A coevolutionary constrained multi-objective algorithm with a learning constraint boundary

TL;DR: This paper proposes CCMOLCB, a coevolutionary constrained multi-objective algorithm that balances convergence, diversity, and feasibility by employing a learning constraint boundary and dynamic weight coefficient, outperforming seven state-of-the-art algorithms on five test suites.

Abstract: When solving constrained multi-objective optimization problems, the balance of convergence, diversity, and feasibility plays a pivotal role. To address this issue, this paper proposes a coevolutionary constrained multi-objective algorithm with learning constraint boundary (CCMOLCB). Firstly, the constrained multi-objective problems are transformed by adding an additional objective using the constraint violation degree. Then, the transformed problem is solved by an improved coevolutionary framework which employs two populations. The main population explores the objective space and repairs infeasible solutions to maintain the feasibility of population. Meanwhile, the feasibility and diversity of solutions are balanced by using a dynamic weight coefficient during the evolution, it changes as the number of iterations increases. The subordinate population selects solutions by taking into consideration the learning constraint boundary (LCB). This boundary guarantees convergence of solutions by constraining the search range of the main population, thereby enhancing the environmental selection pressure. The performance of CCMOLCB is compared with seven state-of-the-art constrained multi-objective evolutionary algorithms on five test suites. The experimental results illustrate that CCMOLCB exhibits competitive performance when dealing with this family of problems.