SEPARATING THE ART AND SCIENCE OF SIMULATION OPTIMIZATION - A KNOWLEDGE-BASED ARCHITECTURE PROVIDING FOR MACHINE LEARNING

The purpose of this paper is to develop an architecture for simulation optimization, building on the work we published in this journal in 1985. The need for a dramatically updated architecture is established by examining the simulation optimization process, traditional approaches to die problem, and difficulties inherent with these methodologies. Our framework directly addresses these problems by exploiting concepts and technologies introduced that have become popular in the last ten years, such as expert systems, to capture heuristic opinions and experience, and neural networks, to introduce machine learning. Three main contributions result from this research. First, the simulation optimization process is examined from a completely new perspective-a strategic overview of the process leads to an unbundling of the ''art'' and ''science'' elements that are co-mingled in current practice, thereby promoting modularity and making the other two contributions possible. Second, the newly unbundled process is translated to a knowledge-based context facilitating the direct inclusion of human expertise. The third contribution of this paper is the incorporation of machine learning into the framework thus permitting the optimizer to teach itself from its experiences.