Diversity based selection for many-objective evolutionary optimisation problems with constraints

Abstract

The paper introduces a novel many-objective evolutionary method, with a diversity-based selection operator and aims to fill the “gaps” in the Pareto Front approximation and to increase its spread. It shows, that guiding the evolution process towards the least explored parts of a space increases overall diversity, but can also lead to increased convergence. A set of experiments is carried out on a many-objective Multi-Skill Resource-Constrained Project Scheduling Problem and a bi-objective Travelling Thief Problem. Both of those problems comprise two interwoven subproblems. Separate optimal solutions to the subproblems do not guarantee the optimal solution to the entire problem. Moreover, it shows that the existing diversity mechanism does not work well in combinatorial spaces, where the domain of each objective has a different size. The results indicate that a novel selection operator circumvents this issue. A set of Quality Measures is used to describe the desirable features of Pareto Front approximation: convergence, uniformity, and spread. The experiments are followed by a set of visualizations. Along with a set of Quality Measures, they give an insight into the characteristics of the method. The paper is concluded by a thorough theoretical analysis and possible directions for future work.