Accelerating the convergence of evolutionary algorithms by fitness landscape approximation
A new algorithm is presented for accelerating the convergence of evolutionary optimization methods through a reduction in the number of fitness function calls. Such a reduction is obtained by 1) creating an approximate model of the fitness landscape using kriging interpolation, and 2) using this model instead of the original fitness function for evaluating some of the next generations. The main interest of the presented approach lies in problems for which the computational costs associated with fitness function evaluation is very high, such as in the case of most engineering design problems. Numerical results presented for a test case show that the reconstruction algorithm can effectively reduces the number of fitness function calls for simple problems as well as for difficult multidimensional ones.
KeywordsFitness Function Covariance Function Fitness Landscape Nugget Effect Engineering Design Problem
Unable to display preview. Download preview PDF.
- 1.Keane, A. J.: Experiences with optimizers in structural design. Proc. Conf. Adaptive Computing in Engineering Design and Control (1994) 14–27Google Scholar
- 4.Ratle, A., Berry, A.: Use of genetic algorithms for the vibroacoustic optimization of plate response. Submitted to the Journal of the Acoustical Society of America.Google Scholar
- 5.Michalewicz, Z., Schoenauer, M.: Evolutionary Algorithms for Constrained Parameter Optimization Problems. Evolutionary Computation 4 (1996) 1–32Google Scholar
- 6.Michalewicz, Z.: Genetic Algorithms + Data Structure = Evolution Programs, 2nd ed. Berlin, Springer-Verlag (1994)Google Scholar
- 8.Matheron, G.: Splines et krigeage: leur équivalence formelle. Technical Report N-667, Centre de Géostatistique, école des Mines de Paris (1980)Google Scholar
- 10.Bäck, T., Schwefel, H.-P.: An Overview of Evolutionary Algorithms for Parameter Optimization. Evolutionary Computation 1 (1993) 1–23Google Scholar