The reason many evacuation models produce different results for repeating exactly the same scenario is because random numbers are used somewhere in the model. This is typically done when either a process is not well understood e.g. conflict resolution when 2 agents compete for the same location, or , gaining a value from a probability distribution e.g. response time range. As such, the number of runs needed will be dependent on the extent these occur and the size/types of distributions used. A sufficient number of runs should reflect a 'suitable' spread of these random numbers about the distributions and the extent that they occur.
One way to give an indication of the level of variability between runs would be to use the change in total evacuation time (TET) between runs. The larger the differences between TET in each run the more that are needed. You could state that enough runs have been performed when the last X runs vary from the average TET by no more than Y%. This method can have problems but it is one way.
I think choosing the number of simulation runs is a bit of a black art and many people just run many simulations to ensure they are representative.
Hope this helps,
Last edited by Michael Kinsey (14 Apr 2012 - 13:30)