A Trust Region Method for Pareto Front Approximation

TL;DR

This paper introduces a trust region-based algorithm for approximating Pareto fronts in black-box multiobjective optimization, effectively maintaining solution distribution uniformity even with more than two objectives.

Contribution

It proposes a novel trust region method that uses density-based reference points to ensure uniform Pareto front approximation in multiobjective black-box problems.

Findings

Algorithm generates well-distributed Pareto solutions

Convergence to Pareto critical points is proven

Effective for problems with three objectives

Abstract

In this paper, we consider black-box multiobjective optimization problems in which all objective functions are not given analytically. In multiobjective optimization, it is important to produce a set of uniformly distributed discrete solutions over the Pareto front to build a good approximation. In black-box biobjective optimization, one can evaluate distances between solutions with the ordering property of the Pareto front. These distances allow to be able to evaluate distribution of all solution points, so it is not difficult to maintain uniformity of solutions distribution. However, problems with more than two objectives do not have ordering property, so it is noted that these problems require other techniques to measure the coverage and maintain uniformity of solutions distribution. In this paper, we propose an algorithm based on a trust region method for the Pareto front approximation in black-box multiobjective optimization. In the algorithm, we select a reference point in the set of non-dominated points by employing the density function and explore area around this point. This ensures uniformity of solutions distribution even for problems with more than two objective functions. We also prove that the iteration points of the algorithm converge to Pareto critical points. Finally, we present numerical results suggesting that the algorithm generates the set of well-distributed solutions approximating the Pareto front, even in the case for the problems with three objective functions.