Objective reduction in many-objective optimization problems

Abstract

Many-objective optimization problems (MaOPs) are multi-objective optimization problems which have more than three objectives. MaOPs face significant challenges because of search efficiency, computational cost, decision making, and visualization. Many well-known multi-objective evolutionary algorithms do not scale well with an increasing number of objectives. The objective reduction can alleviate such difficulties. However, most research in objective reduction use non-dominated sorting or Pareto ranking. However, Pareto is effective in problems having less than four objectives. In this research, we use two approaches to objective reduction: random-based and linear coefficient-based. We use the sum of ranks instead of Pareto Ranking. When applied to many-objective problems, the sum of ranks has outperformed many other optimization approaches. We also use the age layered population structure (ALPS). We use ALPS in our approach to remove premature convergence and improve results. The performance of the proposed methods has been studied extensively on the famous benchmark problem DTLZ. The original GA and ALPS outperform the objective reduction algorithms in many test cases of DTLZ. Among all reduction algorithms, a linear coefficient based reduction algorithm provides better performance for some problems in this test suite. Random based reduction is not an appropriate strategy for reducing objectives.