A flexible, interactive, graphical approach to modeling stochastic input processes

Mary Ann Flanigan, Purdue University

Abstract

One of the main problems in the design and construction of large-scale simulation experiments is the selection of valid input models--that is, probability distributions that accurately represent the behavior of the stochastic input processes driving the system. The problem with using standard distribution families for simulation input modeling is that (a) they frequently do not reflect the true nature of the target input process with sufficient accuracy, (b) they are limited in the number of parameters that are available to control the shape of the distribution, and (c) their parameters are often difficult to estimate. In addition, instances where little data are available force use of intuition, judgment, and subjective information (which are not easily incorporated into the description of standard distributions) to complete the formulation of an input model. In this dissertation, we have developed a flexible, interactive, graphical methodology for modeling the broad diversity of input processes that arise in large-scale simulation studies. We implemented this methodology in a self-contained, Microsoft-Windows-based software system called P scRIME--PRobabilistic Input Modeling Environment. P scRIME integrates graphical and statistical concepts to form a tool that helps an analyst select and visualize an appropriate representation of a random input process using interactive-subjective, data-driven, and visually-based techniques. We have exploited the properties of Bezier curves to develop a flexible univariate and bivariate distribution family that has an open-ended parameterization capable of accurately representing an unlimited number of distributional shapes. Several numerical examples are presented to illustrate the application of this methodology.

Degree

Ph.D.

Advisors

Wilson, Purdue University.

Subject Area

Industrial engineering|Systems design

COinS