Constrained multi-objective optimization via two archives assisted push-pull evolutionary algorithm

Solving constrained multi-objective optimization problems (CMOPs) obtained more and more popularity in the last decade. Various constrained multi-objective optimization evolutionary algorithms were developed for the CMOPs. However, most of them are ineffective in dealing with CMOPs with complex infeasible regions. In this paper, the two archives assisted push-pull evolutionary algorithm (namely PPTA) is proposed to handle the CMOPs with complex infeasible regions effectively. PPTA has the following features: (i) Two archives (A and B) are used, where A is used in the push stage to promote convergence and diversity, and in the pull stage to explore undetected feasible regions; in the pull stage B is used for the promotion of convergence, diversity, and feasibility. (ii) A new push stage strategy is devised for better approximating the unconstrained Pareto front, as well as the constrained Pareto front. (iii) The aggregation function in the pull stage is improved to better explore the objective space. (iv) a new epsilon-constrained method is customized to enhance the ability to explore feasible regions. Results on 33 widely used instances and eight real-world problems verify the superiority of PPTA over 10 related methods.

Automated summary

This paper proposes a new two-archives assisted push-pull evolutionary algorithm (PPTA) to effectively handle constrained multi-objective optimization problems (CMOPs) with complex infeasible regions. Experimental results demonstrate the superiority of PPTA over other related methods.