numerical Models: The classification of models into analytical
and numerical models is based only on the methodology to solve the mathematical model.
The mathematical models which can be solved by employing analytical techniques of
mathematics are called analytical models, while the models which require the application of
numerical methods are called numerical models. The numerical techniques involve the
application of computational procedures to solve the mathematical equations which can be
handles only by computers. The numerical computations in the interactive form are akin to
simulation. Simulation is though used in regard to physical models also, but most of the
10

applications have been found in mathematical modeling and further for modeling the
dynamic and stochastic behavior of systems.
Deterministic and Stochastic Models:
Deterministic Models: The deterministic models have a known set of inputs,
which result into a unique set of outputs. In case of pure pursuit problem, the initial position
of the target and pursuer, their speeds and the path of flight are known and can be expressed
by mathematical equations. For a given set of inputs there can be only one output. This
model is deterministic. Occurrence of the events is certain and not dependent on chance.
Stochastic Models: In stochastic simulation models, there are one or more
random input variables, which lead to random outputs. The outputs in such a case are only
estimates of true characteristics of the system. The working of a production line, where the
operation times at different work station are random is a stochastic process. The random
variables in a stochastic system may be represented by a continuous probability function or
theses may be totally discrete. Depending upon the distribution of random variables,
stochastic models can be further divided into continuous models and discrete models.
Because of the randomness of the variables, the results of stochastic simulations are always
approximate. To obtain reliable results, a stochastic simulation model has to be run for a
sufficiently long duration
1.6 System Simulation:
Forming of a physical model and experimenting on it is simulation. Developing a
mathematical model and deriving information by analytical means is simulation. Where
analytical methods are not applicable numerical methods or specific algorithms are applied
to analyze the mathematical models, which again is simulation. These models, physical as
well as mathematical can be of variety of types. Thus the term simulation described as a
procedure of establishing a model and deriving solution from it covers the whole gamut of
physical, analogue, analytical and numerical investigations. To distinguish this
interpretation of simulation from more general used of technique the term system simulation
is defined as
The technique of solving problems by observation of the performance, over
time, of a dynamic model of the system
11

Thus the dynamic model and time element are two important components of
system simulation. In many simulations, the time element may not be significant parameter.
System behavior may not be a function of time, but still the system is analyzed by step-by-
step calculations of the successive stages of the system.
For example, the size of a repair crew for maintaining a fixed fleet of buses is to
be determined. Size of the repair crew, in this case will be a step or interval in the step-by-
step analysis of the system.
1.7 Simulation A Management Laboratory:
The technique of system simulation is a very important tool of decision making.
The managerial problems are generally too complex to be solved by the analytical
techniques. Various techniques of operations research are applicable to only specific types of
situations, and require many assumptions and simplifications to be made for fitting the
problem into the model. Many of the events occurring in real systems are random with
intricate interrelationships, with their solution beyond the scope of standard probability
analysis. Under the circumstances, simulation is the only tool, which allows the management
to test the various alternative strategies. Since, simulation is a sort of experimentation, and
when used for analyzing managerial problems, it is rightly called the management laboratory.
For training the business executives, simulations called management games are used in many
universities and management institutes.
Comparison of Simulation and Analytical Methods
In contrast to analytical, models, the simulations are run rather than solved.
Running a simulation is like conducting an experiment and an experiment always gives
specific solutions depending upon the input conditions and characteristics of the system,
while the analytical methods give general solutions. For example in an optimization
problem, the mathematical model will give optimum value in a single solution and in
case of simulation model a number of simulations will have to be executed, each
resulting in a different value, one of which will approximate the optimum value.
The results obtained by simulation are approximate while the results obtained by
analytical methods are exact. To increase the reliability of simulation results, simulation
has to be run for longer periods. However in case of complex situation, mathematical
12

modeling becomes difficult and many assumptions and simplifications have to made for
constructing the model. In such cases, the analytical model may give highly approximate
or unrealistic results. It is a question of judgment up to what level the system has to be
approximated or abstracted to fit a mathematical model.
The accuracy of simulation results depends upon the level of details at which the
mode has been developed. More detailed is the model, more complex is its construction.
More detailed simulation mode requires greater time and effort to construct the model
and its execution takes longer run time. Thus a compromise has to be made for the level
of detail, to obtain reasonably realistic results.
The mathematical models can handle only a limited range of problems, while
simulation can handle all sorts of problems. It is said that when every method fails,
simulation can be employed to solve the problem.
When to use simulation:
1. Simulation is very useful for experiments with internal interactions of a
complex system, or of a subsystem within a complex system.
2. Simulation can be employed to experiment with new designs and policies,
before implementing them.
3. Simulation can be used to verify the results obtained by analytical methods
and to reinforce the analytical techniques.
4. Simulation is very useful in determining the influence of changes in input
variables on the output of the system
5. Simulation helps is suggesting modifications in the system under investigation
for its optimal performance
Steps in Simulation Study:
1. Problem formulation
2. Model construction
3. Data collection
4. Model programming
13

5. Validation
6. Design of experiment
7. Simulation run and analysis
8. Documentation
9. Implementation
Problem formulation: The clear and unambiguous description of the problem, definition of the
objectives of the study, identification of alternatives to be considered and methodology for
evaluation the effectiveness of these alternatives needs to be stated at the beginning of any study.
If the statement of the problem is provided by the policy makers, the analyst must ensure that the
problem being describes is clearly understood. Alternatively, if the problem statement is being
formulated by the analyst, the policy makers should be able to understand it and accept it. At this
stage, it should also be ascertained whether the simulation technique is the appropriate tool for
solving problem. The overall plan should include a statement of the alternative systems to be
considered, the measures of performance to be used, the methodologies of analysis to be used,
and the anticipated result of the study.
Model Construction: The model building is much of an art than science. There are no standard
rules for building a successful and appropriate model for all types of situations. There are only
certain guidelines, which can be followed. The art of modeling is enhanced by the ability to
abstract the essential features of the system, to select and modify the basic assumptions and
simplifications that characterize the system, and then improve and elaborate the model. To start
with a simple model is constructed, which is modified step-by-step, every time enriching and
elaborating its characteristics, to achieve an appropriate model, which meets the desired
objectives. In some situations, building block method is employed, where the blocks of
components of system are built and validated. These blocks are then combined to obtain model
for the complete system.
Data Collection: The availability of input data about the system is essential for the construction
of its model. The kind of data to be collected depends upon the objectives of the study, the
required data may be available as past history, or may have to be collected. The construction of
the simulation model and the collection of data have a constant interplay, and the type and
amount of data required may change as the model develops. The data is required not only as an
14

