An efficient procedure for locating the optimal simular response
Proceedings of the fourth annual conference on Applications of simulation
Literature review bibliography of simulation optimitation
WSC '77 Proceedings of the 9th conference on Winter simulation - Volume 1
A gradient—regression search procedure for simulation experimentation
WSC '74 Proceedings of the 7th conference on Winter simulation - Volume 2
Experimental optimization of statistical simulation
WSC '74 Proceedings of the 7th conference on Winter simulation - Volume 2
| Hi-index | 0.00 |
This paper presents an approach designed to increase the efficiency and utility of search for optima of simulation models. Specifically, spline functions (odd-order polynomials fitted between simulation run outputs that match curvature at the end points) are used to approximate the simulation along suitably chosen directions of search. The splines are used to generate "pseudo-experiments" which enrich the data base formed from actual simulation runs. An overall (grand) function is then fit to this data base, yielding new direction(s) of search for the next iteration. Several characteristics of this technique are examined, including its sensitivity to experimental budget, number of iterations allowed, and size of feasible region.This approach results in not only an estimate of the optimal response from the simulation, but also a response surface estimated over a larger domain useful both for sensitivity analysis and in some cases as an approximate representation of the simulation for use in other modeling efforts. The paper describes an application of the technique to a model of a railroad classification yard. The objective is to find the numbers and sizes of inbound and outbound trains, and the dispatching policy within the yard, which minimize total car delay.