Elite archives-driven particle swarm optimization for large scale numerical optimization and its engineering applications

Particle swarm optimization (PSO) is a very simple and effective metaheuristic algorithm. Search operators with similar behavior may lead to the loss of diversity in the search space. All particles in PSO have the same and single search strategy. Therefore, PSO may suffer from premature convergence in solving complex multimodal problems. To improve the global search ability of PSO, this paper reports an elite archives-driven particle swarm optimization (EAPSO). Note that, EAPSO only needs population size and terminal condition for performing the search task, which can distinguish EAPSO over the other reported variants of PSO. In addition, EAPSO has a clear structure, which first builds three types of elite archives to save three different hierarchical particles. Then, six learning strategies for updating the positions of particles are designed by reusing these particles of the three elite archives. To verify the performance of EAPSO, EAPSO is employed to solve CEC 2013 test suite with dimensions 30–100 and three constrained engineering problems. Experimental results show that EAPSO outperforms the compared seven powerful variants of PSO on more than half of test functions and offers highly competitive optimal solutions on the considered engineering problems. That is, experimental results support the validity of the improved strategies and prove the superiority of EAPSO in solving complex multimodal problems. The source code of EAPSO can be found by the following website: https://github.com/jsuzyy/EAPSO.