Title

Simulation and Modeling

Text
Discrete-Event System Simulation

By
Banks Corson II & Nelson Nicle
Fifth Edition 2014
Pearson Edition Limited 2014

Elaborated by

1
:Syllabus

1 Basic concepts and terminology

•Concepts of a system
•System Methodology
•Advantages and disadvantages of simulation
•Simulation terminology

2 Probability and distribution theory

•Probability
•Set theory, compound events
•Conditional probability, independent events
•Discrete distributions
•Continuous distributions
•Function of a random variable
•Moments
•Some common distributions

2
3 Estimation and statistical tests
•Empirical distributions
•Estimation
•Tests of hypotheses
•The Chi-Squire goodness-of-fit test
•The Kolmogorive-Smirnov test
Generation of Random Numbers 4
•Pseudo random numbers
•Congruential generators
•Testing and validating Pseudo random numbers
5 Introduction to queuing theory
•Review of the Poisson and Exponential
distributions
•The M/M/1/∞/FIFO system
•Summary measures for the M/M/1/∞/FIFO
system
•The M/M/1/k/FIFO system
•M/M/C/∞/FIFO system

3
6 Discrete system simulation
•Examples,
•Time management methods,
•Collecting and recording simulation data
•Analysis of simulation results
•Evaluation of the simulation model
7 Languages for discrete system
simulation
• Language characteristics
• Use multipurpose languages
• Special-purpose languages:
1.GPSS
2.SIMSCRIPT II.5
3.SLAM II
4.GASP IV
8 Introduction to continuous system

simulation.
•Models of continuous systems
•Solution of linear differential equations
•Analog computing
•Digital simulation of continuous systems
•Continuous system simulation languages
4
Simulation Principles and Methods

1. Introduction
The basic idea of Simulation is to build an
experimental device (model) that act like
(simulate) the real system in certain important
aspects. The purpose is to understand or
evaluate the behavior of a complex real-world
system over extended period without risk to the
real system performance.

2. Why using Simulation

a) If the experimentation on the real-world
system is not feasible or not possible:

5
• Too expensive, (not feasible)
• Too risky (atomic problems),(not possible)

• Too complex,
• Not existing system,
• Intractable.

b) Simulation is the best (may be the only)

technique available under conditions of
uncertainty due to stochastic (random)
variables, or nonlinearity. Mathematical
treatment of such systems is frequently not
possible.
c) Computer simulation gives control over time,
which may be compressed or expanded, such
as an aircraft; simulation may gather data on
many months of operations in minutes.
6
3. Simulation Application
Simulation can be used in Theoretical problems in
basic science areas such as math., physics, and
chemistry.
It can be useful in practical problems in all aspects
of life such as industrial problems, business,
economic, biomedical systems, social problems
(Population dynamics), space researches, and War
strategies.
4. Basic concepts and terminology
a) System: is defined as an isolated collection of
interacting components or objects with defined set of
interactions among them. A Jet aircraft is an excellent
example of a complex system consisting of mechanical,
electronic, chemical and human components.

7
b) State of a system: It is the minimal
collection of information with which the system future
behavior can be predicted. For the Jet aircraft, the
state of the system can be determined by Speed,
Direction of travel, weather conditions, number of
passengers, and amount of remaining fuel.

c) An Activity: It is the process or Event

Which changes the system state.

8
d) System Classification:
o Natural and Man-Made,

o Deterministic

o Stochastic (Probabilistic)

o Continuous: A system whose changes in its state

occurs continuously over time (liquid flow in a
pipeline),

o Discrete: Growth of world population with respect

to time.

9
5. System methodology
When simulation is used to solve a problem, the
following time-tested steps, or stages are
applied:

A. Observation of the system,

B. Formulation of hypotheses or theories that account
for the observed behavior,
C. Prediction of the future behavior of the system
based on the assumption that the hypotheses are
correct,
D. Comparison of the predicted behavior with the
actual behavior.

10
The system being studied may impose constraints
on certain steps of this scientific method. For
example, consider the simulation of a system that
does not yet exist. Obviously the observation of
such system is not possible, but the simulation of
such system may still be possible if the analysis is
carefully conducted and if the ultimate
requirements are known. The problem solving
process is divided into four phases:
. Planning
. Modeling
. Validation
. Application.

