Making efficient simulation experiments interactively with a desktop simulation package
- PDF / 264,018 Bytes
- 7 Pages / 595 x 842 pts (A4) Page_size
- 83 Downloads / 183 Views
#2000 Operational Research Society Ltd. All rights reserved. 0160-5682/00 $15.00 www.stockton-press.co.uk/jors
Making ef®cient simulation experiments interactively with a desktop simulation package RCH Cheng and JD Lamb* University of Kent at Canterbury, UK It has been shown how a design of simulation experiments methodology can be used interactively with practical simulation models constructed in a desktop simulation package (SIMUL8). The methodology includes new ideas on how to improve the accuracy of a simulation response. It is implemented as a set of computer program modules that are not speci®c to a particular simulation model and provide an interface that lets the modeller construct an ef®cient simulation experiment with only an operational understanding of how the methodology works. The methodology and program modules are illustrated with a practical simulation model, and the results show how they can improve simulation response with negligible increase in computational effort. Keywords: design of experiments; regression metamodels; simulation
Introduction Discrete-event simulation is now well established as a methodology1,2 There are now many interactive simulation packages such as SIMUL83 and Arena4 that can be run on a personal computer. Banks5 has recently written a review of such packages. These packages typically let the user construct complex simulation models quickly without special programming knowledge. There is also now an established theory of design of simulation experiments.6±9 This theory can be used to improve greatly a simulation study, producing more accurate results for the same computational effort, in practice it is not much used at all. One reason for this is that only recently have personal computers and the packages available for them become powerful enough to run the kind of simulation studies that bene®t from this theory. Another is that the theory has not been made easy to use with interactive simulation packages. In this paper we extend the theory and show how it can be used easily and interactively with a simulation package. In particular, we show that the methods we develop can be used with only an operational understanding of how they work and without a noticeable increase in computation time. A simulation model typically has the following features, (in the next section we give examples of how they are realised.) The simulation has an output, which is a random quantity. It is actually the expected value of this output that we will estimate. We call this expected value the response variable. A design variable is a quantity whose value we *Correspondence: Dr JD Lamb, Lecturer in Operational Research, University of Kent, Canterbury, Kent, UK. E-mail: [email protected]
can set in the simulation model. We wish to estimate the value of the response variable as a function of the design variables over a range of values of the design variables; we call this function the response surface. We will use a regression model to estimate the response surface. We call a set of values of the
Data Loading...
