A Probabilistic Solution Generator of Good Enough Designs for Simulation

Abstract

We build a probabilistic solution generator using the learning automata theory, which can generate a small set of "good enough" designs with a predetermined high probability. The main goal of our work is to reduce a large design population to a much smaller subset of good designs that can be analyzed thoroughly in a subsequent simulation study to identify the best design among them. In the process of building the solution generator, a rough-cut design evaluation method with a high noise error is employed in order to screen designs very rapidly _ may it be an approximate method, a heuristic approach, or short simulation runs. The solution generator has been applied successfully to several serious test problems with noisy objectives.