An Empirical Investigation of Optimum-Seeking in the Computer Simulation Situation

Abstract

When computer simulation is being used to investigate a problem, any attempt at obtaining an optimum solution usually involves a search of the relevant factor space. Because a number of search techniques are available, it is desirable to have some criteria for selecting the technique to be used in a specific simulation situation. However, no mathematical basis exists for an evaluation of these techniques relative to characteristics that may vary from simulation to simulation. Thus, it is necessary to perform empirical studies in order to obtain the required information. This paper discusses the framework and results of an empirical evaluation of the performance of some search techniques. It was based on applying the techniques to various multidimensional situations that were constructed so that the true optimum was known, providing comparison of the performance of each technique with the true optimum, as well as with the performance of other techniques. Each of the search techniques was evaluated for a variety of simulation characteristics in the situation where the available number of computer runs was relatively small compared with the number of factors being examined.