| The goals of computer experiments have to do with determining information about a real world process (for which one may have limited real data) by means of a simulator with complex code. Since this code is often expensive to run, the experimenter needs to find ways to get the most information possible from each data point that is evaluated. The objectives of computer experiments considered in this document are the optimization process for finding the optima, characterization of optima, selection of optima, and accurate global sensitivity estimation. Efficient methods are developed and demonstrated for estimating accurate global sensitivities and characterizing optima. A utility function has been developed for selecting optima. Three methods of optimization have been developed and demonstrated: 1) Optimization with the goal of finding the best optimum; 2) Optimization with the goals of finding and characterizing the significant optima of the simulator surface; and, 3) Optimization with the goals of computing global sensitivities, finding and characterizing the significant optima. These methods are then applied to a real world example: Selecting the optimum pump locations and pumping rates for six pumps that minimizes the cost of operation of a system which remediates pollution of the Yellowstone River. |