11
a) Planning or premodeling includes the initial
encounter with the system, the problem to be
solved, and the factors pertaining to the system
and its environment that are likely to affect the
solution of the problem. This means the problem
must be well defined. Obviously the more accurate
and precise the problem statement is, the more
smoothly the solution process can proceed.
Resources must be considered and estimated such
as money, time, personal, and special equipment. If
crucial resources are not available, solution of the
problem can be judged infeasible before a
significant amount of time or money is spent.

12
b) Modeling in this phase the analyst constructs the
system model, which is a representation of the real
system. The characteristics of this model should be
Representative (not identical) of the characteristics of the
real system. You must select some minimal set of the
system's characteristics so that the model approximates
the real system to be cost-effective and manageable.
There are many types of simulation models:

1-Descriptive models (verbalizations of the system's

composition)
2-Physical models (scaled facsimiles of the system:
aircraft, cars, building,..)
Mathematical models (abstract expressions of the -3
relationships among the system variables: Newton's laws
of motion,..)
Flowcharts ( Logical interaction between system's-4
components)
Schematics ( Logical interaction between system's-5
components)
Computer programs : If the system is too complex -6
then the reduction technique or subsystem modeling is
.used
13
c) Validation is to check that the system or the model
is a correct representation of the real system. But
verification means that the logical of a computer program
is ok. So, a verified computer program can represent
invalid model . There are many techniques used to
validate a computer program:

 Compare the results of simulation model with

results historically produced by the real system
operating under the same conditions.
 Use simulator to predict results, and then compare
with the real data for some period.

Various statistical procedures can be used to determine

whether the simulated output could have been produced
by the real systems and vice versa. Two such methods
are the Chi-square goodness-of-fit test and the
kolmogrove-Smirnov test.

14
d) Application Once the model has been properly
validated, it can be applied to solve the problem at
hand. A simulator may work error-free for a long period
until it encounters a new and perhaps unique
combination of program parameters that generates the
next error.

15
6.Advantages and disadvantages of
simulation

Phillips Ravindran and Solberg OR principles and

practice, New York John Wiley,1976] stated that
simulation is one of the easiest tools of
management science to use but probably one of
the hardest to apply properly and perhaps the
most difficult one from which to draw accurate
conclusions.
Adkins and Pooch list 5 advantages of simulation
modeling:
a) It permits controlled experimentation,
b) It permits time compression,
c) It permits sensitivity analysis by manipulation of
input variables,
d) It does not disturb the real system,
e) It is an effective training tools.

16
They also list 4 disadvantages:
a) A simulation model may become expensive in terms of
manpower and computer time,
b) Extensive development time may be encountered,
c) Hidden critical assumption may cause the model to
diverge from reality,
d) Model parameters may be difficult to initialize. These
may require extensive time in collection, analysis, and
interpretation.

Thus although simulation has proved an effective

approach to problem solving, it has its drawbacks.
The researcher should be aware of these
drawbacks before becoming committed to this
approach.

17
 Examples
Example1
The first example we are going to see is the
simulation of a tossing of a fair coin. First step is
analyzing the problem. The fair coin means that
when tossing that coin the probability of head equal
the probability of tail equal 50%. So, using a digital
computer to simulate this phenomenon we are going
to use a uniform Random number generated by the
package you are using or you can write its code.
Uniform random number means that you have a set
of random number between 0 and 1 all with the
same probability. But most languages generate
uniform random number integer from a to b with
equal probability.

18
See the following program written in c++
#include <iostream>
#include <stdlib.h>
void main(void)
{
int x,nuber_or_trials, head=0, tail=0;
float phead,ptail,error_head,error_tail;
cout<<"enter number of trials"<<endl;
cin>>nuber_or_trials;
for(int i=0;i<nuber_or_trials;i++)
{
x=random(2);
if (x==1) head++;
else ++tail;
}
phead=head*1.0/nuber_or_trials;
ptail=tail*1.0/nuber_or_trials;
error_head = abs(((0.5 - phead)/0.5)*100);
error_tail = abs(((0.5 - ptail)/0.5)*100);
cout<<"probability of head= " <<phead<<"
with error ="<<error_head<<"%"<<endl;
cout<<"probability of tai = "<<ptail<<" with
error ="<<error_tail<<"%"<<endl;
cin>>x;
}

19
The output results are:

Enter number of trials = 1

probability of head= 0 with error =100%
Probability of tail = 1 with error =100%
Enter number of trials =5
probability of head= 0 with error =100%
Probability of tail = 1 with error =100%
Enter number of trials = 32767
probability of head= 0.496292 with error =0%
Probability of tail = 0.503708 with error =0%

20
Example2
Get the average daily demand for a small grocery store
selling a fresh bread according to the following table:

Device to generate
Daily demand D Probability Of chance outcomes
demand P(D) Number and color
of balls
100 0.20 20 Red
110 0.50 50 Blue
120 0.30 30 Yellow

Note that the proportion of balls of a specific color

corresponds exactly to the probability of a specific
level of daily demand.

