![]() |
Computational & Technology Resources
an online resource for computational,
engineering & technology publications |
Civil-Comp Proceedings
ISSN 1759-3433 CCP: 76
PROCEEDINGS OF THE THIRD INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY Edited by: B.H.V. Topping and Z. Bittnar
Paper 74
The Effect of Oscillating Population Size and Re-initialization on the Performance of Genetic Algorithms V.K. Koumousis and C.P. Katsaras
Institute of Structural Analysis & Aseismic Research, Department of Civil Engineering, National Technical University of Athens, Greece Full Bibliographic Reference for this paper
V.K. Koumousis, C.P. Katsaras, "The Effect of Oscillating Population Size and Re-initialization on the Performance of Genetic Algorithms", in B.H.V. Topping, Z. Bittnar, (Editors), "Proceedings of the Third International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 74, 2002. doi:10.4203/ccp.76.74
Keywords: genetic algorithms, variable population, micro-GA.
Summary
Genetic Algorithms (GAs) are search algorithms based on the concepts of natural
selection and survival of the fittest (Goldberg 1989) [1]. A number of methods have
been developed, based on the standard GA scheme, to improve the robustness and
computational efficiency of GAs (Smith and Fogarty, 1997) [2]. Moreover, an effort
to categorize the developed methodologies devoted to the parameter control of
evolutionary algorithms is presented by Eiben (Eiben et al. 1999) [3].
One of the main parameters that affect the robustness and computational
efficiency of the GAs is the population size. In this work a Genetic Algorithm (GA)
is proposed that uses a variable population size in the form of a saw-tooth function,
having a specific amplitude The performance of the three different algorithms is evaluated on the basis of the statistics of their maximum and average fitness histories over the generations for a number of GA runs based on different random seeds. The mean curves of these fitness histories for the different GA runs are plotted to reveal the convergence behaviour and performance features of the examined algorithms. The proposed scheme is applied into two categories of problems that are often used as benchmark tests. These correspond to two n-dimensional multimodal peak functions with different features. Numerical results are presented for a wide range of parameters. The main finding is that for large amplitudes and a broad range of values for the period of variation of the population size, the overall performance of the proposed scheme reaches the performance of a Standard GA of substantial bigger population size. This trend is justified also on the basis of schema theorem [5,6].
The previous theoretical and statistical analysis can be used to form guidelines
towards the selection of optimum References
purchase the full-text of this paper (price £20)
go to the previous paper |
|