input to the model, but also some data is used to validate the simulation model. Since data
collection generally takes longer time, it should be started as early as possible.
Model Programming: Any simulation model worth the name requires enormous amount of
computations and information storage, which is possible only with the use of high speed
computers. The translation of model into a computer recognizable format is termed as
programming. Many general and special purpose simulation languages and special purpose and
problem specific simulation softwares are developed which can be used for simulation modeling.
It is for the modeler to decide, whether a simulation language is to be used or special purpose
software is to be used.
Validation: It is essential to ensure that the model is an accurate representation of the system,
which has been modeled. That the computer program performs properly and the results obtained
are identical to the ones from the real system. Validation involves both the validation of the logic
and accuracy of programming. This requires step-by-step modification of the model. It is rarely
possible to develop a reasonably large simulation model in its entirety in first step. Good deal of
debugging is required. The validation is thus an iterative process of comparing the model to
actual system behavior, identifying the discrepancies, applying corrections and again comparing
the performance. This process continues till a model of desired accuracy is obtained. The data
collected from the system is of great help in validation of the model.
Design of Experiment: The simulation is basically experimentation of the model of the system
under investigation. Simulation experiment in most of the situations involves stochastic
variables, which result into stochastic results. The average values of result obtained may not be
of desired reliability. To make the results meaningful it is essential that simulation experiment be
designed in such a way that results obtained are within some specified tolerance limits and at a
reasonable level of confidence. Decisions regarding the length of simulation run, initial
conditions, removal of initial bias, and number of replications of each run; use of variance
reduction techniques etc. has to be made.
Simulation Run and Analysis: The simulation program is run as per the simulation design; the
results are obtained and analyzed, to estimate the measures of performance of the system. Based
on the results, a decision is made, whether or not any modification in the design of simulation
15

experiment is needed. This step is a sort validation of the simulation design. It may reveal that
more runs or more replications are required.
Documentation: Documentation of a simulation program is necessary as the program can be
used by the same or different analyst in future. The program can be used with modifications for
some other identical situation, which can be facilitated if the program is adequately documented.
Documentation of the simulation model, allows the user to change parameters of the model at
will to investigate the influence on outputs, to find optimal combinations. The program should be
documented in such a way that a new analyst can easily understand it.
Implementation: There will not be any problems in the implementation of the simulation
program, if the user is fully conversant with the model, and understands the nature of its inputs
and outputs and underlying assumptions. Thus, it is important that the model user is involved in
the development of the simulation model from the very first step.
1.8 Advantages of simulation:
1. Simulation helps to learn about a real system, without having the system at all. For
example, the wind tunnel testing of the model of an aero plane does not require a full
sized plane.
2. Many managerial decision making problems are too complex to be solved by
mathematical programming
3. In many situations, experimenting with an actual system may not be possible at all. For
example, it is not possible to conduct experiment, to study the behavior of a man on the
surface of moon. In some other situations, even if experimentation is possible, it may be
too costly or risky
4. In the real system, the changes we want to study may take place too slowly or too fast to
be observed conveniently. Computer simulation can compress the performance of a
system over years into a few minutes of computer running time. Conversely, in systems
like nuclear reactors where millions of events take place per second, simulation can
expand the time required level.
16

5. Through simulation, management can foresee the difficulties and bottlenecks, which may
come up due to the introduction of new machines, equipments and processes. It thus
eliminates the need of costly trial and error method of trying out the new concepts.
6. Simulation being relatively free from mathematics can easily be understood by the
operation personnel and non-technical managers. This helps in getting the proposed plans
accepted and implemented.
7. Simulation models are comparatively flexible and can be modified to accommodate the
changing environment to the real situation.
8. Simulation technique is easier to use than the mathematical models, and can be used for a
wide range of situations.
9. Extensive computer software packages are available, making it very convenient to use
fairly sophisticated simulation models.
10. Simulation is a very good tool of training and has advantageously been used for training
the operating and managerial staff in the operation of complex systems. Space engineers
simulate space flights in laboratories to train the future astronauts for working in
weightless environments. Airplane pilots are given extensive training on flight
simulators, before they are allowed to handle real aeroplanes.
Limitations of Simulation Techniques:
1. Simulation does not produce optimum results. When the model deals with uncertainties,
the results of simulation are only reliable estimates subject to statistical errors.
2. Quantification of the variables is another difficulty. In a number of situations, it is not
possible to quantify all the variables that affect the behavior of the system.
3. In very large and complex problems, the large number of variables, and the inter-
relationships between them make the problem very unwieldy
4. Simulation is by no means a cheap method of analysis. Even small simulations take
considerable computer time. In a number of situations, simulation is comparatively
costlier and time consuming.
5. Other important limitation stem from too much tendency to rely on the simulation
models. This results in applications of the technique to some simple situations, which can
more appropriately be handled by other techniques of mathematical programming.
17

Areas of Applications
1. Manufacturing: Design analysis and optimization of production system, materials
management, capacity planning, layout planning and performance evaluation, evaluation
of process quality
2. Business: Market analysis, prediction of consumer behavior, optimization of marketing
strategy and logistics, comparative evaluation of marketing campaigns.
3. Military: Testing of alternative combat strategies, air operations, sea operations,
simulated war exercises, practicing ordinance effectiveness, and inventory management.
4. Healthcare applications: Such as planning of health services, expected patient density,
facilities requirement, hospital staffing, and estimation the effectiveness of a health care
program.
5. Communication applications: Such as network design and optimization, evaluating
network reliability, manpower planning, sizing of message buffers.
6. Computer applications: Such as designing hardware configurations and operating system
protocols, sharing and networking.
7. Economic applications: Such as portfolio management, forecasting impact of Govt.
policies and international market fluctuations on the economy, budgeting and forecasting
market fluctuations.
8. Transport applications: Design and testing of alternative transportation policies,
transportation networks - roads, railways, airways etc., evaluation of timetables, traffic
planning.
9. Environment applications: Solid waste management, performance evaluation of
environment programs, evaluation of pollution control systems.
10. Biological applications: Such as population genetics and spread of epidemics.
1.8 Continuous and Discreet Systems:
From the viewpoint of simulation, the systems are usually classified into two
categories:
Continuous systems
Discrete systems
18

