Improving constraint handling for multiobjective particle swarm optimization

In this paper, a novel particle swarm algorithm for solving constrained multiobjective optimization problems is proposed. The new algorithm is able to utilize valuable information from the infeasible region by intentionally keeping a set of infeasible solutions in each iteration. To enhance the diversity of these preserved infeasible solutions, a modified version of adaptive grid is introduced. In addition, a voting mechanism is designed to balance the preference of infeasible solutions with smaller constraint violation and the exploration of the infeasible region. The effectiveness of the proposed method is validated by simulations on several commonly used benchmark problems. By using the hypervolume indicator, it is shown that the proposed algorithm is more powerful than two other state-of-the-art algorithms.