A memetic Differential Evolution approach in noisy optimization

This paper proposes a memetic approach for solving complex optimization problems characterized by a noisy fitness function. The proposed approach aims at solving highly multivariate and multi-modal landscapes which are also affected by a pernicious noise. The proposed algorithm employs a Differential Evolution framework and combines within this three additional algorithmic components. A controlled randomization of scale factor and crossover rate are employed which should better handle uncertainties of the problem and generally enhance performance of the Differential Evolution. Two combined local search algorithms applied to the scale factor, during offspring generation, should enhance performance of the Differential Evolution framework in the case of multi-modal and high dimensional problems. An on-line statistical test aims at assuring that only strictly necessary samples are taken and that all pairwise selections are properly performed. The proposed algorithm has been tested on a various set of test problems and its behavior has been studied, dependent on the dimensionality and noise level. A comparative analysis with a standard Differential Evolution, a modern version of Differential Evolution employing randomization of the control parameters and four This research is supported by the Academy of Finland, Akatemiatutkija 00853, Algorithmic Design Issues in Memetic Computing. Ernesto Mininno E-mail: [email protected] Ferrante Neri Tel.: +358-14-2602764 E-mail: [email protected] University of Jyväskylä Department of Mathematical Information Technology P.O. Box 35 (Agora) 40014 University of Jyväskylä Finland Fax: +358-14-2604981