Continuous Systems: Systems in which the state of the system changes continuously
with time are called continuous systems. The pure pursuit problem which represents a
continuous system since the state variables, the locations of target and pursuer, varies
continuously with time. Generally the systems in which the relationships can be
expressed by mathematical expressions as in engineering and physical sciences turn out
to be discreet systems. In continuous systems, the variables are by and large
deterministic.
Discrete Systems: Systems in which the state changes abruptly at discrete points in time
are called discrete systems. The inventory system and queuing systems are examples of
discrete systems. In inventory system, the demand of items as well as the replenishment
of the stock occurs at discrete points in time and also in discrete numbers. Similarly in
case of queuing systems the customers arrive and leave the system at discrete points in
time. Generally the systems encountered in operations research and management sciences
are discreet systems. The variables in discrete systems generally deal with stochastic
situations.
There are also systems that are intrinsically continuous but information about
them is only available at discrete points in time. These are called sampled-data systems.
The study of such systems includes the problem of determining the effects of the discrete
sampling, especially when the intention is to control the system on the basis of
information gathered by the sampling.
1.9 Monte Carlo Method
The term Monte Carlo Method is a very general term and the Monte Carlo
Methods are used widely varying from economics to nuclear physics to waiting lines and to
regulating the flow of traffic etc. Monte Carlo methods are stochastic techniques and make
use of random numbers and probability statistics to solve the problems. The way this
technique is applied varies from field to field and problem to problem. Monte Carlo method
is applied to solve both deterministic as well as stochastic problems. There are many
deterministic problems also which are solved by using random numbers and interactive
procedure of calculations. In such a case we convert the deterministic model into a stochastic
model, and the results obtained are not exact values, but only estimates. (Examples:D S Hira)
19










UNIT 2
SIMULATION OF CONTINUOUS SYSTEM











20

2.1 Characteristics of Continuous Systems:


In these system, the state of the system changes continuously generally with time.
But the time is not an essential parameter of a continuous system. In many situations, some
parameter, other than time, is continuously changed to determine the behavior of the system. The
recursive procedures for solving problems do not involve time, but are examples of continuous
systems.
A continuous system is one in which the predominant activities of the system
cause smooth changes in the attributes of the system entities. When such a system is modeled
mathematically, the variables of the model representing the attributes are controlled by
continuous functions. Generally, in continuous systems, the relationships describe the rates at
which attributes change, so that the model consists of differential equations.
The simplest differential equation models have one or more linear differential
equations with constant coefficients. It is then often possible to solve the model without the use
of simulations. Even so, the labor involved may be so great that it is preferable to use simulation
techniques. However, when nonlinearities are introduced into the model, it frequently becomes
impossible or, at least, very difficult to solve the models. Simulation methods of solving the
models do not change fundamentally when nonlinearities occur. The methods of applying
simulations to models where the differential equations are linear and have constant coefficients,
and then generalizing to more complex equations.
2.2 Comparison of numerical integration with continuous simulation system:
What is it that makes one study of a dynamic system a computation and another
study, a simulation? Although the dividing line between simulation and ordinary computation
often becomes quite thin. There are two distinctive features of simulation:
(i) In simulation, we always keep track of the state of the system explicitly. The outcome
of each step (of numerical calculations) in a simulation experiment can be interpreted
directly as the state of the system at some point in time. Simulation, essentially,
consists of constructing a state history of the system a succession of explicit state
descriptions at each instant as we move forward in time. Thus there is a one to one
21

correspondence between what the computer does and what takes place in the system.
In a numerical solution of equations no such correspondence is preserved. Usually in
pure computations shortcuts are taken, parameters program is developed. These
destroy the one-to-one correspondence between the computer steps and the original
system from which the equations were derived. Consequently the output data have to
be interpreted in the light of earlier manipulations before conclusions can be drawn
about the system.
(ii) Secondly, there is also a difference of attitude. In case of a pure numerical calculation
we only see the given set of differential equations as mathematical object and proceed
to integrate them. In simulation we look upon the equations as one of the steps in the
entire process. We know the real-life system, and we are aware of the approximations
in the model that is being simulated. Finally, by looking at the output data (which
directly represent the dynamics of the system) we are also prepared to modify the
model, if necessary.
2.3 Selection of an integration formula:
Continuous dynamic systems are generally represented by means of differential
equations. Numerical solution of these differential equations involves some integrating
procedure. Many different integration formulas are available for this purpose, and the selection
of an appropriate formula for integration is the most crucial decision to be made in a continuous
system simulation. One of the simplest possible integration formulas is the Euler formula. There
are much more efficient ways of performing numerical integration which do not rely simply on
the last known values to predict the rate at which the variables are changing and also in some
formulas the integration step size t is adjusted to match the rate at which the variables are
changing. As a matter of fact, the simple Euler formula is rarely used in practice because of the
rapid accumulation of errors. There are improved versions of the Euler formula available, such as
Euler-Richardson formula, Euler-Gauss formula. Some of the integration formulas commonly
used in simulation is: Simpsons rule (1/3 rule, 3/8 rule); trapezoidal rule; the Runge-Kutta
family of formulae (second, third and fourth order); predictor-corrector methods, and so on.
Many of these are available as standard subroutines. There is no integration method which is the
22

