Multi-Point Simulated Annealing with Adaptive Neighborhood

Keiko ANDO  Mitsunori MIKI  Tomoyuki HIROYASU  

Publication
IEICE TRANSACTIONS on Information and Systems   Vol.E90-D    No.2    pp.457-464
Publication Date: 2007/02/01
Online ISSN: 1745-1361
DOI: 10.1093/ietisy/e90-d.2.457
Print ISSN: 0916-8532
Type of Manuscript: Special Section PAPER (Special Section on Foundations of Computer Science)
Category: Optimizing Algorithms
Keyword: 
optimizing algorithm,  simulated annealing,  adaptive neighborhood,  continuous optimization problem,  

Full Text: PDF(700.5KB)>>
Buy this Article



Summary: 
When Simulated Annealing (SA) is applied to continuous optimization problems, the design of the neighborhood used in SA becomes important. Many experiments are necessary to determine an appropriate neighborhood range in each problem, because the neighborhood range corresponds to distance in Euclidean space and is decided arbitrarily. We propose Multi-point Simulated Annealing with Adaptive Neighborhood (MSA/AN) for continuous optimization problems, which determine the appropriate neighborhood range automatically. The proposed method provides a neighborhood range from the distance and the design variables of two search points, and generates candidate solutions using a probability distribution based on this distance in the neighborhood, and selects the next solutions from them based on the energy. In addition, a new acceptance judgment is proposed for multi-point SA based on the Metropolis criterion. The proposed method shows good performance in solving typical test problems.


open access publishing via