Optimization of Highly Non-Linear Simulation Functions Using Improved Adaptive Response Surface Methodology
Adaptive Response Surface Methodology (ARSM) is a new developed method sequentially estimates the optimum of a complex function in a gradually reduced design space. In this paper, previously developed approach in ARSM has been compare and contrast with a new approach proposed by authors: using inherited Latin Hypercube Design (LHD) points and generating new Maximin LHD points by solving facility location problems in each iteration. Minimum of the response surface which is a second order approximate model based on LHD points in each iteration is an estimation of the real optimum of the complex function. The computation results reveal that the new suggested approach demonstrate more precise estimation of minimum with less number of function calls compare to the ARSM with random LHDs which is an improvement in the efficiency and accuracy of ARSM.
M. S. Manesh et al., "Optimization of Highly Non-Linear Simulation Functions Using Improved Adaptive Response Surface Methodology", Applied Mechanics and Materials, Vols. 110-116, pp. 2662-2669, 2012