Mode pursuing sampling method using coordinate perturbation for high-dimensional expensive black-box optimization

This article presents a novel mode pursuing sampling method using coordinate perturbation, notated as MPS-CP, to further alleviate the computational cost in solving highdimensional, expensive, and black-box (HEB) problems. Different from the conventional mode pursuing sampling method (MPS), the proposed MPS-CP employs a novel coordinate perturbation strategy for sequential sampling. In this approach, cheap points are generated using adaptive perturbation on a subset of design variables. The extent of perturbation is adaptively adjusted to make a trade-off between local exploitation and global exploration. Expensive sample points are selected sequentially from the cheap points according to specific criteria. The overall performance of MPS-CP is tested on several numerical benchmarks and compared with original MPS and several state-of-the-art algorithms including DYCORS and OMID. The results indicate that the proposed MPS-CP generally outperforms the competitive algorithms in terms of efficiency, convergence, and robustness. Finally, the proposed MPS-CP is applied to an all-electric GEO satellite (AEGS) multidisciplinary design optimization (MDO) problem to demonstrate its effectiveness and practicality in engineering practices.