A Surrogate-Assisted Evolutionary Algorithm Based on K-Means Clustering and Lévy Flight

Expensive optimization problem refers to a category of problems for which numerical simulations or physical experiments are expensive and intractable. Evolutionary algorithm (EA) cannot be directly applied to such problems due to the large number of fitness evaluations consumed to determine the global best solution. Hence, this paper proposes a surrogate-assisted evolutionary algorithm based on k-means clustering and Lévy flight (SAEA-KL), where the Radial Basis Function network (RBFN) is selected as a promising surrogate for saving the number of expensive evaluations, and the dual-start Lévy flight and k-means clustering method are deployed to initialize the population of the competitive swarm optimizer in consideration of both optimal and fitness landscape information. Experimental results on common benchmark problems indicate that the proposed method is competitive with seven state-of-the-art algorithms under a limited computational budget.