Fuction Approximation for GA Based Optimization

Abstract

*† Optimization of large-scale practical structures is a challenge even to this day. Mathematical programming techniques have been found to be inadequate for the solution of large practical problems with different types of structural elements and design variables. Genetic Algorithms (GAs), owing to its robustness and global nature along with other advantages over mathematical techniques are being increasingly used for solution of such problems. However, using GAs for large-scale optimization could be very expensive due to its requirement of large number of function evaluations for convergence. This would result in a prohibitive cost for computation of functional evaluations even with the best computational resources available today. A response surface method for GA based structural optimization is presented in this paper to enable solution of large practical optimization problems. The proposed method for function approximation is simple and effective. The method has no limitations on number of design variables in the problem. The method was illustrated by optimising a ten bar truss with 10 design variables and a fuselage of a light transport aircraft. The optimum values obtained from function approximation approach were found to be in good agreement with that obtained from search without function approximations. The total number of actual evaluations was reduced by 86% in the large-scale problem, which is a significant saving in computational time. This method is very promising and can be a tremendous advantage for GA based optimization of large-scale practical structures. Nomenclature