An Optimization Method for Stochastic Systems with Multiple Objectives

Abstract

This paper presents an algorithm for the optimization of multi-objective problems with or without a closed form representation and with or without stochastic responses. The algorithm combines direct search methods, single-objective formulations, and a method to locate missing Pareto solutions in a Pareto approximation, to extend the BiMADS approach beyond two objectives. The algorithm is designed to find a reasonably complete, yet efficient approximation of the Pareto set, and is evaluated on systems with up to eight objectives. A discussion of parameter settings, limitations, and improvements is also included.