Multi-phase constrained multi-objective optimization via heterogeneous transfer

Despite the achievements of multi-objective evolutionary algorithms in solving Constrained Multi-Objective Optimization Problems (CMOPs), their efficiency deteriorates when dealing with complex constraints. This deficiency is mainly due to the lack of diversity among the generated offspring, resulting from the singularity of the evolutionary operator. To address this issue, we introduce an Adaptive Multi-Phase Constrained Multi-objective Optimization Algorithm using Heterogeneous Transfer (AMPHT). By using knowledge from unconstrained Multi-Objective Problems (MOPs) to identify beneficial solutions for CMOPs, AMPHT generates diverse offspring through traditional genetic and difference operators, as well as by leveraging a heterogeneous transfer learning method. To prevent computational resource strain from excessive offspring generation, AMPHT integrates an adaptive multi-phase optimization framework. This evaluates the efficiency of each offspring generation method and dynamically adjusts their implementation frequencies. Our comprehensive testing of AMPHT on sixty-six problems, benchmarked against twelve cutting-edge algorithms, validates its enhanced performance in solving complex CMOPs and handling a wider problem range.