A surrogate-assisted variable grouping algorithm for general large-scale global optimization problems

Problem decomposition plays an important role when applying cooperative coevolution (CC) to large-scale global optimization problems. However, most learning-based algorithms only apply additively separable problems, while the others insensitive problem type perform low accuracy and efficiency. Given this limitation, study designs a general-separability-oriented detection criterion, further proposes novel algorithm called surrogate-assisted variable grouping (SVG). The new criterion detects separability between some other variables by checking whether its optimum changes with latter. Consistent definition of general separability, endows SVG strong applicability high accuracy. To reduce expensive fitness evaluations, locates help surrogate model rather than original high-dimensional model. Moreover, it converts variable-grouping process into search in binary tree taking subsets as nodes. This facilitates reutilization historical information, thereby reducing times. Experimental results on benchmark suite indicate that compared six state-of-the-art algorithms, achieves higher efficiency both nonadditively Furthermore, can significantly enhance performance CC.