21
To simulate the daily demand (5 days)
 Draw one ball at a time, notice its color and then place
it back in the bowl. Then translate the outcomes into
unique values of demand.

Sample Color of Day of the Simulate

Number ball Week d
Demand
1 Blue Monday 110
2 Blue Tuesday 110
3 Yellow Wednesday 120
4 Yellow Thursday 120
5 Red Friday 100
Total 560

 Expected value of simulated demand

= 560/5=112 units / day
 Analytical solution
Expected daily demand
=100(0.2) + 110(0.5) +120(0.3) = 111 units / day

22
Example3
Use the one-digit Random Number (6, 3, 5, 0, 8)
to generate random observations for :
a)Throwing an unbiased coin,
b)Throwing a die,
c)The color of a traffic found by a randomly arriving car
when it is green 40% of the time, yellow 10% of the
time, and red 50% of the time.

23
Solution

a) To simulate a coin using one digit R.N.(0-9), let 0 - 4

Represent a Head and 5 – 9 Represent a Tail; so the
solution will be
R.Observations
(6, 3, 5, 0, 8) (T, H, T, H, T)

b) To simulate a die using number 0 – 9. Let 1 – 6

represents the faces of the die; then 0, 7, 8 and 9 are
rejected. So the solution will be

R.Observations
(6, 3, 5, 0, 8) (6,3,5,reject,reject)

24
c) Divide 0 – 9 into 3 classes according to the given
probability:
o Green 40% 0, 1, 2, 3
o Yellow 10% 4
o Red 50% 5, 6, 7, 8, 9

R.Observations
So (6,3,5,0,8) (Red,Green,Red,Green,Red)

25
Probability and Distribution Theory
Definition1 A random variable is a quantity whose value
is determined by the outcome of a Random
Experiment ,i.e., the Random Variable X is a function
whose domain is the event space (set of all possible
outcomes of a random experiment) and whose range is
some subset of the real numbers.
X : E R
Set Theory and compound Events
Definition2
An event is some subset of the event space of a
random experiment.
Ω : Event space or the collection of all the possible outcomes of

the experiment.

Example
Consider the experiment of tossing two dice and noting
the sum of the faces. The event space is Ω = {2, 3, 4,
5, 6, 7, 8, 9, 10, 11, 12} and some events are
E1 = {2} E2 = { 3, 4, 5} E3 = {2,4,6,8,10.12} (an even
number appears.

Definition3
The Complement of an event E is Ē which contains the
set of elements that are in Ω but not in E.
Note that: E + Ē = Ω , E * Ē = Φ (null set )
26
Example

Ē1 = {3,4,5,6,7,8,9,10,11,12}
Ē2 ={2,6,7,8,9,10,11,12}
Ē3 = {3,5,7,9,11}

Definition4
Let a random experiment has N equally likely
outcomes. Let some event E is composed of n
outcomes, so the probability of E is denoted by
P(E) = n/N , P(Ē ) = 1 – P( E ) = (N – n )/N
Note : P(Ω ) = 1 , P( Φ ) = 0.

Definition5
The Union of two events E1 and E2 is denoted
by E1 + E2 and it is defined as the outcomes in
either E1 OR E2 OR both.
The intersection of two events E1 & E2 is denoted by
E1 . E2 and is defined as the outcomes that the
events have in common .Two events that have no
outcomes in common are said to be
mutually exclusive.

27
 Discrete Distribution

Example Tossing a single die Let X a random Variable

that counts the spots of the side facing up. Then the
values of X is 1, 2, 3, 4, 5, 6
X 1 2 3 4 5 6
P(X = x ) 1/21 2/21 3/21 4/21 5/21 6/21

P(x)

6/21

5/21

4/21
3/21
2/21
1/21
x
1 2 3 4 5 6

28
The properties that every discrete probability
function must have :
1) P(xi ) ≤ 0 for all x It is called probability density
function (p.d.f.)

