Cooperative Evolutionary Framework With Focused Search for Many-Objective Optimization

Abstract

When dealing with many-objective optimization problems, Pareto-based approaches suffer from the loss of selection pressure toward Pareto front. In this study, a general cooperative evolutionary framework with focused search is proposed to make Pareto-based approaches perform better for many-objective optimization problems. The proposed framework has two evolutionary populations, a focused evolutionary population and a Pareto-based evolutionary population, and these two populations work collaboratively. The focused evolutionary population focuses on searching for the corner solutions that are important for convergence and spread (focused search), guiding the Pareto-based evolutionary population to evolve toward the Pareto front, and promoting Pareto-based evolutionary population to spread along the Pareto front. Pareto-based evolutionary population aims to obtain the solutions with well convergence and diversity (global search), providing some undeveloped but potentially promising solutions to focused evolutionary population. As a general framework, any Pareto-based approaches can be adapted to the proposed framework. As a case study, four representative Pareto-based approaches are selected to instantiate the framework. Experimental results show that Pareto-based algorithms with a focused evolutionary population can be appropriate for many-objective optimization problems, and thus the proposed framework paves a new way to improve the performance of Pareto-based approaches for many-objective optimization problems.