A knowledge transfer-based strategy for constrained multiobjective optimization

The complex constraints in constrained multiobjective optimization problems may cause the Pareto front to be distributed on disconnected feasible boundaries. Most existing evolutionary algorithms encounter challenges in obtaining the entire Pareto front due to inappropriate cooperation between the populations. The ideology of knowledge transfer provides inspiration for addressing complex optimization problems. Based on this, this paper proposes a knowledge transfer-based coevolutionary algorithm, which adopts the idea of divide-and-conquer and two combined into one. The algorithm derives the original constrained multiobjective optimization problem into two problems, both of which share the same optimization objective but follow distinct search trajectories. Specifically, one problem focuses on global search, while the other emphasizes local search. A knowledge transfer strategy is proposed to achieve the exchange of complementary information between these two problems in the evolutionary directions. This strategy assists in solving the derived problem by transferring promising individuals that remain undiscovered in the search trajectories. The optimal solution of the original constrained multiobjective optimization problem is obtained. Experiments conducted on 56 benchmark problems show superior or competitive performance compared with 11 state-of-the-art algorithms.