An improved atom search optimization with dynamic opposite learning and heterogeneous comprehensive learning

Abstract

Atom search optimization (ASO) is a newly developed metaheuristic algorithm inspired by molecular basis dynamics. The paramount challenge in ASO is that it is prone to stagnation in local optima due to premature convergence. To solve this issue, a modified atom search optimization with dynamic opposite learning and heterogeneous comprehensive learning is proposed in this paper, which is named DOLHCLASO. First, the asymmetry of dynamic opposite learning can increase the probability that the population will obtain an optimal solution, while dynamic characteristic can enrich population diversity and enhance exploration capability. Second, the heterogeneous comprehensive learning divides the population into two subpopulations, exploration-subpopulation and exploitation-subpopulation, and the division of labor and cooperation effectively balances exploration and exploitation. Finally, the proposed DOLHCLASO was evaluated in the CEC2017 benchmark functions and real-world engineering cases, compared with some classic and excellent variants of algorithms to confirm its performance. Experimental results and statistical analysis demonstrate that the performance of the DOLHCLASO is significantly better than other selected optimizers.