An Adaptive Two-Stage MOEA/D For Solving Constrained Multi-Objective Problem

The crux of solving constrained multi-objective optimization problems (CMOPs) lies in balancing the interplay between constraints, convergence, and diversity. However, the relationship among these three elements is intricately complex, altering one can inadvertently lead the other two towards undesirable outcomes, making the constrained multi-objective problem very challenging. To address this, this paper proposes a novel two-stage multi-objective evolutionary algorithm based on decomposition (MOEA/D), TCBCMO, designed for addressing constrained multi-objective optimization problems (CMOPs) through constraint classification. The algorithm employs MOEA/D in the first stage to identify the unconstrained Pareto front (UPF), followed by a new constraint handling technique (CHT) in the second stage that classifies constraints and dominance relations, facilitating the transition to the constrained Pareto front (CPF). In addition, an adaptive stage-switching method is proposed to retain more genetically good individuals during population iterations and improve algorithmic stability. Lastly, we validate the effectiveness of TCBCMO in benchmark tests of CMOPs, where the evaluated performance demonstrates superior results on most of the test problems compared to the other six related algorithms.