best for simulations. One has to consider several factors when choosing an integration formula.
The following are the few important ones.
1. Accuracy
2. Speed of computation
3. Self starting ability
4. Solution stability
5. Presence or absence of discontinuity
6. Ease with which error can be estimated
Errors: There are basically two types of computational errors that occur at each step of
integration:
(i) Round off errors are introduced because of the limited size of the computer word.
Every number has to be represented within this size. For example, suppose a number
obtained as a product of two 4-digit numbers
.2102 x .8534 = .17938468
has to be accommodated within 4-digits. It is therefore rounded off to .1794,
introducing thereby an error. Some computer systems (compiler such as ALPS,
BASIC and some versions of FORTRON) have automatic rounding built into them,
but many systems do not round off; they simple chop off which is even worse. Thus
the product in the example would end up being 0.1793.
(ii) Truncation errors are caused when an infinite mathematical series is approximated by
a finite number of terms. In continuous system simulation truncation errors arise
mainly due to the inadequacy of an integration formula being used. For example,
when Euler integration formula is used we are using only the first two terms of the
Taylor series (which is infinite).
(t + t) = (t) +
J(t)
Jt
. t +
J
2
(t)
Jt
2
.
(t)
2
2!
+
J
3
(t)
Jt
3
.
(t)
3
S!
+
Where t is sufficiently small the truncation error in Euler formula can be
approximated with
J
2
(t)
Jt
2
.
(t)
2
2!

23

The second derivative can be approximated to


J
2
(t)
Jt
2
=
1
t
_
J(t + t)
Jt
-
J(t)
Jt
_
On dividing the error term with value of f(t) to get a relative error and on getting rid
of the sign, we get
Rcloti:c crror = Er = _
J(t + t)
Jt
-
J(t)
Jt
_
t
2(t)

In each individual step Er may be small but when accumulated over hundreds of
consecutive steps in integration the truncation error could make the simulation results
meaningless.
Integration step: Choice of the step size t of integration is another very important decision.
The smaller the integration step the smaller is the error. The relative integration error in using
Euler formula, for example is given by above equation. Clearly Er can be made as small as we
please by making t sufficiently small. But the number of computation steps and therefore the
computation time will increase inversely in proportion to t. More steps would also accumulate
an increased amount of round off errors. A compromise has to be made between the conflicting
requirements of speed and accuracy. For a given accuracy, it is often possible to arrive at an
optimal combination of an integration formula and the step size.
Sometimes, when the error varies widely, it is advisable to use a varying step
integration formula. That is, the error is evaluated at each integration step and the step size t for
the next step automatically adjusted accordingly.
It often happens that the time intervals for which we require the output of
integration are much larger than the integration step size. In that case the program is so written
that values of the variables are saved once, say every 100 steps of integration.




24










UNIT 3
SIMULATION OF DISCRETE SYSTEM







25


3.1 Time Flow Mechanism:
Because of the dynamic nature of discrete event simulation models, we must
keep track of the current value of simulated time as the simulation proceeds, and we also need a
mechanism to advance simulated time from one value to another. The variable in a simulation
model that gives the current value of simulated time is called the simulation clock. The unit of
time for the simulation clock is never stated explicitly when a model is written in a general
purpose language such as FORTRAN or C, and it is assumed to be in the same units as the input
parameters. Also, there is generally no relationship between simulated time and the time needed
to run simulation on the computer.
Historically, two principal approaches have been suggested for advancing the
simulation clock:
1. Next-event time advance
2. Fixed-increment time advance
The first approach is used by all major simulation software and by most people
coding their model in a general purpose language. The second is a special case of the first.
With the next-event time advance approach, the simulation clock is initialized to
zero and times of occurrence of future events are determined. The simulation clock is then
advanced to the time of occurrence of the most imminent (first) of these future events, at
which point the state of the system is updated to account for the fact that an event has
occurred, and our knowledge of the times of occurrence of future events is also updated.
Then the simulation clock is advanced to the time of the (new) most imminent event, the
state of the system is updated, and future even times are determined, etc. This process of
advancing the simulation clock from one event time to another is continued until eventually
some pre-specified stopping condition is satisfied. Since all state changes occur only at event
times for a discrete-event simulation model, periods of inactivity are skipped over by
jumping the clock from event time to event time. It should be noted that the successive jumps
of the simulation clock are generally variable (or unequal) in size.
26


In Fixed-increment time advance approach, the simulation clock is advanced in
increments of exactly t time units for some appropriate choice of t. After each update of the
clock, a check is made to determine if any events should have occurred during the previous
interval of length t. If one or more events were scheduled to have occurred during this interval,
these events are considered to occur at the end of the interval and the system states (and
statistical counters) are updated accordingly. The fixed-increment time-advance approach is
illustrated in fig (1.49) where the curved arrows represent the advancing of the simulation clock
and e
i
(i=1,2,) is the actual time of occurrence of the ith event of any type (not the ith value of
the simulation clock). In the time interval [0,t], an event occurs at time e
1
but is considered to
occur at time t by the model. No events occur in the interval [t,2t], but the model checks to
determine that this is the case. Events occur at the times e
2
and e
3
in the interval [2t,3t], but
both events are considered to occur at time 3t, etc. Two disadvantages of fixed-increment time
advance are the errors introduced by processing events at the end of the interval in which they
occur and the necessity of deciding which event to process first when events that are not
simultaneous in reality are treated as such by the model. These problems can be made less severe
by making t smaller, but this increases the amount of checking for event occurrences that must
be done and results in an increase in execution time. Because of these considerations, fixed
increment time advance is generally not used for discrete-event simulation models when the
times between successive events can vary greatly.