2)
 P(x )  1
i 1
i

Another useful function is the Cumulative distribution

CDF=F(x) = P ( X ≤ x ) {probability that X ≤ x }

F(x)

3/4

1/2

1/4

1 2 3 4 5 6
29
The CDF has the following properties:
1) 0 ≤ F ( X ) ≤ 1 - ∞ <x<∞
2) If X1 ≤ X2 F ( X 1) ≤ F ( X 2) that is F is
monotonically increasing.
3) Lim F ( X ) =F(∞) = 1
x ∞

Lim F ( X ) =F(-∞) = 0
x -∞

Expected value (Mean)



E(X) =  X P( X
i 0
i i
)

For the above example

E(X) = (1)(1/21) + (2)(2/21) + (3)(3/21) + (4)(4/21) +
(5)(5/21) + (6)(6/21) = 91/21

30
Variance

V ( X )   [ xi  E ( X ) ] 2 P ( x i )
i 0

 E [ ( X  E ( X ) )2 ]
 E ( X 2 )  ( E ( X ) )2

For the above example V(X) = 2.222

1/ 2
Standard Deviation of X is given by S(X) =
V( X )
For the above example
S(X)=[2.222]^1/2 b= 1.4907

31
 Continuous Distribution
• 1- f(x) =0 if x is not in the range of X,

• 2- f(x) ≥0,


• 3-  f ( x)dx  1


X1

• 4- F(x1) =  f ( x)dx



• 5- E(X) =  xf ( x)dx



2
( x  E ( X )) f ( x)dx
• 6- V(X ) = 


2 2
x f ( x ) dx  ( E ( X ))
• = 

32
 Function of a Random Variable

• Let X denote a R.V. Suppose the R.V. Y is related to

X by Y = aX + b where a, b are arbitrary constant.

• E(Y) = E(aX+b)
 

• =  
(ax  b) f ( x) dx  a  xf ( x)dx  b  f ( x)dx
 

• So E(Y) = aE(X) + b

• V(X) = V(aX + b)

•   (ax  b) 2 f ( x)dx  ( E (aX b)) 2



33
   
  a x f ( x)dx  2ab  xf ( x)dx  b
2 2 2
 f ( x ) dx  ( E ( aX  b ) 2

  

  a 2 x 2 f ( x)dx  2ab E ( X )  b 2  ( E (aX  b) 2


  a 2 x 2 f ( x)dx  a 2 ( E ( X )) 2


 a 2 [  x 2 f ( x)dx  ( E ( X )) 2 ]


 a2 V ( X )

34
Some Common Distribution

1) Bernoulli Distribution
It is a discrete distribution having only two
outcomes. Let X has been defined and it takes
the value o with probability p and 1 with
probability q=1-p i.e.,
X p(x)
0 p
1 q
CDF = F(x) = 0 x< 0
p 0≤x
<1
1 x≥1
μ = E(X) = 0.p + 1.q = q
 2 V ( X )  E ( X 2 )  ( E ( X ) ) 2
 0. p  1.q  q 2  q (1  q )  pq
  ( pq )1 / 2

35
p(x) F(x)

q 1

p p

x x
0 1 0 1

Example
Consider the experiment of tossing a fair coin. Let X
assume the value o if a head appears and 1 for tail.
Then p = q = ½ and μ(mean) = p = ½ &
 2 (V ( X ))  pq  1 / 4

36
 Uniform Distribution

f(x)
Let variable X has a uniform distribution
between a and b X ≈ U(a,b)
f(x) = 1/(b-a) a≤x≤b a b
= 0 otherwise
•Rectangular distribution
•The interval [a,b] is called the range of F(x)
the distribution 1
•CDF is given by:
x
F(X )   f (t )dt  0 xa
 0
xa a
 a xb b
ba
1 xb
ab 2(b  a ) 2
Mean    Var .   
2 12

Prove the above mean and

variance

37
Poisson Distribution
It is a discrete distribution.

( t ) k e  t
P( X = k ) = PX ( k ) 
k!


k e 

k 0 k!
1

Mean λ Variance = λ

38
 Exponential Distribution

y
1 
e

f x ( y) 

