A flexible two-stage constrained multi-objective evolutionary algorithm based on autonomic regulation

The core element in solving constrained multi-objective problems (CMOPs) with evolutionary algorithms is simultaneously balancing objective optimization and constraint satisfaction. Maintaining this balance becomes more challenging for existing algorithms when dealing with complex CMOPs, as various complex feasible regions often result in CMOPs with very different characteristics. To address this issue, we propose a more flexible two-stage evolutionary algorithm based on automatic regulation (ARCMO), which can effectively control evolutionary trends to adapt to complex CMOPs. Specifically, the first stage performs a fast global search and passes the population information to the second stage. The second stage consists of two dynamically complementary sub-processes: the exploration subprocess and the convergence subprocess. The ratio of these two subprocesses is dynamically adjusted based on information from the first stage, allowing ARCMO to adapt to CMOPs with different complexities. Experiments on several recently proposed benchmark suites and real-world application problems show that ARCMO is more adaptable than the contender algorithms when solving different complex CMOPs.