Two-Stage Dynamic Cooperative Evolution Algorithm for Constrained Multi-objective Optimization

Compared with multi-objective optimization problems(MOPs), constrained multi-objective optimization problems(CMOPs) have to consider the impact of constraints on seeking Pareto front in addition to convergence and diversity, thus how to balance these factors is a challenging task of CMOPs. To confront this challenge, a new two-stage dynamic collaborative evolutionary algorithm, called TSDCEA, is proposed to solve the CMOPs. TSDCEA divides the evolutionary process of the population into two stages. In the first stage, the two populations evolve collaboratively so that the auxiliary population rapidly explores infeasible regions while the main population explores feasible regions. The auxiliary population stops evolving when it basically reaches the unconstrained pareto front(UPF) and the population evolution process switches to the second stage. Then the main population applies a dynamic ε-constraint method, while the auxiliary population is utilized for differential evolution with elite individuals from the main population. The generated new offspring participate in environmental selection to explore feasible regions close to the main population and encourage the main population to disperse widely in the explored feasible regions. Comparative experiments show that the TSDCEA algorithm is more competitive than other state-of-the-art constrained multi-objective evolutionary algorithms(CMOEAs).