Sequential Design of Computer Experiments for Multiobjective Optimization

• Activity Number: 522
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Joshua Svenson*+ and Thomas J. Santner+
• Companies: The Ohio State University and The Ohio State University
• Keywords: Computer Experiments ; Pareto Optimization ; Expected Improvement ; Pareto Front ; Nondominated Sets

Abstract

Sequential designs for computer experiments have been used for a variety of goals, such as global optimization, improving model fit, and contour estimation. Here, we focus on sequential designs for computer experiments with multiple outputs, where the goal is to approximate the set of nondominated points in objective space, commonly referred to as the Pareto front. New methods for approximating the Pareto front, which combine nondominated set comparison methods from the multiobjective evolutionary algorithm literature with expected improvement criteria from the statistics and optimization literature, are introduced. The properties of these new methods are discussed, and several examples are included to show how these new methods perform.