Pressure point driven evolutionary algorithm for many-objective optimization

Abstract

Abstract How to form good trade-offs between convergence and diversity for many-objective optimization is an ongoing challenge. With the increase in objectives, the ratio of non-dominated solutions in a population increases sharply, which challenges individual discrimination and selection pressures. Besides, the Pareto fronts of many-objective optimization problems (MaOPs) have various shapes, such as disconnected, degenerate, biased deceptive, and mixed shapes, which further challenge the trade-offs between convergence and diversity. To address the above issues, we propose a pressure point driven Evolutionary Algorithm (proEA). Specifically, a pressure point based strategy is developed to update the pressure point, such strengthening the selection pressure. Then, the reference vector based environmental selection strategy is improved by integrating an angle-based selection strategy to account for the complicated shapes of Pareto fronts. Finally, a series of numerical experiments are conducted to compare the proposed proEA with five representative algorithms in the context of 36 test instances with 5, 7, 10, and 15 objectives. The experimental results demonstrate the superiority of algorithm proEA.