27

3.2 Discrete and Continuous probability:


Stochastic Activities:
The description of activities can be of two types, deterministic and stochastic. The
process in which the outcome of an activity can be described completely in terms of its input is
deterministic and the activity is said to be deterministic activity. On the other hand, when the
outcome of an activity is random, that is there can be various possible outcomes, the activity is
said to be stochastic. In case of an automatic machine, the output per hour is deterministic, while
in a repair shop, the number of machines repaired will vary from hour to hour, in a random
fashion. The terms random and stochastic are used interchangeably.
A stochastic process can be defined as an ordered set of random variables, the
ordering of the set usually being with respect of time. The variation in the ordered set may be
discrete or continuous.
A random variable x is called discrete if the number of possible values of x (i.e.,
range space) is finite or countably infinite, i.e., possible values of x may be x
1
, x
2,
. , x
n
. The list
terminates in the finite case and continues indefinitely in the countable infinite case. The demand
of an item can be say 0, 1, 2, 3 or 4 per day, each having its own of occurrence. This demand is a
discrete activity. Color of a traffic signal light encountered by a randomly arriving vehicle may
be red, amber or green, is a discrete activity. A random variable is called continuous if its range
space is an interval or a collection of intervals. A continuous variable can assume any value over
a continuous range. For example, heights of school children, temperatures and barometric
pressures of different cities and velocity of wind are examples of continuous variables.
A stochastic process is describes by a probability law, called probability density
function.
Discrete Probability Functions:
If a random variable x can take x
i
(i=1, n) countably infinite number of values,
with the probability of value x
i
being p (x
i
), the set of numbers p (x
i
) is said to be a probability
distributions or probability mass function of the random variable x. The numbers p (x
i
) must
satisfy the following two conditions:
28

(i) p (x
i
) 0 for all values of i.
(ii) p (x

) = 1
n
=1

The probability distribution may be a known set of numbers. For example, in case
of a dice, the probability of each of the six faces is 1/6.
x 1 2 3 4 5 6
p(x) 1/6 1/6 1/6 1/6 1/6 1/6
If two coins are flipped together, the probabilities of both heads, one head one tail
and both tails are , and respectively.
Example 4.1: A pair of fair dice is rolled once. The sum of the two numbers on the dice
represents the outcome for a random variable x. Determine the probability distribution of x. What
is the probability that x is odd?
Solution: The sum of the two numbers on the dice, x will range from 1+1=2 to 6+6=12, or the
range of x is 2 to 12 both inclusive.
x = 2 can occur in one way (1+1)
= 3 can occur in two ways (1+2, 2+1)
= 4 can occur in three ways (1+3, 2+2, 3+1)
= 5 can occur in four ways (1+4, 2+3, 3+2, 4+1)
= 6 can occur in five ways (1+5, 2+4, 3+3, 4+2, 5+1)
= 7 can occur in six ways (1+6, 2+5, 3+4, 4+3, 5+2, 6+1)
= 8 can occur in five ways (2+6, 3+5, 4+4, 5+3, 6+2)
= 9 can occur in four ways (3+6, 4+5, 5+4, 6+3)
= 10 can occur in three ways (4+6, 5+5, 6+4)
= 11 can occur in two ways (5+6, 6+5)
= 12 can occur in one way (6+6)

29

Total number of ways = 1 + 2 + 3 + 4 + 5 + 6 + 5 + 4 + 3 + 2 + 1 = 36


Thus the probability of x being 2 is 1/36, being 3 is 2/36 etc. The distribution is
given in table below
x: 2 3 4 5 6 7 8 9 10 11 12
p(x): 1/36 2/36 3/36 4/36 5/36 6/36 5/36 4/36 3/36 2/36 1/36

Probability that x is an odd number is,
= p (x =3) + p (x =5) + p (x =7) + p (x =9) + p (x =11)
= 2/36 + 4/36 + 6/36 + 4/36 + 2/36 = 18/36 = 1/2 or 50%
Cumulative Distributive Function:
A cumulative distribution function is defined as a function that gives the
probability of a random variables being less than or equal to a given value. In the case of
discrete data, the cumulative distribution function is denoted by P (x
i
). This function implies that
x takes values less than or equal to x
i.

Let us consider the following data, which pertains to the demand of an item by the
customers. Total 200 customers demand has been recorded.
Number of items
Demanded (x)
Number of customers Probability
Distribution
Cumulative
Distribution
0 7 .035 .035
1 25 .125 .160
2 50 .250 .410
3 72 .360 .770
4 33 .165 .935
5 13 .065 1.000


30

The probability of zero demand is 7/200 = .035, while that of one item is 25/200
= .125, and that of two items is 50/200 = .250, and so on. The cumulative probability, P (2) = p
(0) + p (1) + p (2) = .035 + .125 + .250 = .410. The probability distribution and the cumulative
probability distribution are given in third and fourth columns of table.


0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 1 2 3 4 5
P
r
o
b
a
b
i
l
i
t
y
Numberofitemsdemanded
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5
C
u
m
u
l
a
t
i
v
e

P
r
o
b
a
b
i
l
t
y
AxisTitle
31


The probability distribution (p.m.f) and the cumulative probability distribution
(CDF) are commonly represented graphically. The date of table is shown plotted in fig above.
Since p.m.f. is defined at discrete points only, the CDF is a step function type of graph.
Continuous Probability Functions:
If the random variable is continuous and not limited to discrete values, it will
have infinite number of values in any interval, howsoever small. Such a variable is defined by a
function f(x) called a probability density function (p.d.f.). The probability that a variable x, falls
between x and x + dx is expressed as f(x) dx, and the probability that x falls in the range x
1
to x
2
is
given by
p(x) = _ (x)Jx
x
2
x
1

If x is a continuous variable on the range - to its p.d.f.f(x) must satisfy the
following two conditions:
(i) f(x) 0 - < x <
(ii) ] (x)Jx = 1

-

Example: Ref example 5.3 in D S Hira, Pg 79
Measures of probability Function:
The two important characteristics of a probability density function are the central
tendency and dispersion.
1. Central Tendency:
Three most important measures of central tendency are mean, mode and median.
Mean:
If x is a random variable, then its expected value E(x) itself is called the mean or average
value of x, and is identified as x . In the case of a discrete distribution, if there are N
observations, taking the individual values x
i
(i = 1,2, . , N), the mean is given by
32

x =
1
N
x

=N
=1

If the N observations are divided into I groups, where the ith group takes the value x
i
and
has n
i
observations, then mean is given by
x =
1
N
n

=N
=1

In case the probability p(x
i
) of occurrence of each observation x
i
is given, then the mean
is given by,
x = p(x

) x


For a continuous variable, the mean value is defined as,
x = _ x (x)Jx

-

Where f(x) is the p.d.f. of x i.e.
_ (x)Jx

-
= 1
MODE:
Mode represents the most frequently occurring value of a random variable. In other
words, when the probability density function has a peak the value of x at which the peak
occurs is called mode. A distribution may have more than one peaks, that is may be
multimodal. Depending upon the number of peaks, it may be called unimodal, bimodal,
trimodal etc. The highest peak, that is the most frequently occurring value is then called
the mode of the distribution.