1


fy(x)

0
y

39
  y y 
1 
 x
f ( y ) dy  0  e 
dy   e 
1
0

1 1
  E( X )  ,  2  V (X ) 
 2


CDF  Fx (b)  f

x ( y )dy

 0 b 0

 y b
 1 
0  e dy  1  e 
b 0

40
 Normal Distribution

( y  )2
1 2 2
f x ( y)  e    y
2 
it is called a N  (  ,  ) i., e., it has two parameters.

fy(x)

x
μ
∞- ∞

41
  ( y   )2
1
Fx (b)  
 2 
e 2 2
dy

This integral does not exist in closed form. But there

exist tables of values for the standard normal distribution
with mean=0 and var.=1 which is given by:

z2
1
f ( z)  e 2
(   0 ,  1)
2

If X is N  (  ,  ) then the random var iable Z is

Z  ( X   ) /  is a N  (0,1)
 X    Z i., e., X is N  (  ,  )

42
Variance Reduction Techniques
1 Increase sample size N where
V = E(x2) – (E(X))2 V  1/N

Examples
use f(x) = e –x it is exponential distribution with mean
λ=1 i.,e.,F(X)= 1 – e –x

i R.N. Xi = - ln ( 1 - ri )
1 0.495 0.684
2 0.335 0.408
3 0.791 1.568
4 0.469 0.633
5 0.279 0.326
6 0.698 1.199
7 0.013 0.014
8 0.761 1.433
9 0.290 0.343
10 0.693 1.183

Total = 7.793 so mean = 7.793/10 = 0.779

with error = ((1-0.779)/1)*100 = 22.1%
43
2-Use Complement Random Numbers for the
above example f(x) = e –x

i ri xi=-ln(1-ri) ŕi=1-ri xi=-ln(1- ŕi)

1 0.495 0.684 0.505 0.702
2 0.335 0.408 0.665 1.092
3 0.791 1.568 0.209 0.234
4 0.469 0.633 0.531 0.756
5 0.279 0.326 0.721 1.275
6 0.698 1.199 0.302 0.359
7 0.013 0.014 0.987 4.305
8 0.761 1.433 0.239 0.272
9 0.290 0.343 0.710 1.236
10 0.693 1.183 0.307 0.366
Total = 7.793 =10.597

Estimated mean = 0.5(0.7793+1.0597)

= 0.92
With Error = ((1-0,92)/1)*100 = 8.05%
44
Example:
determine the mean of the random variable with
F(X)= x2 0≤x≤1
f(x) = 2x, 0≤x≤1
1 1
x   x f ( x)dx   2 x 2 dx  2 / 3
0 0

F(X) = X2 = r So Xi = + ri0.5

Ri Xi = ri0.5

0.628 0.792
0.207 0.455
0.662 0.814
0.774 0.880
0.360 0.600
0.951 0.975
TOTAL= 4.516 Estimated mean = 4.516/6 = 0.753 with
error = 13%

45
Using Complement Random Variables:

ri ŕ=1-ri xi= ri , (1  ri )
0.628 0.792
0.372 0.61
0.207 0.455
0.793 0.891
0.662 0.814
0.338 0.581

Total = 2.061 + 2.082 = 4.143

So the mean = 4.143/6 = 0.691
with Error = ((2/3-0.691)/(2/3))*100 = 4%

46
 Introduction to Queuing Theory

A queuing system is a system in which customers arrive,

wait if that service is not immediately available,
receive the necessary service, and then depart.

Customers Service Customers

arriving facility departing

Characteristics of any queuing system

(A) The arrival pattern (distribution of arriving
customers)
(B) The service process (distribution of service time)
(C) The number of parallel servers
(D) The system capacity
(E) The queuing discipline (FIFO, LIFO, priority)
So any queuing system can be described as A/B/C/D/E

47
:The items of interest concerning any queuing model

• 1. Queuing length (Lq)

• 2. Time in the system (W)
• 3. Idle and busy time of a server
• M/M/1/∞/FIFO system
• Derivation using stochastic balance

λpo λp1 λp2

S
S
o
S
1
S
2 32

μp1 μp2 μp3

• Using balance

• at point So μp1=λpo p1= λ/μpo

• at point S1 (λ+ μ)p1=λpo+ μp2

p2= (λ/μ)2po

• and so on we get pn = (λ/μ)npo

48
  
 n

n 0
p n  1   ( ) p0
n0 

1 1
 p0     
 n

n 0
(

) 
n 0
 n

1
 p0   1 
1
1 
 pn  (1   )  n for n  0,1,2,3,...

