A Multi-objective Evolutionary Algorithm Based on Two-Stage Search and Constraint-Dominance Indicator

In recent years, there has been increasing attention towards solving Constrained Multi-objective Optimization Problems (CMOPs), leading to the proposition of various constrained multi-objective evolutionary algorithms. However, most methods struggle to achieve both desirable convergence and diversity when dealing with CMOPs with complex infeasible regions. This paper proposes a constrained multi-objective evolutionary algorithm based on Two-Stage Search and Constraint-Dominance Indicator (TSCDI). In the first stage, the algorithm disregards constraints to traverse infeasible regions; subsequently, in the second stage, it filters infeasible solutions using a constraint-dominance indicator to retain high-quality solutions within the infeasible region. Simulation experiments conducted on LIRCMOP and DASCMOP series test problems demonstrate that compared to five representative algorithms (PPS, C-TAEA, DPPPS, c-DPEA, ICMA), the proposed algorithm demonstrated superior performance, achieving the best IGD and HV scores in 15 instances each. This indicates that the proposed algorithm can effectively address problems with large infeasible regions and relatively small feasible regions.