A tri-population constrained multi-objective evolutionary algorithm based on multitask co-evolution strategy
Jianlong Xu, Yongkuan Yang, YuKai Hua, Zhao Jing · 2022
Convergence, diversity and feasibility are difficult to balance in constrained multi-objective optimization problems（CMOP）, especially in some CMOP with relatively small feasibility. In order to solve this problem, this paper proposes a tri-population evolutionary algorithm with multitask co-evolution strategy. The main task is responsible for the original constrained multi-objective optimization. The second task called the help task executes the unconstrained search, and the third task named the transition task fills the gap between the main and the second tasks by treating the degree of constraint violation as another optimization objective. Finally, through the knowledge transfer between the population offspring of the three tasks, the offspring with excellent convergence are transferred to the main task population and get fully exported. In experiments, The algorithm is also compared with four high-level constrained multi-objective evolutionary algorithms (CMOEA) on 24 bench-mark CMOPs. The experimental results demonstrate the competitiveness of the proposed algorithm.