Abstract/Summary

Due to the complex topology of the search space, expensive multi-objective evolutionary algorithms (EMOEAs) emphasize enhancing the exploration capability. Many algorithms use ensembles of surrogate models to boost the performance. Generally, the surrogate-based model either works out the solution’s fitness by approximating the evaluation function or selects the solution by weighting the uncertainty degree of candidate solutions. This paper proposes a selection operator called Cheap surrogate selection (CSS) for multi-objective problems by utilizing the density probability on a k-dimensional tree. As opposed to the first type of surrogate models, which approximate the objective function, the proposed CSS only estimates the uncertainty of the candidate solutions. As a result, CSS does not require extensive sampling or training. Besides, CSS makes use of neighbors’ density and builds the tree with low computational complexity, resulting in an accelerated surrogate process. Moreover, a new EMOEA is proposed by integrating spherical search as the core optimizer with the proposed selection scheme. Over a wide variety of benchmark problems, we show that the proposed method outperforms several state-of-the-art EMOEAs.