Micro Differential Evolution Performance Empirical Study for High Dimensional Optimization Problems

• Published in: Large-Scale Scientific Computing pp. 281 - 288
• Main Authors: Olguin-Carbajal, Mauricio, Herrera-Lozada, J. Carlos, Arellano-Verdejo, Javier, Barron-Fernandez, Ricardo, Taud, Hind
• Format: Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2014
• Series: Lecture Notes in Computer Science
• Subjects: Differential evolution; High dimensionality; Micro-algorithm; Reduced population
• ISBN: 3662438798; 9783662438794
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-662-43880-0_31

This paper presents an empirical study of a micro Differential Evolution algorithm (micro-DE) performance versus a canonical Differential Evolution (DE) algorithm performance. Micro-DE is a DE algorithm with reduced population and some other differences. This paper’s objective is to show that our micro-DE outperforms the canonical DE for large scale optimization problems by using a test bed consisting of 20 complex functions with high dimensionality for a performance comparison between the algorithms. The results show two important points; first, the relevance of an accurate set of the optimization algorithms parameters regarding the problem itself. Second, we demonstrate the superior performance of our micro-DE with respect to DE in 19 out 20 tested functions. In some functions, the difference is up to seven orders of magnitude. Also, we show that micro-DE is better statistically than a simple DE and an adjusted DE for high dimensionality. In several problems where DE is used, micro-DE is highly recommended, as it achieves better results and statistic behavior without much change in code.