Abstract
An automated approach to optimizing multiple-response simulation experiments is presented. AMOS (Automated Multiple-Response Optimum-Seeker for Simulations) is an integrated system of Prolog and Fortran programs, implemented on a microcomputer, which seeks to automatically optimize a weighted linear combination of response variables produced by a constrained number of simulation runs. The available simulation runs are allocated to "stages" of the experiment. Within each stage, a modified Response Surface Methodology (RSM) is used to maximize a "created" objective function over a bounded search region. The size of the search region is reduced from stage to stage, until available simulation runs are exhausted, or desired convergence is achieved. AMOS automatically performs all the steps of the modified RSM. It generates the experimental designs and executes the simulations to produce the response variable observations. Each response variable is modeled and tested for "fit" using Least Squares Regression and Analysis of Variance. Failure of any one response model to "fit" causes AMOS to design and execute the next higher resolution experiment, up to a central composite design. The fitted models are "weighted" and summed to form the created objective function. The weights are normalized values derived from user-specified priorities. The created objective function is then maximized, over the current search region, using Golden Section Search or Canonical Analysis, as appropriate to the order of the polynomial model. The "optimal" point in any stage becomes the starting point for the REM in the next stage. AMOS executes on any standard IBM-compatible microcomputer, with at least 640k bytes of memory and the MS-DOS operating system. Three examples illustrate the AMOS approach.
Matwiczak, Kenneth Matthew (1990). AMOS, Automated Multiple-Response Optimum-Seeker for Simulations. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -1117210.