Self-adaptation of mutation distribution in evolution strategies for dynamic optimization problems

Issue title: Feature and algorithm selection with Hybrid Intelligent Techniques

Guest editors: Teresa B. Ludermir, Ricardo B.C. Prudêncio and Cleber Zanchettin

Article type: Research Article

Authors: Tinós, Renato^{a; *} | Yang, Shengxiang^b

Affiliations: [a] Department of Computing and Mathematics, FFCLRP, University of São Paulo (USP), 14040-901, Ribeirão Preto, SP, Brazil | [b] Department of Information Systems and Computing, Brunel University, Uxbridge, Middlesex UB8 3PH, UK | Federal University of Pernambuco, Recife, Brazil

Correspondence: [*] Corresponding author. E-mail: [email protected]

Abstract: Evolution strategies with q-Gaussian mutation, which allows the self-adaptation of the mutation distribution shape, is proposed for dynamic optimization problems in this paper. In the proposed method, a real parameter q, which allows to smoothly control the shape of the mutation distribution, is encoded in the chromosome of the individuals and is allowed to evolve. In the experimental study, the q-Gaussian mutation is compared to Gaussian and Cauchy mutation on experiments generated from the simulation of evolutionary robots and on dynamic optimization problems generated by the Moving Peaks generator.

DOI: 10.3233/HIS-2011-0136

Journal: International Journal of Hybrid Intelligent Systems, vol. 8, no. 3, pp. 155-168, 2011