CCGSK-PSO: Constrained Multi-objective Particle Swarm Optimization Algorithm Based on Co-evolution and Gaining-Sharing Knowledge

Constrained multi-objective optimization is facing huge challenges due to the complex constraints and the restraint of multiple optimization objectives. The most difficult problem need to solve is to achieve the balance between optimization objective convergence and feasibility under constraints as far as possible. To solve this problem, a constrained multi-objective particle swarm optimization algorithm based on co-evolution and gaining-sharing knowledge mechanism is proposed, which is called CCGSK-PSO. Based on the evolutionary strategy of co-evolution, the evolutionary process conducts two types of co-evolutionary in each iteration: the co-evolution of unconstrained evolutionary population and constrained evolutionary population, and the co-evolution of objective convergence subpopulation and constrained subpopulation in constrained evolutionary population. The objective convergence subpopulation is mainly responsible for exploring the objective convergence boundary information, while learning from the individuals in constrained subpopulation in the neighborhood to guide the population to effectively cross the infeasible regions. The constrained subpopulation is mainly responsible for exploring the constraint boundary information, while guiding the population to converge to the elite individuals. Finally, the obtained solution set is widely and uniformly distributed along the constrained Pareto front. The experimental results on 37 test problems in three benchmark suites show the effectiveness of CCGSK-PSO.