L   1,
1 
2
Lq   1
1 

49
• Expected number in the system (L)

• L = E(X)
 
  xp n   x (1   )  x
x 0 x 0

L  (1   ) x x
x 0

 (1   )   x x 1

d  x
 (1   )  
d x 0
d 1
 (1   ) 
d 1  
1
 (1   ) 
(1   ) 2
 
L     1 i., e.,   
1   

50
• Expected Number in the queue Lq


Lq  op0   ( x  1) p x
x 1
 
  xp x   p x
x 1 x 1

L  (1  p0 )
 L  1  p0

 11 
1 
2 2
 Lq     1, i.e.,   
1    (   )

51
Using Little’s formula

The Expected time in the system (W)

L
W

1
 
 

The Expected time in the Queue (wq)

1
Wq  W 


  
 (   )

52
Random Number Generation (Pseudo R.N.)

There are many algorithms for random number

generations, but we are going to emphasize on the most
common one which is known as a Congruential method.

Ri+1 = ( a Ri + c ) modulo T

where

Ri+1 is the new uniform random number,

Ri is the previous uniform random number

a, c, and T are arbitrary constants

53
Example
For a = 1, c = 4, T = 5, R0= 2(seed); generate 10 random
numbers.

Solution

R0 = 2

R1= Rem[ {(1)(2) + 4}/5 ] = 1

R2= Rem[ {(1)(1) + 4}/5 ] = 0

R3= Rem[ {(1)(0) + 4}/5 ] = 4

R4= Rem[ {(1)(4) + 4}/5 ] = 3

R5= Rem[ {(1)(3) + 4}/5 ] = 2

R6= Rem[ {(1)(2) + 4}/5 ] = 1

R7= Rem[ {(1)(1) + 4}/5 ] = 0

R8= Rem[ {(1)(0) + 4}/5 ] = 4

R9= Rem[ {(1)(4) + 4}/5 ] = 3

R10= Rem[ {(1)(3) + 4}/5 ] = 2

54
Generation of a Random Observation
Following a certain distribution
1- For a Simple Discrete Distribution we use the
allocation method, see the previous examples.
2- For continuous Distribution we use the inverse method
if and only if the distribution is reversible, see the
following examples.
Example 1
Generate a random observation from the uniform
distribution between a and b, i;, e.,U(a,b)
1 f(x)
f ( x) 
ba
xa
F ( x) 
ba a
x
U (0,1) U ( a, b)
xa
R  F ( x)  F(x)
ba 1
 x  a  R (b  a ) R1
i., e.x  F 1
( R) R2
R3
0
x1 x2 x3
a b x
55
 Most important Example:
Generate Random Observations from Exponential
distribution with mean μ  x
f ( x)   e
F ( x )  1  e  x
R  F ( x)  1  e  x
ln(1  R )
x 

ln(1  R1 )
so x1 

ln(1  R2 )
x2 
f(x) 
.....
where R1 , R2 ,..., are U (0,1)

x
1
F(x)
R1
R2
R3

0
x1 x2 x3 x
56
Illustrative Example:

Sales of life insurance problem:

From the previous sales history, each time a house call is

made:

50% not interested in purchasing,

50% willing for further discussions.

After discussions ½ of time the visit will result in “No

Sale”, 1/3 of time will result in sale of 10,000 L.E. policy,
1/6 of time will result in sale of 20,000 L.E. policy. Using
a simulated sample of 20 visits, determine the probability
that a sale is made and the expected policy value for
each policy, assuming a sale is made.

