Diversity preference-based many-objective particle swarm optimization using reference-lines-based framework

Abstract A simple yet effective diversity preference approach is developed and coupled with many-objective particle swarm optimization (PSO) for solving box-constraint optimization problems. Since the selection of global leaders in PSO is crucial for many-objective optimization, an approach is developed using the reference-lines-based framework to update these leaders. In the proposed approach, diversity is ensured first by making clusters of solutions for every reference line. Thereafter, only one solution from each cluster with minimum penalty-based boundary intersection (PBI) fitness is selected. In case any cluster of a line is empty, a solution with minimum PBI fitness with respect to the line is selected. This ensures the selection of isolated but sometimes dominated solutions. The proposed approach is then used for updating the archive of global leaders and for assigning them to particles in the swarm. The proposed algorithm, which is referred to as MaOPSO-DP, is tested on 3-, 5-, 8-, 10-, and 15-objective instances of DTLZ and WFG test problems. Results demonstrate the effectiveness of MaOPSO-DP over eight many-objective evolutionary and PSO algorithms from the literature.