Authors: Pavel Novoa-Hernández; Carlos Cruz Corona; David A. Pelta
Addresses: Department of Mathematics, University of Holguin, Holguin, Cuba ' Department of Computer Science and Artificial Intelligence, Center for Research in Information and Communication Technologies (CITIC-UGR), University of Granada, Granada, Spain ' Department of Computer Science and Artificial Intelligence, Center for Research in Information and Communication Technologies (CITIC-UGR), University of Granada, Granada, Spain
Abstract: Self-adaptation is a popular parameter control technique in evolutionary computation, which has been extensively studied in stationary optimisation. In the context of dynamic optimisation problems (DOPs), there are research works that suggest the application of such technique. Nevertheless, some important issues remain open, for example, how self-adaptation can be more profitable for a given algorithm. From the survey we made, it is possible to distinguish three main application levels of self-adaptation in dynamic environments: metaheuristic level, 'mechanism for DOPs' level, and the combination of both. While most of the related works belong to the first level, a small number can be grouped in the second one. However, in contrast to previous two, unfortunately, very little or nothing has been done with the third one. Based on these motivations, in this paper we empirically analysed the role of several self-adaptive models in these levels using multipopulation differential evolution algorithms as baseline. The results suggest that self-adaptation has a significant impact when applied at least to the 'mechanism for DOPs' level.
Keywords: self-adaptation; dynamic optimisation; differential evolution; bio-inspired algorithms; parameter control; evolutionary computation; self-adaptive models.
International Journal of Bio-Inspired Computation, 2016 Vol.8 No.1, pp.1 - 13
Available online: 09 Feb 2016 *Full-text access for editors Access for subscribers Purchase this article Comment on this article