Constrained multi-objective optimisation with collaboration-driven strategy and infeasible knowledge-based cooperation

Most existing constrained multi-objective evolutionary algorithms adopt unitary environmental selection and reproduction operator, which face the issue of striking a proper balance between objectives and constraints. To this end, we developed a new method with collaboration-driven strategies and infeasible knowledge-based cooperation. The main idea of this approach is using heterogeneous frameworks and operators. Specifically, the main population is maintained using the domination-based framework and constrained-domination principle. An auxiliary population is maintained using the decomposition-based framework and epsilon-constrained method. Meanwhile, the infeasible knowledge-based genetic algorithm and differential evolution are employed to generate offspring. The superior of this algorithm is validated by comparing with some popular methods with distinct characteristics on four suites. Experimental results indicated that our method performs best among all competitors on 62.5%, 50%, 50%, and 25.56% for CTPs, MWs, LIRCMOPs, and DOACMOPs, respectively. The effectiveness of collaboration-driven and infeasible knowledge-based cooperation are also verified by some ablation studies.