Adaptive Design Space Reconstruction Method in Surrogate Based Global Optimization

Surrogate-based global optimization (SBO) has gained rapid dominance in engineering design. However, traditional SBO method over entire design space with large size interval would be considerably time-consuming. In order to improve the optimization efficiency in SBO, an adaptive design space reconstruction (ADS) method based on fuzzy clustering method and effective sample points is proposed in this paper. Fuzzy c mean clustering method is applied to divide the initial design space into several sub-regions from which we choose the sub-region which is most likely to contain the global optima. During the optimization process, effective sample points are collected to be the center of new space constructed by trust region method, instead of a single sample point, to keep optimization from getting trapped in local minimums. Then the optimization search will be managed in the reconstructed promising sub-region. We test and verify the proposed method with the airfoil drag minimization problems proposed by Aerodynamic Design Optimization Discussion Group (ADODG), which could demonstrate that better results can be obtained within the reconstructed design space with high efficiency.