33

In case of discrete distribution, mode is determined by the following inequations:


p(x = x

) p(x = x), x

x onJ
p(x = x
]
) p(x = x), x
]
x
For a continuous distribution, mode is determined as,
J
Jx
|(x)] = u
J
2
Jx
2
|(x)] = u
Median:
Median divides the observations of the variable in two equal parts. Half the values of a random
variable will fall below the median, and half above the median. For a discrete or continuous
distribution of random variable x, if X denotes the median, then
p(x X) = p(x X) = u.S
The median is easily found from the cumulative distribution, since it is the point at which
P(x)=0.5.
If any two of the mean, mode and median are known, the third can be computed by the following
relation.
Hcon - HoJc = S(Hcon - HcJion)
The relative values of mean, mode and median depend upon the shape of the probability density
function. If the probability distribution curve is symmetric about the center and is unimodal as in
fig below, then all the three mean, mode and median will coincide. If the curve is skewed to the
left as in fig, mean Is greater than mode and if it is skewed to the right mode is greater than mean.
34


A comparison of the three measures of central tendency will reveal that, the mean involves
weighting the data according to their magnitude. The median depends only on the order, while
the mode depends only on the magnitude. Though all the three have their own importance in
describing a distribution, but the mean is the most significant measure of central tendency.
Fractiles and Percentiles:
The probability density functions are sometimes described in terms of fractiles, which are
generalization of the median. While median divides the observations in two equal parts, fractiles
divide the data into a number of parts. For example, ten fractiles divide the data into ten parts as
the first tenth, second tenth and so on. The fractile taken as percentage are called percentiles
2. Dispersion:
Dispersion indicates the degree of variability of the data about the central value. In other words it
represents the spread of the distribution. Mean deviation, standard deviation and variance are
important measures of the dispersion or variability.
Mean Deviation:
The mean deviation (M.D.) of a sample from the central value a is given as,
35

H. . =
1
N
|
x

- o
|
N
=1

H. . =
1
N

(x

- o)
N
=1

The central value a may be mean, mode or median.
Variance:
Variance is associated with the mean value. If there are N observations x
i
(i = 1, 2 . N) and
having the mean value x , the variance is estimated with the formula
I =
1
(N - 1)
(x - x

)
2
N
=1
, For lorgc N.
Which can further be written as,
I =
1
(N - 1)
(x

2
-Nx
i
)
2
N
=1

When the observations fall into groups of the same size.
I =
1
(N - 1)
n

(x -x

)
2
I
=!

Where I is the number of groups and n
i
is the number of observations in group i. If the difference
between N and N-1 is ignored the ratio becomes n
i
/N, which can be replaced by p(x
i
)
I = p(x

)(x -x

)
2
I
=1

Sincc p(x

)
I
=1
= 1 onJ p(x

)
I
=1
x

= x
36

I = p(x

)
I
=1
x

2
- x
2

For a continuous variable,
I = _ (x)x
2
Jx - x
2

-

Variance is also called second moment of dispersion.
Standard Deviation:
The positive square root of variance is called standard deviation (S.D.) and is denoted by or S.
Because of the square root occurring in the definition, the standard deviation has the same
dimensions as the observations. It can therefore be compared directly with the mean of the
distribution.
Coefficient of variation:
It is the ratio of standard deviation of the mean. Since, it is a relative term without any dimension,
it is a very useful measure of dispersion.









37









UNIT 4
DENSITY FUNCTION, GENERATION OF
RANDOM NUMBERS





38


4.1 Testing of random numbers for randomness and for auto correlation
Random Numbers: Random numbers are a sequence of numbers obtained from a stochastic
process. Earlier, random numbers were generated by throwing dice, dealing cards, drawing
balls from an urn and roulette wheel. Mechanized devices were used in early 1900s, followed by
methods based on electronic circuits and schemes such as phone books, and so on. Advent of
computing led to interest in numerical and arithmetic methods. These methods generate numbers
sequentially and are easily programmed.
Classical random number generators have major defects such as short period
length and lack of higher dimension uniformity. However, nowadays there are a class of rather
complex generators which is as efficient as the classical generators while enjoy the property of a
much longer period and of a higher dimension uniformity.
Properties of Random Numbers: A sequence of random numbers, R
1
, R
2,
. R
n
must
satisfy two important properties (or statistical properties): (i) Uniformity (ii) Independence
(i) Uniformity: Random numbers should appear to be ~U(0,1) where U= Continuous
uniform distribution
The ability to generate U(0,1) random variables is of fundamental importance since
they are usually the building block for generating other random variables.
(ii) Independence: Each random number R
i
must be an independent sample drawn from a
continuous uniform distribution between 0 and 1 i.e. ~U(0,1). The Pdf (probability
density function) is given by:
Pdf : (x) = ]
1,u x 1
u, otbcrwisc

The density function is as shown in fig below:
39


Expected value of each R
i
is given by,
E(R) = _ x. Jx
1
0
= _
x
2
2
1
0
=
1
2

Variance is given by,
I(R) = _x
2
1
0
. Jx - |E(R)]
2
= _
x
3
S
1
0
-
1
2
2
=
1
S
-
1
4
=
1
12

If the interval between 0 and 1 is divided into n equal parts or classes of equal length
then,
The probability of observing a value in a specified interval is independent of
the previous values drawn.
If a total of m observations are taken, then the expected number of
observations in each interval in m/n, for uniform distributions.
Pseudo Random Numbers
The pseudo means false. But here word pseudo implies that the random
numbers are generated by using some known arithmetic operation. Since, the arithmetic
operation is known and the sequence of random numbers can be repeatedly obtained, the
numbers cannot be called truly random. However, the pseudo random numbers generated by
many computer routines, very closely fulfill the requirement of desired randomness.
If the method of random number generation that is the random number generator
is defective, the generated pseudo random numbers may have following departures from ideal
randomness.
The generated numbers may not be uniformly distributed.
40

