Adaptive properties in Genetic Algorithms applied to structural optimization

Abstract

From pioneer works of Holland (1975) and Goldberg (1989) until now the objective of research in Genetic Algorithms (GAs) has been to increase the efficiency of algorithms. The two most relevant conclusions can be extracted from literature are: first, the importance of randomness of the main operators namely selection, crossover and mutation, and second the referred randomness improves the initial population fitness inducing its evolution towards the global optimum. However, some aspects of GAs are not explained and the optimality conditions of the method stay unknown. Most of the remaining information on efficiency of algorithms has heuristic nature or is deduced from numerical tests applied to simple examples. Despite the above considerations it is recognized that GA efficiency improves clearly if some adaptive rules are included. In the present work, adaptive properties in GAs applied to structural optimization are studied. Here, adaptive rules perform using additional information related with the behavior of state and design variables of the structural problem. At each generation the adaptation of genetic parameters to evolutionary conditions aims to improve the efficiency of genetic search. The introduction of adaptive rules occurs at three levels: (i) when defining the search domain at each generation; (ii) considering a crossover operator based on commonality and local improvements; and (iii) by controlling mutation including behavioral data.