Constraints on Control Parameters of Asynchronous Differential Evolution
The efficiency of an algorithm to find the global minimum depends on its ability to keep population diversity during evolutionary iterations. Statistical variance can serve as a measure of population diversity. We analyse the expected population variance after mutation and crossover for best/1/bin strategy of Classical Differential Evolution and for new strategies of a novel Asynchronous Differential Evolution. Relations between the control parameters (N p , F, C r ) of algorithms and the extension factor of population variance are derived. Constraints on control parameters to prevent premature convergence of the algorithm are suggested and compared with phase portraits (convergence domains) for several benchmark functions.
Keywordsglobal optimization derivative-free optimization premature convergence control parameters evolution strategy
Unable to display preview. Download preview PDF.
- 4.Suganthan, P.N., et al.: Problem definitions and evaluation criteria for the CEC05 special session on real-parameter optimization. Technical report, Nanyang Technological University, Singapore (2005), http://www.ntu.edu.sg/home/epnsugan/index_files/CEC-05/CEC05.htm
- 5.Zaharie, D.: Critical values for the control parameters of differential evolution algorithms. In: Proc. 8th Int. Mendel Conf. Soft Comput., pp. 62–67 (2002)Google Scholar
- 6.Zhabitskaya, E.I., Zhabitsky, M.V.: Asynchronous Differential Evolution. In: Adam, G., Buša, J., Hnatič, M. (eds.) MMCP 2011. LNCS, vol. 7125, pp. 328–333. Springer, Heidelberg (2012)Google Scholar