The generated numbers may not be continuous


The mean of the generated numbers may be too high or too low
The variance may be too high or too low
There may be cyclic patterns in the generated numbers, like:
(a) Auto correction between numbers

(b) A group of numbers continuously above the mean, followed by a group continuously
below the mean.
Thus, before employing a pseudo random number generator, it should be
properly validated, by testing the generated random numbers for randomness.
Generation of Random numbers
Random numbers can be obtained by the following methods:
(a) Random numbers may be drawn from the random number tables stored in the memory of
the computer. This process, however, is neither practicable nor economical. It is a very
slow process, and the numbers occupy considerable space of computer memory. Above
all, in real system simulation, a single run may require many times more random numbers
than available in tables.
(b) An electronic device may be constructed as part of the digital computer to generate truly
random numbers. This, however, is considered very expensive.
(c) Pseudo Random Numbers be generated by using some arithmetic operations. These
methods most commonly specify a procedure, where starting with an initial number, the
second number is generated and from that a third number and so on.
Mid-square Random Number Generator: (D S HIRA pg 58)
Congruence Method or Residue Method: (D S HIRA pg 59)
Qualities of an Efficient Random Number Generator


41

4.2 Generation of Random Variates


Bernoulli Trial:
Bernoulli trial is a trial which results in two outcomes, usually called a success
and a failure. Bernoulli Process is an experiment performed repeatedly, which has only two
outcomes, say success (S) and failure (F). Here the probability that an event will occur (success)
remains constant.
In rolling a die, the probability of face i (1, 2, ., 6) coming up is always 1/6.
The probability of a head while tossing a coin is 0.5 in each flip. Firing a target (hit or miss),
contesting an election (win or lose) etc. are examples.
It is very easy to simulate a Bernoulli case. For example, if the probability that a
machine will fail in the next one hour is 0.3, we just draw a random number, if it is less than or
equal to 0.3, machine will fail otherwise it will not fail in the next hour.
Binomial Distribution:
If n Bernoulli trials are made at a time, then the distribution of the number of
successes is the Binomial distribution. For example, if a sample of N balls is drawn from
infinitely large populations of balls, having a proportion p of say red balls, then the distribution
of red balls in each sample is given by Binomial distribution.
The Binomial probability mass function of a variable x is given by,
(x) = [
n
x
p
x
(1 - p)
n-x
, x = u, 1, 2, . , n
Where p is the probability of success in an independent trial and thus 1 p is the probability of
failure. x, a positive integer represents the number of successes in n independent trials. To
generate a random variate x, n random numbers are generated. A random number less than or
equal to p i.e., r p gives a success and the total numbers of such successes gives the value of x.
The properties of Binomial distribution are
Hcon x = n. p & Iorioncc o
2
= n. p. (1 - p)
42

Negative Binomial Distribution:


In Binomial distribution, the number of trials N is fixed and the number of
successes (or failures) determines the random variate. In negative Binomial distribution, the
random variable is given by the number of independent trials to be carried out until a given
number of successes occur. If X denotes the number of trials to obtain the fixed number of K
successes then,
Probability of X trials until K successes occur = Probability of (K 1) successes
in (X 1) trials x probability of a success in X
th
trial.
p(X) = _
X - 1
K - 1
] P
K-1
((1 -P)
X-K
P)
= _
X - 1
K - 1
] P
K
(1 -P)
X-K

where P is the probability of success in a trial.
When K is an integer, this distribution is also called a Pascal distribution. When
K is equal to one, it is called a geometric distribution.
Geometric Distribution:
It is a special case of Negative Binomial distribution with K = 1. Independent
Bernoulli trials are performed until a success occurs. If X is the number of trials carried out to get
a success, then X is the geometric random variable.
p(X) = P. (1 - P)
X-1

Mean =
1
P
and Variance =
1-P
P
2
, where P is the probability of success in a trial.
Hypergeometric Distribution:
In case of Binomial distribution, the probability of success is same for all trials. In
other words, the population with proportion P of the desired events is taken to be of infinite size.
However, if this population is finite and samples are taken without replacement, the probability
of proportion of desired events in the population will vary from sample to sample. The
43

distribution, which describes the distribution of success X, in such a case, is called the
hypergeometric distribution.
If N is sample size and M is the finite size of population to start with, and P is the
probability of success in a trial in the beginning, then the hypergeometric probability function is
given by,
p(X) = H.
P!
X!
(H. P -X)!
(H. )! (H - N + X)
(N - X)!
H!
N!
(H - N)!

Where Q = 1 P
The mean is given by x = N. P and variance o
2
=
M-N
M-1
N (P. )
Poisson distribution:
Poisson distribution is the discrete version of the Exponential distribution. Let x
be a random variable which takes non-negative integer values only, then the probability mass
function,
p(x) =
z
x
c
-x
x!

Is called the Poisson distribution with parameter , where > 0
The properties of Poisson distribution are
Hcon = z
Iorioncc = z
One of the first applications of this distribution was in representing the probable
number or Prussian cavalry troopers killed each year by being kicked in the head by horses. This
distribution finds applications in a wide variety of situations where some kind of event occurs
repeatedly but haphazardly. Number of customers arriving at a service station, number of fatal
accidents per year per kilometer of highway, number of typographical errors on a page, number
44

of flaws per square yard of cloth, number of -particles emitted by a radioactive substance etc. is
some of the situations.
To generate an observation X, a random number is generated. The cumulative
probabilities of X for values 0, 1, 2, .. etc. are computed and each time compared with the
random number. As soon as the cumulative probability becomes greater than or equal to the
random number, the value of X is returned.
Empirical Distribution:
It is a discrete distribution constructed on the basis of experimental observations.
The outcomes of an activity are observed over a period of time, and the probability of each
outcome is established from the observations. For example, the demand of an item as observed
over a long period is 1, 2, 3 or 4 pieces per hour. Out of say 400 observations, 40 times the
demand is for one piece, 160 times for 2 pieces, 180 times for three pieces and 20 times for four
pieces. In other words the probabilities of demand being 1, 2, 3 or 4 pieces are
4u
4uu
= u.1,
16u
4uu
= u.4,
18u
4uu
= u.4S, onJ
2u
4uu
= u.uS rcspccti:cly.
These discrete probability distributions are an empirical distribution.
The general approach of working with such a distribution is called the integral
inverse approach. Integration, the process of adding the values is statistically called cumulative
distribution. A variate is generated using a random number, that is taking inverse of the law.
If F(x) represents C.D.F of variable x, and a random number r is such that
F (i 1) r F
i
, then the discrete variable x assumes the value x
i

