Sequential Screening Methods for Discrete-Event Simulation
by
Hong Wan
School of IE, Purdue University

Factor screening is performed to eliminate unimportant factors so that the remaining important factors can be more thoroughly studied in later experiments. In simulations of complex systems, there are often a large number of input factors that users can vary in computational experiments. Consequently, an efficient effective screening design will be critical for prompt analysis and strategy making. This talk will introduce several sequential screening methods with specific error control for discrete-event simulation experiments and compare their performance in different scenarios. Based on these methods, we propose a hybrid framework integrating several designs for screening purposes. The algorithm is “smart” enough to dynamically learn the system and determine what the optimal design is for each factor at each step. While the sequential nature of the design of experiments has been promoted for a long time, the on-time selection and integration of different methods have rarely been discussed and heavily relies on human intervention at each stage. The research discussed automates this selection and integration process.

Date received: February 27, 2007