Space Partition based Response Surface Method and Its Performance Measures

摘要

This paper proposed a space partition based response surface method (SPRSM) for global approximation in highly non-linear optimization problems. Under this method, a max-error guided partition (MEGP) strategy is adopted to divide the target space into small blocks and response surfaces are repetitively constructed over the blocks during the optimization process. Experimental results showed that SPRSM had a significant improvement over conventional RSM in both accuracy and efficiency.