Release Time:2019-03-09  Hits:

Indexed by: Journal Papers

Date of Publication: 2015-10-01

Journal: JOURNAL OF GLOBAL OPTIMIZATION

Included Journals: Scopus、EI、SCIE

Volume: 63

Issue: 2

Page Number: 229-251

ISSN: 0925-5001

Key Words: Lipschitz constant; Approximate function; Response surface; Global optimization; Black box function

Abstract: This article develops a novel global optimization algorithm using potential Lipschitz constants and response surfaces (PLRS) for computationally expensive black box functions. With the usage of the metamodeling techniques, PLRS proposes a new approximate function to describe the lower bounds of the real function in a compact way, i.e., making the approximate function closer to . By adjusting a parameter (an estimate of the Lipschitz constant ), could approximate in a fine way to favor local exploitation in some interesting regions; can also approximate in a coarse way to favor global exploration over the entire domain. When doing optimization, PLRS cycles through a set of identified potential estimates of the Lipschitz constant to construct the approximate function from fine to coarse. Consequently, the optimization operates at both local and global levels. Comparative studies with several global optimization algorithms on 53 test functions and an engineering application indicate that the proposed algorithm is promising for expensive black box functions.