57
Insurance Example

50% not interest 50% for further discussion

½ of time no sale Given sale

time(10000)

time(20000)
1/3 of

1/6 of
Probability of No sale=0.75

Prob. of sale=0.25

Theoretical Solution:
Probability of No Sale = 0.75
Probability of Sale = 0.25
Pro(10000 policy | sale) =2/3 = 0.667
Pro(20000 policy | sale) =1/3 = 0.333
Expected value = 0.667x10000 + 0.333x20000
= 13333 L.E.
58
Simulation of 20 insurance calls
Trial Flip Interest Not Roll 0 10 20 Value
Interest
1 H X 4 X 10,000
2 H X 4 X 10,000
3 T X
4 H X 1 X
5 H X 2 X
6 T X
7 H X 3 X
8 H X 6 X 20,000
9 T X
10 T X
11 H X 2 X
12 T X
13 H X 3 X
14 T X
15 H X 6 X 20,000
16 H X 4 X 10,000
17 H X 2 X
18 H X 4 X 10,000
19 T X
20 H X 6 X 20,000

59
From the table, 13 of the visits will result in a “No sale”.
P (No sale) = 13/20 = 0.65
Given Sale, we have 4x10,000 and 3x20,000 policies
soled so
P(10000) = 4/7 = 0.5714
P(20000) = 3/7 = 0.4286
The Expected Vale = 0.5714 x 10000 + 0.4286 x 20000
= 14,286 L.E.

Comparison between Theoretical & Simulation

(Validation):

Prob.of Prob.of Prob.of Prob.of Expected

No Sale Sale 10,000 20,000 Value

Simulation 0.65 0.35 0.5714 0.4286 14,285

Theoretical 0.75 0.25 0.667 0.3333 13,333

60
Time Management Methods
There are two time references used in simulation:
1) Simulation Time: The period of time simulated by
the model – 30min., 5years, 100 years whatever
interval the researcher is interested in. This time is
reset to zero at the beginning of the simulation run
and acts as a counter to the number of simulation
time units.
2) Run time: The time that the computer takes to
simulate the period of interest.

We have two ways for time changes

(periodic &Event Scan):
i) The Periodic Scan or Fixed time increment
technique: In this technique the simulation clock is
adjusted by one predetermined uniform unit and
then examines the system to determine whether any
events occurred during that interval. If any occurred,
the event or events are simulated; otherwise no
action is taken.

61
The simulation clock is then advanced another unit,
and the process is repeated. In this technique, the
exact time of the occurrence of particular event is
largely ignored. All events that occur during a given
interval are treated as if these events occurred at the
end of that interval.
Example :
Consider the simulation of a single-bay-
service station for a period of ten minutes. Assume
that the system is empty at the beginning and at the
end of the period. Assume also that four customers
are serviced, their arrivals occur at simulation times
1.8, 3.2, 6.1 and 7.4, and that service completions
occur at simulation times 2.6, 4.8, 7.3 and 8.1.
If a time interval of one minute is used, arrival events
are denoted by A1, A2, A3 and A4. While completion
events are denoted by C1, C2, C3 and C4(see the
following figure.).

62
 A1 C1 A2 C2 A3 C3 A4 C4
1.8 2.6 3.2 4.8 6.1 7.3 7.4 8.1

S0 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10
C3 and A4 are both considered to have occurred at clock
time 8.0; although C3 actually occurred before A4.

The problem of periodic scan is that events

separated in time appear to occur simultaneously. In this
example at time period S7 – S8 the system appears to be
idle while actually it is not.
• The big problem is to determine the length of the
interval to be used.
• The time increment used in the periodic scan
approach should be as small as possible to minimize
the possibility of lost information. But the smaller
the time unit, the longer the number of calculation
necessary to complete the simulation.

63
Thus the actual run time of the simulation and hence
the cost of simulation is increased. So there is a trade-
off between the precision and maximum run time.
ii ( The Event Scan
In this technique the clock is advanced by
the amount necessary to trigger the occurrence of the
next event. Thus the time advance intervals are of
variable lengths.
While some information can be lost in the
periodic scan, the event scan avoids this problem;
because the clock is advanced only to the next
occurrence time.
In periodic scan, the exact occurrence time
of events is not known, because each event is treated
as if occurred at the end of the time interval in which
it occurred.. This problem is overcome in Event Scan
approach.

64
Example
A bank of machines in a manufacturing shop breaks down according to the
following inter arrival time distribution. The time it takes one repair person
to complete the repair of a machine is given in the service time distribution.
Inter arrival P(X) Mapping Service P(X) Mapping
time (hours) of R.N. Time (hours) of R.N.
0.5 0.30 00 – 29 0.5 0.25 00 – 24
1.0 0.22 30 – 51 1.0 0.20 25 – 44
1.5 0.16 52 – 67 2.0 0.25 45 – 69
2.0 0.10 68 – 77 3.0 0.15 70 – 84
3.0 0.14 78 – 91 4.0 0.10 85 – 94
4.0 0.08 92 – 99 5.0 0.05 95 – 99
1.00 1.00

Simulate the breakdown of 5 machines. Calculate the average machine

downtime using two repair persons and the following random numbers
sequence. Both repair persons cannot work on the same machine.
R.N.: 30, 81, 02, 91, 51, 08, 28, 44, 86, 84, 29, 08, 37, 34, 99.

