Constrained many-objective evolutionary algorithm based on adaptive infeasible ratio

Constrained many-objective optimization problems (CMaOPs) pose great challenges for evolutionary algorithms to reach an appropriate trade-off of solution feasibility, convergence, and diversity. To deal with this issue, this paper proposes a constrained many-objective evolutionary algorithm based on adaptive infeasible ratio (CMaOEA-AIR). In the evolution process, CMaOEA-AIR adaptively determines the ratio of infeasible solutions to survive into the next generation according to the number and the objective values of the infeasible solutions. The feasible solutions then undergo an exploitation-biased environmental selection based on indicator ranking and diversity maintaining, while the infeasible solutions undergo environmental selection based on adaptive selection criteria, aiming at the enhancement of exploration. In this way, both feasible and infeasible solutions are appropriately used to balance the exploration and exploitation of the search space. The proposed CMaOEA-AIR is compared with the other state-of-the-art constrained many-objective optimization algorithms on three types of CMaOPs of up to 15 objectives. The experimental results show that CMaOEA-AIR is competitive with the compared algorithms considering the overall performance in terms of solution feasibility, convergence, and diversity.