On increasing computational efficiency of evolutionary algorithms applied to large optimization problems

This paper presents new advances in development of dedicated Evolutionary Algorithms (EA) for large non-linear constrained optimization problems. The primary objective of our research is a significant increase of the computational efficiency of the standard EA. The EA are understood here as Genetic Algorithms using decimal chromosomes, three standard operators: selection, crossover, and mutation, as well as additional new speed-up techniques. So far we have preliminarily proposed several general concepts, including smoothing and balancing, a´posteriori solution error analysis and related techniques, as well as an adaptive step-by-step mesh refinement. We discuss here the efficiency of chosen speed-up techniques using simple but demanding benchmark problems, including residual stress analysis in elastic-perfectly plastic bodies under cyclic loadings, and physically based smoothing of experimental data. Particularly, we consider a smoothing technique using average solution curvature, new criteria for selection based on global solution error, as well as a step-by-step mesh refinement combined with smoothing. Preliminary numerical results clearly indicate a possibility of significant acceleration of calculations, as well as practical application of the improved EA to the optimization problems considered.