65
Solution
RN Interarrival Time of RN Service Repair Man1 Repair Man2 Waiting
Down
Time Arrival Time On Off On Off Time Time

+
30 1.0 1.0 81 3 1 4 0 3
+
02 0.5 1.5 91 4 1.5 5.5 0 4
+
51 1.0 2.5 08 0.5 4 4.5 1.5 2
+
28 0.5 3.0 44 1 4.5 5.5 1.5 2.5
86 3.0 6.0 84 3 6 9 0 3

Total Down Time 14.5

Average Down Time = 14.5 / 5 = 2.9 hours.

66
Example for a continuous distribution:
Simulate a system M/M/1 with arrival rate λ = 20/h
and service rate μ = 25/h. Assuming system
empty at time zero.
•Using time scan (periodic scan)
•Event scan
Solution:

1. Using Time Scan :

P (arrival in one minute period) = 1 – e-λt
= 1 – e-1/3(1)
= 0.281
P (departure in one minute period) = 1 – e -5/12(1)
= 0.341
So let 0 ≤ R.N. ≤ 0.281 Arrival,
0 ≤ R.N. ≤ 0.341 Departure.

67
 Time Customer Random number Arrival Departure
in min. In system Arrival Departure Yes/No Yes/No
0 0 0.096 - Yes -
1 1 0.569 0.665 No No
2 1 0.764 0.842 No No
3 1 0.492 0.224 No Yes
4 0 0.950 - No -
5 0 0.610 - No -
6 0 0.145 - Yes -
7 1 0.484 0.552 No No
8 1 0.350 0.59 No No
9 1 0.450 0.410 No Yes
10 0 0.802 - No -

68
2. Using Event Scan (Event Driven):
The time until next arrival is given by

 ln(1  R ) 1
X arr .    min .
 3
  3 ln(1  R )

The time until next departure is given by

 12 ln(1  R ) 5
X dep.   min .
5 12

69
Time in Event Customer Random Number Min. until next Next
min. In system Arrival Departure Arr. Dep. Event
0 0 0.096 - 0.303 - A
0.303 A 1 0.569 0.665 2.525 2.625 A
2.828 A 2 0.764 - 4.332 0.1 C
2.928 C 1 - 0.842 4.232 4.428 A
7.160 A 2 0.492 - 2.032 0.196 C
7.356 C 1 - 0.224 1.836 0.609 C

70
 Example for discrete distribution:
Two-stage assembly line are producing a large product. There
is no room in between the two line ; so the output of the first
line is fed to the second line directly else the first line must
wait until second line to be free. The service distribution for
both line are given below:
Service Time Freq. R.N. Time Freq. R.N
seconds for 1st line Allocation for 2nd line Allocation
10 4 00 – 03 4 00 - 07

20 6 04 - 09 5 08 - 17
30 10 10 – 19 6 18 - 29
40 20 20 – 39 7 30 - 43
50 40 40 – 79 10 44 - 63
60 11 80 – 90 8 64 - 79
70 5 91 – 95 6 80 - 91
80 4 96 - 99 450 92 - 99
100

71
Item First Line Waiting Second Line
Time
No. R.N. Start Service Finish R.N. Start Service Finish
Time Time Time Time Time Time
1 56 00 50 50 83 50 70 120

2 55 50 50 100 20 47 120 50 170

3 84 120 60 180 08 180 20 200
4 36 180 40 220 05 220 10 230
5 26 220 40 260 42 260 40 300
6 95 260 70 330 95 330 80 410
7 66 330 50 380 30 17 410 20 430
8 03 410 10 420 10 21 430 30 460
9 57 430 50 480 31 480 40 520
10 69 480 50 530 90 530 70 600
470 60
•The output rate average = 600/10 = 60 seconds
•Utilization of 1st line = 470/530 = 88.7%
•Utilization of 1nd line = 430/550 = 78.2%
72