A dual-grid dual-phase strategy for constrained multi-objective optimization

摘要

Constrained multi-objective optimization problems (CMOPs) appear frequently in engineering applications. In some cases, feasible regions are narrow and/or disconnected. For this kind of problems, existing constraint-handling methods, integrated with multi-objective evolutionary algorithms, are easily stuck at local optima. Aiming to strengthen the global search ability, a dual-grid dual-phase strategy is proposed, which is termed dual-grid push and pull search (DPPS). In the DPPS, two populations, corresponding to dual grids, are used individually to explore the feasible and infeasible spaces. Specifically, one population maintains feasible solutions, and the other explores the whole search space without considering constraints. Then, the two populations share useful information and pull each other so as to enable the algorithm to search for the optimal feasible region (i.e., Pareto solution set). To demonstrate the effectiveness of the proposed algorithm, the MOEA/D integrated DPPS (MOEA/D-DPPS) is tested on a frequently-used benchmark suite as well as a newly-constructed suite. Experimental results clearly show the superiority of MOEA/D-DPPS compared with six state-of-the-art algorithms.