Lakhasly

For each scenario that is to be simulated, decisions need to be made concerning the length of the simulation run, the number of runs (also called replications), and the manner of initialization that is required.The simulation analyst will need to understand the difference between Sue and Bob's performance and how that can be included in the simulation as the service time of the attendant, and then make changes in the simulation to include that as an input parameter and not hardwired into the equations for the attendant.However, if the simulation has tens or hundreds of input parameters, conducting a run for each combination of the parameters can be unmanageable.If there are only a few input parameters, this approach will work.The simulation analysts may also want to use a technique called Monte Carlo analysis (described in a later chapter) that relies on repeated random sampling to compute the results of the simulation.One approach to conducting experiments is to vary the input data for each simulation run and observe the behavior of the system.To illustrate the importance of thinking through experiments during the conceptual model development, what would happen if you were presenting the results of these runs to the client (i.e., gas station management), and they ask you whether the results change if Sue is working as an attendant versus Bob?The changes needed may not be that difficult, but it requires time and resources to change the conceptual model, collect data, make the changes to the simulation, do the verification and validation, and run the experiments again.

For each scenario that is to be simulated, decisions need to be made concerning the
length of the simulation run, the number of runs (also called replications), and the
manner of initialization that is required. One approach to conducting experiments is
to vary the input data for each simulation run and observe the behavior of the system. If there are only a few input parameters, this approach will work. However, if
the simulation has tens or hundreds of input parameters, conducting a run for each
combination of the parameters can be unmanageable. In this case, a formal process
24 M. L. Loper
called design of experiments (DoE) can be used to better define which runs are
needed. DoE is a statistical approach to experimental design (described in a later
chapter) that can help minimize the number of simulation runs needed to understand
the behavior of a system under investigation. The simulation analysts may also want
to use a technique called Monte Carlo analysis (described in a later chapter) that
relies on repeated random sampling to compute the results of the simulation.
Defining the experiments for the gas station model should be considered while
developing the conceptual model. The reason is that the simulation analyst needs to
make sure the right assumptions, behaviors, and data are collected and included in
the simulation in order to support the necessary experiments. A set of experiments
an analyst might run for the gas station management would be to vary the number
of vehicles that come into the station. This could include increasing the number of
passenger cars, light trucks, and vans separately, or increasing them in different
combinations.
To illustrate the importance of thinking through experiments during the conceptual model development, what would happen if you were presenting the results of
these runs to the client (i.e., gas station management), and they ask you whether
the results change if Sue is working as an attendant versus Bob? Since the behavior
of the attendant was simplified and does not include personal attributes of specific
people, the simulation cannot answer that question directly. The simulation analyst
will need to understand the difference between Sue and Bob’s performance and
how that can be included in the simulation as the service time of the attendant, and
then make changes in the simulation to include that as an input parameter and not
hardwired into the equations for the attendant. The changes needed may not be that
difficult, but it requires time and resources to change the conceptual model, collect data, make the changes to the simulation, do the verification and validation,
and run the experiments again. Upon presenting the new results to the gas station
management, they ask if the results are different if there is a football game at the
stadium down the block. If the simulation does not allow traffic to be input hourly
(say it was originally implemented as a probability distributed), then more time and
resources will be needed to make the changes. Defining the sets of experiments
early in the simulation process can prevent many unnecessary hours or reworking
the simulation