Global Optimization Method with Enhanced Adaptive Response Surface Method for Computation-Intensive Design Problems

摘要

The drawbacks of standard adaptive response surface method (ARSM) for global optimization of computation-intensive design problems including low efficiency and poor global convergence capability are firstly discussed. The cutting plane approach employed in ARSM is the root caused such drawbacks of ARSM. This study addresses development of the enhanced adaptive response surface method (EARSM). The algorithm and overall process of EARSM are represented in detail, and a novel design space reduction approach named significant design space (SDS) approach is proposed to replace cutting plane approach for reducing design space during optimization process. Due to use of SDS approach, no extra auxiliary optimizations are required in EARSM. The validation results of numerical benchmark problems and simple beam optimization problem prove that both efficiency and global convergence capability of EARSM are evidently improved compared with ARSM. In addition, the limitations of EARSM are also pointed out and analyzed to reveal the reason for the limitations. Finally, future works for enhancing EARSM are recommended as to elimination such limitations.