Variance Reduction Techniques (VRTs) are methods of reducing the variance of a simulation output random variable (e.g., the average delay in a queue) so that the number of replications can safely be reduced, the confidence interval for that output performance metric narrowed, and/or the level of confidence for such a confidence interval increased. One very useful VRT is Common Random Numbers (CRN). By using the same random numbers for the same purposes (e.g., the 7th draw from random number stream #12 should be used to calculate the 4th downtime duration of machine X in all scenarios), CRN tests all scenarios under the same conceptual experimental conditions. An important step in using CRN is to synchronize usage of random number streams. WITNESS, ProModel, and ARENA make this synchronization easy to achieve: the last parameter in calls to standard random distributions (like the triangular) specifies which random number stream to draw from. This parameter is optional in ProModel and ARENA (but should be used) and required in WITNESS. So, for example, stream 11 can be dedicated to cycle times at machine X, stream 12 to downtime durations at machine Y, stream 13 to intervals between downtimes at machine Y, etc.
__________________
E. Williams, PMC
|