Dynamic cooperative coevolution based on variable importance for non-separable large-scale global optimization

A wide range of large-scale optimization problems have emerged across various fields. Among these problems, non-separable problems pose significant challenges in finding global optimal solutions. To tackle these challenges, cooperative coevolution has become a widely utilized algorithmic framework for solving large-scale problems by decomposing them into smaller components. However, existing methods for problem decomposition struggle with effectively handling the inherent structural characteristics of non-separable problems. To address this issue, this paper presents a dynamic grouping approach based on variable importance. Throughout the entire solution process, this approach periodically assesses the importance of each variable by quantifying the impact of their perturbations on the objective function. The variable group is then formed by selecting a small number of highly important variables and combining them with randomly selected variables. Additionally, a variable reset method is introduced to reset some of the already-converged variables, enhancing the exploration capabilities. The proposed algorithm is evaluated against five state-of-the-art algorithms using a self-established non-separable function benchmark suite. The results demonstrate that the proposed algorithm performs exhibits significant advantages on non-separable functions. Specifically, the proposed algorithm achieves a success rate of 13 out of 16 functions.