This preview shows pages 1–3. Sign up to view the full content.
MIE360 Computer Modeling and Simulation
Lecture Notes
Lecture 10– Make Pilot Runs
Yes
Determine # of
replications
No
Determine the
run lengths
Terminating
Simulation
Determine the
Warmup period
No
Include
Transient
Yes
Determine # of
replications
The WarmUp period
In many real life situations we rarely start the system from scratch. When plants close, its status is
frozen as is until it opens up again the next day. Or it runs 24/7 shifts.
When we simulate such systems in a computer model, there is an initial period which resembles the
operation of the plant when it first started many years ago.
In order to ensure that the initial simulation period is representative of the steady state operation of a
system, we have to wait some time until the simulation itself reaches this steady state. Until it
reaches steadystate the results have to be ignored. We can only start tracking the states and counters
once this stead state period is reached.
Computer simulation languages always allow a user to specify a warmup period.
For now we delay dealing with this issue, by assuming that we wish to include the transient.
Terminating vs. Nonterminating Simulations
A terminating simulation is one that has a natural end point, such as a closing time or natural cycle.
A nonterminating simulation is one that has no such natural endpoint.
Clearly the problem of the run length only exists for nonterminating simulations  for terminating
simulation the run length is decided upon by the user or situation.
While there are very sophisticated methods for determining the run lengths for nonterminating
simulations, these are beyond the scope of introductory simulation courses, and common practice.
The standard practice in simulation is for the analyst to negotiate a runlength with the client that is
considered “appropriate” and thus to negotiate the client into using a terminating simulation.
Daniel Frances © 2010
1
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document MIE360 Computer Modeling and Simulation
Lecture Notes
Using Replications to derive confidence intervals
Consider simulation results in the table above
i
123456789
1
0
1
1
1
2
1
3A
v
g
X
i
143
7
8
41
27
5
9
81
1
7
6
.
6
2
What about reporting a confidence interval for the average of 6.62? Perhaps we can calculate a
sample standard deviation for these observations, and then create a confidence interval around the
6.62 average!
DEFINITELY NO!
Outputs of simulation models are notoriously autocorrelated! Large readings tend to come in
bunches, and so do small readings. It is the same reason why waiting lines randomly grow, stay long
for a while, and then randomly return to shorter lines. The theory behind confidence intervals
requires the data to be IID, and assumes no autocorrelation. With this assumption blatantly violated
reporting confidence intervals based on a stream of simulation outputs would be totally invalid!
This is the end of the preview. Sign up
to
access the rest of the document.
This note was uploaded on 09/20/2011 for the course MIE 360 taught by Professor D.frances during the Fall '10 term at University of Toronto Toronto.
 Fall '10
 D.Frances

Click to edit the document details