April 23, 2018 at 1:55 pm #3509
In learning about simulation in one of my classes, we briefly talked about how having a warm-up period can increase a simulation’s accuracy in determining true means for waiting time and average queue length. I have a few questions:
April 23, 2018 at 3:51 pm #3516
- Is there certain cases where it would be better not to factor in a warm-up period
- What is the most common way to determine a warm-up period? There seem to be conflicting views online. For reference. I have been using one full simulation replication time as my warm-up period. (e.g. if the replications are for 700 hours my replication with a warm-up is 1,400 with the first 700 hrs removed)
- Would making the warm-up period to large “hurt” your study?
I’m interested to hear the answers to 2 and 3 of this as I am also curious what people have found to be the most effective ways to determine an optimal or at least adequate warm-up period. I think in regards to your first question, warm-up periods are usually used for instances when we are assuming stead state for a process.April 23, 2018 at 4:18 pm #3518
Hi Veronica, those are some good questions.
To address the first question, I would first say that it’s best to separate the warm-up and stable state and analyze them separately. There are some situations where you would not want to factor in a warm-up with the rest of the simulation because the system does not behave as it would if it were in stable state. Some examples are in airport check-in area or a manufacturing facility. For manufacturing, there may be some interesting patterns in machine/employee learning behaviors in the warm-up.
The second question, I thought I remember using an equation to determine warm-up period, but an alternative could be to use the data where variance and mean “settle down”.
Regarding the third question, if the warm-up period is too large, then it may misrepresent data on the actual warm-up and steady state periods.
Finally, I think some simulation softwares (e.g. simul8, ARENA) have some useful guidelines and tools on how to choose a warm-up period.
I look forward to seeing the responses from others.April 23, 2018 at 4:44 pm #3524
For question 1, I would say yes, in some certain cases where warm-up period are not necessary. For example, when you are simulating a shop that starts and ends the day empty. However, warm up period is required in a case where you are simulating a week in a factory.
For question 3, I would not say large warm up period will end up hurting the studies. From my experiences, different software have different warm up setups, I used to use Simio rather than Arena for setting the warm up period, and I would say it depends on which studies you are conducting. For instance, we used Simio to set up simulation warm up period to monitor a music festival (Floydfest, we want to monitor the average waiting time for customers who were transported from the parking lot to the actual music festival site through adding additional buses and vans during the peak time by analyzing the number of tickets sold at each day. In this case, we need to make the warm up period large enough so that we could collect the data that we need because the show did not last for 24 hours.
You must be logged in to reply to this topic.