P-lsGOF: A parallel learning-selection-based global optimization framework

Article type: Research Article

Authors: Zuo, Mingcheng^a | Dai, Guangming^{a; b; *}

Affiliations: [a] School of Computer, China University of Geosciences (Wuhan), Wuhan, China | [b] Hubei Key Laboratory of Intelligent Geo-Information Processing, China University of Geosciences, Wuhan, China

Correspondence: [*] Corresponding author. Guangming Dai. E-mail: cugdgm@126.com.

Abstract: When optimizing complicated engineering design problems, the search spaces are usually extremely nonlinear, leading to the great difficulty of finding optima. To deal with this challenge, this paper introduces a parallel learning-selection-based global optimization framework (P-lsGOF), which can divide the global search space to numbers of sub-spaces along the variables learned from the principal component analysis. The core search algorithm, named memory-based adaptive differential evolution algorithm (MADE), is parallel implemented in all sub-spaces. MADE is an adaptive differential evolution algorithm with the selective memory supplement and shielding of successful control parameters. The efficiency of MADE on CEC2017 unconstrained problems and CEC2011 real-world problems is illustrated by comparing with recently published state-of-the-art variants of success-history based adaptative differential evolution algorithm with linear population size reduction (L-SHADE) The performance of P-lsGOF on CEC2011 problems shows that the optimized results by individually conducting MADE can be further improved.

Keywords: Parallel optimization framework, real-world problems, learning-based differential evolution

DOI: 10.3233/JIFS-200753

Journal: Journal of Intelligent & Fuzzy Systems, vol. 39, no. 5, pp. 7333-7361, 2020