i 1

2 3 . n
x
i
x
1
x
2
x
3
.. x
n
F(x) F
1
F
2
F
3
.. F
n


45

Continuous Distribution:
A random variable can be generated from any of the continuous distribution by
making use of cumulative density function of the variable. If F(x) is the C.D.F. of variable x,
then the new random variable,
= F(x)
is uniformly distributed over 0, 1 interval. If a random number r is generated, then
r = F(x) or
x = F
-1
(r)
The generation of random observations from some important continuous
distributions is given below.
Normal Distribution:
There are two probability laws, the normal probability and exponential
probability that describe most of the behavior that can be observed in real life systems. There are
many other probability laws derived from these two, but they are used only when finer precision
is needed in simulation.
The normal distribution is also called the Gaussian distribution, after the
mathematician who first described it. Because of its bell type shape, normal distribution is also
called the bell curve. Normal distribution is used most frequently to describe the distributions as
the marks obtained by a class, dimensions of parts made on a machine, number of light bulbs
which fuse per time period, heights of male or female adults etc.
The density function of the normal curve is given by
(x) =
1
2no
exp_ -
1
2
[
x - p
o

2
_ , - < x <
where is the mean of the distribution and is the standard deviation.
46

The parameter can take any real value, while standard deviation s is always
positive. The shape of a normal curve depends upon the values of and . The effect of
variations in the values of and is illustrated in fig below.
Bell Curves with = 0.25 and Different Values of

Bell Curves with = 6 and Different Values of

47

The normal random variates can be generated by a number of methods. The


commonly used methods are developed from the central limit theorem, which states that the
probability distribution of the sum of n independent and identically distributed random numbers
r
i
with mean
i
and variance
i
2
(i =1, n), approaches the normal distribution with mean p
I
n
=I

and variance o
I
2 n
=I
asymptotically as N becomes large.
Let r
1,
r
2,
.. r
n
be the random numbers (0 < r
I
< 1) and
I = r
I
n
=I

Then E(I) =
n
2

Ior (I) = :or(r

) =
n
12
n
=I

Thus V is asymptotically normal with mean n/2 and variance n/12
Let
Z =
I -
n
2
_
n
12

which is normal with zero mean and unit variance.
Now any random variate y, corresponding to the above n random numbers can be
obtained from the expression
y -
o
=
I -
n
2
_
n
12

Or
48

y = +
o
_
n
12
[I -
n
2
= +
o
_
n
12
_r
I
n
=I
-
n
2
_
Since, according to Central Limit Theorem, the normality is approached quite
rapidly even for small values of n, the value n = 12 is commonly used in practice.
y = +o _r

12
=1
- 6_, i = 1, 2, . , 12
Exponential Distribution:
The exponential distribution is used to represent an activity where most of the
events take place in a relatively short time, while there are a few which take very long times. The
p.d.f for this distribution is as shown in fig below.

The service times in queuing systems, inter arrival times of vehicles on a
highway, the life of some electronic parts, and the orders for a product received per day etc. are
some of the examples where exponential distribution can be used. Exponential distribution is
49

analogous to Geometric distribution. While geometric distribution represents a variable as the


number of customers arriving up a specific time, the exponential distribution gives the time of
next arrival of the customer.
The exponential p.d.f. is given by
(t) = _
c
-t
, t > u
u, otbcrwisc

The random variate from this distribution is generated by drawing a random
number, r.
r = (x) _c
-x
Jx
x
0
= 1 - c
-x

x = -
1

ln(1 -r) = -
1

ln r
The above expression is justified since 1 r is also a random number and r and
1 r are equally likely.
Erlang Distribution:
There is a class of distribution functions named after A.K. Erlang, who found
these distributions to be representative of certain types of telephone traffic. If there are k
independent random variables v
i
(i = 1, 2, ., k), having the same exponential distribution,
(:

) = kc
k
i

with v
i
> 0, > 0 and k a positive integer then
I = :

k
=1
has the Eilang uistiibution


50













If there are k independent variates having the same exponential distribution, then
their combined distribution is called Erlang distribution. Fig above illustrates the Erlang
distribution for several values of k. The parameter k is called the Erlang shape factor.
A random variate from Erlang distribution can be obtained by obtaining k random
variates from the exponential distribution and adding them.
Thus if,
:

= -
1
k
lnr

, i = 1,2 . . k
Then
: = :

k
=I
= -
1
k
ln r

k
=1
= -
1
k
ln_r
I
k
=I

51

Uniform Distribution:
The uniform probability law states that all the values in a given interval are
equally likely to occur. Suppose, we say that service times are uniformly distributed between 30
to 80 seconds. It means that a service will not take less than 30 seconds and more than 80
seconds. On the average service time will be 55 seconds.
Let a variable x be uniformly distributed in the interval (a,b), that is all the values
of x between a and b, (b > a) are equally likely, then
x = o + (b - o)r
where r is a random number between 0 and 1. The uniform distribution is also called Rectangular
distribution or Homogeneous distribution.
Beta Distribution:
The beta distribution is an extension of Erlang distribution and exists only
between the limits 0 and 1. A random variable x is said to have beta distribution with parameters
p and q, if its density function is given by:
(x) = _
x
p-1
(1 -x)
q-1
[ (p , q)
, i u < x < 1
u, x u onJ x 1
witb p > u, q > u
Properties of the beta distribution are:
Hcon =
p
p + q

Iorioncc =
p(p +1)
(p + q)(p +q +1)

Problems : D S Hira book

You might also like